Evolutionary development (evo devo, ED)
This page offers a general systems definition of the phrase "evolutionary development", and an introduction to its application to autopoetic (self-reproducing) complex systems, including the universe as a system. Its primary author is EDU community co-founder, systems theorist John M. Smart. The EDU community seeks to reconcile evolutionary, developmental, physical, informational, and adaptive approaches to universal structure, function, and complexity, and its members hold many different views. Links to alternative definitions for this phrase, and alternative applications in complex systems and philosophy may appear here, as they are proposed. For more on our community, see our homepage.
- 1 Definition and overview
- 2 Two polar categories and tensions
- 3 Contingency vs. inevitability: The two extremes of scientific and societal bias
- 4 The VCRIS dynamical model of natural selection in autopoetic systems
- 5 Evolutionary development in organisms: The 95/5 rule
- 6 The riddle of development and the challenge to cosmology
- 7 Do we live in an evo devo universe? The EDU hypothesis
- 8 The fine-tuned universe hypothesis: Early evidence for universal ED
- 9 The partially fine-tuned universe hypothesis: Intelligence is a weak selector, not a designer
- 10 The riddle of convergent evolution: Limited forms most beautiful
- 11 Less-optimizing convergence (LOC) versus optimizing convergence (OC)
- 12 Optimizing convergence as accelerating and stabilizing evolutionary development, on many scales
- 13 “Tape of life” (“identical Earths”) experiments: Simulating ecogeophysical ED
- 14 “Tape of the cosmos” (“identical Universes”) experiments: Simulating Universal ED
- 15 Physical and informational adaptation: Autopoesis and intelligence
- 16 Evo devo models require advances in a variety of theories, especially our theories of intelligence
- 17 Observation selection effects: The challenge of assessing them for fine-tuning and convergence
- 18 Ambiguity of the word “evolution”, and the modern evolutionary synthesis
- 19 The value of religion, and the fallacy of intelligent design as an explanation for adapted complexity
- 20 Research questions
- 21 Acknowledgments
- 22 References
Definition and overviewautopoetic (self-reproducing) complex system.
The hyphenated “evo-devo” is commonly used for living systems, most prominently in evo-devo genetics and epigenetics, and the unhyphenated “evo devo” can be used for the theory of any potentially replicating and adapting complex system (star, prebiotic system, gene, cell, organism, meme (concept), behavior, technology), whether living or nonliving. Occasionally, the hyphenated term “eco-evo-devo” is used to place evo-devo biology within ecological systems with their own evo devo dynamics (e.g., Pigliucci 2007; Gilbert, Bosch, and Ledón-Rettig 2015). This is controversial, since the conventional neo-Darwinian Modern Synthesis does not recognize multi-level selection, and maintains that ecological dynamics are secondary to species competition. But the rise of theoretical and systems ecology and its models, including ecological energetics, panarchy, and ascendancy, can be viewed as supporting the idea that ecologies themselves both evolve and develop. Finally, future evo devo models may require what Lucia Jacobs refers to as “cog-evo-devo” (Jacobs 2012), the recognition that both information and cognition evolve and develop, are causal agents in the dynamics of complex replicators, and are increasingly important in determining their future, via niche construction, as higher intelligence emerges.
Inspired by the work of evo-devo biologists, evo devo systems theorists look for processes of evolutionary creativity and developmental constraint in any autopoetic complex systems, at any scale. Evo devo systems theory thus redefines the much-used but increasingly multi-causal term “evolution”, to restrict evolutionary process to contingent, stochastic, information-creative, experimental, diversifying, and nonhierarchical processes of system change, when we are referring to evolution within the boundaries of any proposed autopoetic system. These processes are the dynamical and informational opposite of the predictable, information-conservative, convergent, unifying, and hierarchical processes of “development,” which work to replicate and maintain that system. Redefinitions of long-used words are never a popular choice, but this redefinition is potentially clarifying for autopoetic dynamics, from the perspective of information theory. If evolutionary processes necessarily generate new information, and developmental processes conserve and build upon old information, and we can determine “new” or “old” only in relation to the life cycle of the system under analysis, we may have a useful new perspective on both dynamics and their intrinsic predictability to observers within any self-reproducing system.
When we apply these definitions to the life cycle of an individual organism, as in the figure at right, we can observe evolutionary, information-creative processes in such events as stochastic gamete production, and in the stochastic cellular, molecular, and genetic (via such processes as transposable elements and somatic recombination) architecture within any frog, by comparison to its neighbors. Simultaneously, we can observe developmental, information-conservative processes in any replicative dynamics, informatics, and morphology that we observe in all frogs of a specific species. Both evolutionary and developmental processes can thus be empirically differentiated in any living complex system via these definitions. Both processes are presumably fundamental to adaptation, and the ways each system encodes representations (models, intelligence) of itself and its environment.
Again, within any particular autopoetic system, evo devo models redefine the word “evolution” to refer specifically to variety-generating, experimental, divergent, and other “soon-unpredictable” processes that generate combinatorial explosions of contingent possibilities. They use the word “development”, to refer to variety-reducing, conservative, convergent, and other “statistically predictable” processes that manage replication. These developmental processes are intrinsically predictable if you have the right models and sufficient computation capacity, the right perspective (often a collective, big picture, or long-term view) or if you have empirical experience, having seen a prior life cycle of the developing system in question (a cell, a tree, a human, a stellar system, a galaxy, a universe).
Independently, and via similar reasoning, some scholars occasionally use the term "evolutionary development" as a replacement for "evolution" as it as it juxtaposes two dynamically- and informationally-opposing concepts — "random", unguided Darwinian evolution and nonrandom, guided development — and thus is a more conservative and humble descriptive term to use when one is uncertain whether the change one is talking about is random or directional. For either reason, in addition to occasional early use by physicists (Turchin 1977) and origin of life scholars (Oparin 1968), a small but growing group of ecologists (Salthe 1993), biologists (Losos 2017), paleontologists (Conway-Morris 2004) theoretical biologists (Reid 2007), cosmologists (Munitz 1987), complexity theorists (Levin 1998) and systems theorists (Heylighen 2007; Smart 2008) find it valuable to use the “evolutionary development” term.
The start of the journal Evolution & Development in 1999 signaled rise of evo-devo biology to a formal subdiscipline. Evo devo systems science and philosophy presently has no journal. If it did, Evolution, Development, and Complexity, the name we use for our Evo Devo Universe (EDU) research and discussion community's Satellite meetings at the Conference on Complex Systems, would be a reasonable title. This would be a journal within which the complexity science and systems theory of such topics as universal Darwinism, evo devo cosmology, evo-devo biology, eco-evo-devo, cog-evo-devo, living systems theory, technological evo devo, artificial and biological intelligence, hierarchy theory, accelerating change, and related topics might be modeled and critiqued. Biologically-Inspired Complexity Science and Philosophy (BICS&P) would be reasonable title for our emerging field itself. BICS&P is the self-description of our Evo Devo Universe research and discussion community (EvoDevoUniverse.com).
Want something fun? Watch this lovely 4 min video, Evo-Devo, by Tim Blais of A Capella Science. It will open your mind to the amazing insights we're learning from evo-devo biology today.
Two polar categories and tensions
Table 1 (Smart 2008) introduces sets of two polar (equal and opposite) word pairs that can be associated with evolutionary and developmental processes in a range of complex systems. As you look them over, think of all the events, processes, and systems you have previously described with these words. These and similar words, and the concepts behind them, are often useful starts at categorizing social, economic, and technological events and processes into one of two camps.
Some systemic processes operate by chance, others by necessity, and some by both. Some processes are random, others predestined. Some events are indeterminate, others predetermined. Some processes are segregating, others integrating. Some act bottom-up, others are top-down. Some systems appear to be branching, others funneling. Some changes look reversible, others irreversible. Some are generating novelty, others conserving sameness. Some are exploring possibilities, others running into constraints. Some promote variability, others stability. Some degrade hierarchies, others create hierarchy. In the organization, good foresight and strategy requires a continual balance between divergent (innovative, experimental) and convergent (predictive, conservative) thinking. We can see these twin tensions, and their mixture, in all the ways humans use for knowing the world.
In the twentieth century, we learned that even our scientific laws fall neatly into these two categories. From our reference frame, not only have we discovered deterministic (developmental) types of laws that precisely describe the far future, like the equations of classical mechanics and relativity, we’ve also found stochastic and statistical (evolutionary) physical laws, like quantum mechanics, thermodynamics, and nuclear physics.
We have learned we can view physical and informational systems as either deterministic or stochastic, depending on the analytical reference frame we adopt. Deterministic laws are highly conserved and predictable at the individual level (i.e., the laws of motion for individual objects), yet become unpredictable at the collective/emergent level (i.e., the N-body problem in physics). Stochastic laws are random, novel, and creative at the individual level (the quantum state or entropy of any particular system, the decay of any particular nucleus), and yet are probabilistically predictable at the collective level. We see a simple example in radioactive half-life, and more complex examples in non-equilibrium thermodynamics, self-organized criticality, and phase transition thermodynamics. Such factors as the reference frame of the observer with respect to the system, the scale at which they are observing the system, and the duration of observation relative to a (presumed) autopoetic cycle all seem to influence the ease and extent of predictability in nature.
Many social, economic, and political processes historically alternate between dominantly unpredictable and divergent (evolutionary) and predictable and convergent (developmental) phases, depending on the process under study (cf. Vermeij 2009). For every social issue, we can find processes simultaneously generating "evolutionary" variety and "developmental" convergence, in comparative analyses of different cities, counties, states, countries, or regions. For example, regarding economic inequality, we find great "evolutionary" variations, country by country, in the levels and quality of social services available to each citizen, and in the cycles of increasing or decreasing inequality. Yet we also find a long-term "developmental" trend of predictably increasing total economic inequality (relative and absolute rich-poor divides) the greater the flow rates of capital, goods, and information in any societies we analyze (Bejan and Errera 2017). The two opposing perspectives and tensions of evolution and development (unpredictability or predictability) appear to be equally fundamentally useful ways to view the world.
Use of the paradox-containing evo devo term also communicates our humility and ignorance when we are asked whether evolutionary (divergent, contingent) or developmental (convergent, inevitable) process are presently dominating in any system or environment. We usually don’t know which processes are most in control of physical or informational dynamics, at first glance. Careful study, modeling, and data collection are often required to see where any complex system is presently headed, process by process.
Contingency vs. inevitability: The two extremes of scientific and societal bias
One might think that the existence of inevitable, developmental physical and informational processes of change is as obvious, in modern scientific practice and philosophy, as are unpredictable contingent, evolutionary processes of change. It has been more than three hundred years since universally inevitable celestial mechanics was elucidated (Newton 1687), a century and a half since we discovered the Second Law of Thermodynamics (Clausius 1851), and a century since Einstein reformulated mechanics into an even more general inevitabilist framework (Einstein 1915), allowing us a deeper understanding of both space-time and energy-matter, and predicting such still incompletely-understood emergent phenomena as black holes, and perhaps even dark energy, via the cosmological constant. These and many other physically well-characterized processes are developmental constraints within which all of life's stochastic evolutionary processes must occur. Surely such examples must lead us to realize that there are likely to be many other statistically predictable macrotrends and inevitable emergences in life, society, and technology, waiting to be discovered and measured empirically, and eventually more rigorously characterized by physical and informational theory and simulation.
Unfortunately, there is presently a strong practitioner and philosophical bias against inevitabilist thinking in most scientific, technical, economic, political and cultural communities, particularly since the rise of chaos theory and nonlinear science in the 1970s, and of subjectivist postmodernism in our academic institutions in the late 20th century. Humanity is guilty of periods and domains of overapplied developmentalist thinking, as in the various clockwork universe models of the 18th and 19th centuries (most famously, Laplace 1812). More recently, many Western nations overapplied reductionist and logical positivist thinking in our think tanks, corporate strategy, and government plans in the mid-twentieth century. The above brief history illustrates that our dominant societal biases have tended to each of two extremes (contingency or inevitability in various human futures) in a chaotic and cyclic dynamic. A few decades or generations hence, perhaps after some particularly predictive scientific or technical advance, we may again swing to the opposite extreme, and adopt an overly developmentalist bias, at least in particular scientific or societal subcultures.
A more adaptive position, rather than swinging to extremes, might be to recognize that both predictable and unpredictable processes are always occurring within any complex system, and to try to better understand each. We can identify at least a few inevitable (developmental) processes, as well as a much larger number of contingent (evolutionary) processes in any complex system we analyze, including the universe as a system. Chaos theory and sensitive dependence on initial conditions apply to some universal processes, but certainly not to all processes. Even our modern philosophy of science, while it acknowledges a "contingentist" and "inevitabilist" debate with respect to the results of scientific experiments (cf. Martin 2013) does not yet acknowledge that both positions are always true, in any complex system, from different perspectives, as we will discuss. More disturbingly, modern science and complexity theory also rarely ask how each apparently fundamental process interrelates, and how each must contribute to selection and adaptation.
Today, we are primarily contingentists, and so we are biased to under-recognize and under-seek statistically inevitable processes of change, and there are social blocks and professional costs to significant inevitablist thinking in the social and technical sciences. For one example of the costs of this bias, consider the following potential sociotechnical developmental process, one that I study and find particularly important. We can characterize a "general Moore's law" of exponentially growing computational capacity per dollar, observable since the 1890s at least (Kurzweil 1999), involving exponential growth in performance and resource efficiency the further our computing processes move into meso-, nano- and quantum-scale realms (Smart 2000). This computational performance and efficiency acceleration via physical miniaturization appears to be a developmental macrotrend in human history, one likely to occur on all Earth-like planets, within some predictable stochastic envelope, if those planets harbor intelligent technology-using life, regardless of their political economies and cultures.
This example of various forms of technical acceleration, one very socially relevant potential statistical inevitability, still poorly characterized in our physical and informational theory, has been largely ignored by academic and complexity science communities alike. Only futurists like myself, and a handful of philosophers and social theorists, seem interested in writing about it and asking about its causal dynamics. I have followed the literature in this area since the late 1990s, and I can assure you that the number of funded science or engineering researchers considering this process is minuscule, even today. Contingentist bias, in my view, is the simplest likely explanation for this state of affairs.
The Santa Fe Institute, a leading US complexity science research organization, tried at least three times (2009-2011) to get the NSF to fund a Performance Curve Database (PCDB) project (see http://pcdb.santafe.edu/), simply to collect better data on predictable exponential trends in technological performance efficiency, to aid in empirical and theoretical models of these fascinating and still accelerating processes. The requested modest funds were denied, and the postdoc leading the grant applications, Bela Nagy, a personal friend, left his scientific career soon afterward, in part due to his disillusionment with the conservative funding priorities of Big Science. The PCDB remains unfunded today, and I know of no other similar project yet in any nation. Perhaps collecting data on technical exponentials, the fastest-changing and most economically and socially disruptive processes in human society today, wasn't considered a high enough priority for the grantors. More likely, NSF politicians didn't want the controversy of being seen as aiding the inevitablist perspective (see Kurzweil 1999 and 2005) on scientific and technical acceleration, due to our current scientific and societal biases toward a primarily contingentist view of social change. I do not know the details, but would be curious to see a causal study done.
Given the reality of contingentist bias, those who write about technological development or accelerating change from a macrohistorical perspective today are often pejoratively labeled as technophiles, utopians, or positivists, when all they are trying to do is establish that both unpredictable evolutionary paths, wherein we must exercise our free moral choice, and predictable yet causally opaque developmental processes, like technical acceleration, and destinations, like societal electrification, digital computers, or machine intelligence, appear to coexist in our complex universe. Our ability to see not only evolutionary change, but also simultaneous processes of ecological, societal, technical, global, and universal development suffers greatly as a result.
The VCRIS dynamical model of natural selection in autopoetic systems
If we wish to understand natural selection in autopoetic systems, both living and nonliving, we must better characterize dynamical change, and develop better theories of information and intelligence. The VCRIS (“vee-kriss”) evo devo conceptual model (Smart 2017a) may be a useful, small step toward these challenges, especially when contrasted to the classic VIST model (variation, inheritance, selection, time/cumulative replication, Russell 2006) of dynamics offered by traditional evolutionary theory. The VCRIS model proposes that three sets of physical and informational dynamics must be modeled to understand and predict the outcomes of natural selection in autopoetic systems. The first two are fundamentally oppositional processes, and the third arises from their interaction. These are:
- Variational or "Evolutionary" processes that generate, maintain and manage diversity, divergence, and experiment. When we observe them from within any autopoetic system, these processes grow increasingly unpredictable over time.
- Convergent or "Developmental" processes that attract, constrain, and guide the system through hierarchical stages of form and function. When we observe them from within any autopoetic system, these processes grow increasingly predictable over time.
- "Evo Devo" processes that are Replicative, with Inheritance of physical and informational parameters, under Selection for adaptation. Selection can favor either or both “evolutionary” (variational) or “developmental” (convergent) dynamics in the replicator, depending on context. Adaptation, in turn, depends on the encoding of information (intelligence) in three places: Replication (organism, autocatalytic) processes, Inheritance (seed, gene, parametric) processes, and Selection (environment) processes.
In the VCRIS model, physical and informational processes that change unpredictably in successive replication cycles, to generate, maintain, and manage Variety, are in fundamental tension and opposition with physical and informational processes that change predictably in successive replication cycles, and thus generate, maintain, and manage Convergence. V and C are the first two terms in the VCRIS model, as these two oppositional processes are proposed as root perspectives in autopoetic systems, including our universe itself, if it is a replicating and adaptive system, as various theorists have proposed (Smolin 1992, 1997, 2004; Vaas 1998; Vidal 2010; Price 2017). Standard evolutionary theory offers no model of this fundamental opposition, of the inheritance and tension between two classes of informational-physical initiating parameters (evo and devo) at every scale at which replication occurs, including gene, epigene, organism, group, niche, environment, and universe.
If our universe is an autopoetic system, the VCRIS model offers us a new term to understand selection, a term that juxtaposes two fundamental binaries, those things that change, and those that converge to stay the same, in any replication cycle. “Unpredictable predictable” is a term a physicist might favor, yet evolutionary development (evo devo) seems more precise, as it uses our model of replication (life cycle) as a way to define those things that predictably stay the same, in prior and parallel life cycles.
In toy cellular automata universe models, like Conway’s Game of Life, the spatiotemporally repetitive structures and dynamics that we see in each successive game (replication cycle), can be defined as predictable, convergent and developmental. Such reliably emergent structures and dynamics are robust to variation in most of the game's initial conditions (occupied configurations within the initiating matrix), yet they are also finely sensitive (finely tuned) to be critically dependent on a few of those conditions, such as the rules of the automata. The morphology and dynamics of other emergent structures in this game are essentially unpredictable, divergent, and can be thought of as sources of evolutionary variety within the game. See Poundstone 1985 for an account of Conway's game from a universal perspective.
In real-world systems, such as individual living organisms, we can observe that the features of two genetically identical twins that look the same are (in theory) predictable, convergent, and developmental. The morphological, dynamical and functional features that are stochastically different, which include their fingerprints, brain wiring, organ microarchitecture, and many (not all) of their ideas and behaviors, are unpredictable, variety-generating (within bounds), and “evolutionary,” in an evo devo model. Most dynamical processes in two identical twins, when we observe them at the molecular scale, appear stochastic and evolutionary. It is only when we look at the twins from across the room (a great increase in observational space and time, from the molecular perspective) that we see a subset of developmental similarities. We will discuss this as the 95/5 rule in the next section, and then consider how it may apply to the universe as an autopoetic system.
In living systems, Selection always appears to involve a majority of “tree-like” evolutionary processes (think of Darwin’s “tree of life”) driven by Variation, and a minority of “funnel-like” developmental processes (any cyclically stable attractors in phenospace) driven by Convergence. From the perspective of information theory, the first process generates new information, and the latter conserves old information, expressed in a prior cycle. These two informational and dynamical processes appear to work both cooperatively and competitively with each other, in service to adaptation. Consider how Replicating organisms are sometimes driven to variation, and sometimes to convergence in both their systems and subsystems. Inheritance units (seeds, genes) sometimes duplicate (think of gene duplication) and vary, and sometimes converge (with gene loss). Selection in the environment sometimes favors creation of phenotypic diversity, and sometimes favors convergence to a particular dominant phenotype. In the VCRIS model, evo and devo (variational and convergent) replication and inheritance under selection are the root source of adapted order.
Perhaps most promisingly, from my perspective, the RIS terms at the center of the model allow us to think of information, learning and intelligence, all processes that may be central to the maintenance of autopoesis, from three separate systems perspectives, that of the Replicator (organism, as an autocatalytic system), the Inheritance system (informational parameters that guide variation and convergence) and the Selective environment (environmental conditions). This seems particularly appropriate, and a clue toward a better autopoetic information theory, as all sufficiently complex organisms (such as any metazoans with culture) appear to store the fruits of their learning and intelligence in these three, partially decomposable systems. In other words, we can say that adapted intelligence (encoded information) in any evo devo system always appears opportunistically partitioned between three complex actors, Seed (inherited parameters), Organism (autocatalytic replicator), and selective Environment (SOE partitioning). Intelligence is never resident in only one of these actors. It always straddles all three (Smart 2008).
For a basic example of Environmental intelligence partitioning, genes use historically metastable features of the local environment to reliably guide the evolving and developing organism to its future destinations. Much information for embryo construction is not specified in the genome, but in the replication-stable features of the environment. For a more complex example, metazoans externalize their intelligence in “niche construction” of their local environment to make it more co-adapted to their needs (Odling-Smee et al. 2003). This process is also called “stigmergy” by scholars (Heylighen 2008, 2016). Niche construction/stigmergy is a key informational process that appears to grow with the complexity of the replicator. It presumably exerts selective pressure toward certain forms of variation and of convergence, in ways not yet well characterized in evo devo theory.
Consider also that environments may also replicate, on some higher systems level, just as organisms and seeds replicate. This happens, for example, when we replicate an urban architecture or idea-complex (like capitalism or democracy) in global society, when stars replicate, when continents drift apart, or if our universe itself replicates. In this model, our selective environment is much more similar to an organism, one fated to produce a new seed or seeds in special high-complexity locales, than is commonly understood in complexity theory.
We may also use the VCRIS model to gain a new perspective on another long-used term in the complexity literature, self-organization. Self-organization is typically defined as the emergence of "spontaneous" order from a previously apparently disordered system. When a complexity theorist uses the term self-organization, they are calling attention to poorly understood, partly hidden ordering processes. In the VCRIS model, these ordering processes must be partly evolutionary (via inherited mechanisms of variation) but largely developmental (via inherited mechanisms of convergence). Both ordering processes interact to produce an autocatalytic life cycle (replication) and both appear to require inheritance factors that are selected upon. These five VCRIS processes, then, are the key ones we must strive to better understand in any autopoetic system.
To understand self-organization, we must find the hidden evolutionary (to some extent) and developmental (to a major extent) dynamics that have been tuned into the initial and boundary conditions of the replicating system, as a result of selection that occurred upon that system in previous autopoetic cycles. For example, when we randomly cut up viral DNA and proteins in a petri dish, and place those molecular fragments in another dish, many fragments will appear to “self-organize” (spontaneously form structure), at a rate much greater than chance. They do so because those molecules have become finely tuned, under prior selection, to use physically and informationally metastable features of the universal environment to produce both contingent evolutionary variety and robustly predictable developmental order (self-assembly), using processes of both bottom-up and top-down causation. In an evo devo universe, such classical self-organization discoveries as Rayleigh–Bénard convection and the Belousov–Zhabotinsky reaction can be called previously hidden, now understood forms of evolutionary developmental ordering. Once we have the appropriate model, such order is no longer spontaneous but becomes predictable, in a broad range of environmental conditions.
If our universe is an autopoetic system, it too must have many such hidden evolutionary and developmental ordering processes at work as well, most of which we do not yet model well. Complexity theorists who argue that self-organization under far-from equilibrium conditions is as much a source of biological complexity as genetic variation and natural selection can be classified as universal evolutionary developmentalists, though they may not self-describe with this term. See Jantsch (1980), Haken (1984) and Kauffman (1993) for three promising yet still early theoretical efforts exploring self-organization from what I would call a universal evo devo frame.
Self-organization is thus a helpful term to remind us that both evolutionary and developmental processes are occurring in any autopoetic system, and I will use it in that sense in this article. At the same time, it should be most helpful to use the full set of VCRIS terms as our models improve, as we should be able to model replication, inheritance, and selection in evolutionary and developmental terms. Again, in the VCRIS model of selection in autopoetic systems, adaptive processes are not called “evolutionary” but rather “evolutionary developmental” or evo devo, to remind us they are always a balance between diverging and converging dynamical processes. This small change in terms helps to correct a bias of standard models, which ignore or minimize convergence, particularly at the level of the universal environment. Even today, the study of convergent evolution (planetary, biogeographic, and ecosystem development) remains controversial and understudied in evolutionary (developmental) biology. This neglect is no longer acceptable, in my view.
Many biologists today would argue that macroevolutionary dynamics are overwhelmingly contingent, diversity generating, and unpredictable. So it is a small change in definition for us to restrict the term evolution to only such processes, within any autopoetic system. Many evolutionary biologists might not like that restriction, but from my perspective, evolutionary biology today offers a view of life and selection that is dangerously incomplete. It has long neglected the physical and informational roles of development in macroevolutionary change, and developmental processes in the selective environment. Fortunately, evo-devo biology is rehabilitating development as a process in living systems. We can hope this will lead us to better see development in the universe as well.
Finally, if autopoesis turns out to be the most efficient and effective way to generate advanced complexity that is both intelligent and stable to time and change, as I presently believe but cannot prove, then it seems most parsimonious to expect both that our future AI must be autopoetic (evo devo) in nature, and that our universe itself is an autopoetic system. Our reality may be, as Rod Swenson (1992) argues, autopoetic “turtles all the way down.”
Evolutionary development in organisms: The 95/5 rule
Since the mid-1990s, the interdisciplinary field of evolutionary developmental, or “evo-devo” biology has emerged to explore the relationship between evolutionary and developmental processes at the scale levels of single-celled and multicellular organisms (Steele 1981, 1998; Jablonka and Lamb 1995, 2005; Raff 1996; Sanderson and Hufford 1996; Arthur 2000; Wilkins 2001; Hall 2003; Müller and Newman 2003; Verhulst 2003; West-Eberhard 2003; Schlosser and Wagner 2004; Carroll 2005; Callebaut and Rasskin-Gutman 2005). Evo-devo biology includes such issues as:
- how developmental processes evolve
- the developmental basis for homology (similarity of form in species with a common ancestor)
- the process of homoplasy (convergent evolution of form and function in species with unique ancestors)
- the roles of modularity and path dependency in evolutionary and developmental process
- how the environment impacts evolutionary and developmental process.
Conceptual and technical advances in scientific disciplines including comparative phylogenetics, morphology and morphometrics, and statistics are allowing better insights into the evolutionary relationships among organisms, and inferences about how developmental processes influence those relationships. The best work in evo-devo recognizes that natural selection is a net subtractive process. Natural selection generates increasing physical diversity, as seen in ever-growing evolutionary "trees", but at the same time, an even greater reduction in potential physical diversity (Johnson 2011).
A foundational concept in evo-devo biology is deep homology. This is the recognition that certain original developmental gene-protein regulatory networks that control growth and differentiation, like the Hox genes and proteins that control metazoan body plans, and the Pax6 genes that control eye morphology and function, are deeply conserved across species, in ways that few evolutionary biologists who adhere to the Modern Synthesis have realized until recently. For example, human beings share these and other developmental genetic networks with flies, and thus we share many developmental quirks and constraints, and even exhibit similar clinical syndromes when these genes are dysfunctional. Lewis Held Jr's three volume evo-devo series, Quirks of Human Anatomy, How the Snake Lost its Legs, and Deep Homology? (Held 2009, 2014, 2017) offers a beautiful introduction to deep homology.
A key insight of deep homology is that these conserved networks are accretive, meaning they represent developmental regulatory systems that generally cannot be removed, only added to with additional regulation. Under selective pressure for innovation, brief episodes of developmental deregulation can be expected to occur. See our discussion of heterochrony and neoteny below. But it also seems likely that all such deregulation, taken too far, will threaten organism viability. Over macroevolutionary time, as organisms grow more complex, we can predict that accretive developmental networks must increasingly limit the kinds of evolutionary change that can occur. For example, the thirty-five metazoan body plans that emerged shortly after the Cambrian are a long-proposed example of such a developmental limitation on evolution. No new body plans have emerged in the half billion years since. Even if other body plans may be theoretically possible, they must have increasingly serious adaptive disadvantages. At a certain point, Earth's niche-dominant forms may competitively prevent further innovation in practice.
Accretive developmental networks must also greatly increase the likelihood of similar solutions to common adaptive problems, even in species greatly separated in evolutionary space or time, when those species are exposed to similar environments. Thus they help us solve the riddle of evolutionary convergence. Due to developmental regulatory accretion, evo-devo mechanisms must become increasingly constrained, able to do less evolutionary innovation, the more complex any organism becomes. This growing constraint can be opposed temporarily, as when the Hox genes were duplicated in metazoans to create an exciting new era of unpredictable brain plan variation within fixed body plans. This variation eventually led life to human consciousness. But over time, evolutionary options must be increasingly reduced as developmental complexity grows, until the entire system is capable of resetting itself. Our presumed future migration from biological to technological life may allow a deep homology reset, to some degree, but perhaps less than we first imagine. If physical development itself must follow certain universals in the creation of morphology, and if evo-devo approaches will naturally win in AI, as I suspect, then accretive developmental constraint can also be expected to occur in the minds and bodies of our AI successors as well.
In this context, the fundamental role of evolution can be hypothesized as cumulative mechanisms that generate experimental ("good bet") types of diversity, to improve the odds of survival under environmental selection. Evolutionary systems harness stochasticity in an increasingly information-driven and intelligent way as organic complexity grows, but evolutionary innovation itself is largely unpredictable (Shapiro 2011; Noble 2017). Living systems continually sense their internal states and environment, and they react to catastrophe and stress with bursts of such poorly-predictable, information-driven innovation (a divergent form of "intelligence"), a pattern some evolutionary biologists call punctuated equilibrium (Eldredge and Gould 1972).
The fundamental role of development can be hypothesized as cumulative mechanisms that conserve and execute a small subset of (in-principle) predictable processes that have worked in the past to guarantee replication, under a range of chaotic internal and external environmental conditions. Developmental systems encode future-predictive probabilistic models of themselves and their environment, models which we assume follow the rules of Bayesian probability in nervous systems and presumably even in single celled organisms. Developmental prediction (a convergent form of "intelligence") is generated from special initial conditions (developmental genes), tuned via informational constancies that exist in genes, developing organisms, and the environment.
The theory of facilitated variation (Gerhart and Kirschner 2005; 2007), in which the genetic processes in living systems are assumed to sort into two groups, a conserved core, which regulate critical elements of development and physiology, and a set of changing genetic elements, whose variation is “facilitated” by the conserved core, presumably in ways that both reduce the lethality of experimental change and increase the utility of genetic variation (“experiments”) subsequently retained by populations, is a model consistent with this view. In evo devo language, the conserved core are conserved developmental genetic, allelic, and epigenetic processes, and evolutionary genetic processes are those that facilitate genetic, allelic, and epigenetic variation within and across generations. Such processes presumably act in tension with and opposition to each other in very fundamental informational and dynamical ways.
In this model, natural selection can be argued to be a composite of two more fundamental kinds of selection. Evolutionary selection biases the system toward potentially useful, yet also potentially dangerous, intelligence-guided innovation and disorder when needed, and developmental selection biases the system toward convergences and order that have historically allowed complexity conservation and replication. In this view, we must see both of these selective and often opposing processes, apparently at work at many scales in every system that replicates, to truly understand biological change. For example, we should be able to identify both structure-or-function-divergent and structure-or-function-convergent classes of gene flow operating between species in the terrestrial biosphere, via such processes as genetic drift and horizontal (or lateral) gene transfer. Such transfer is well documented in Prokarya, and has been greatly facilitated by viruses in Eukarya (Zimmer 2015).
One of the clarifying features of developmental selection is that it is always critically dependent on a small subset of control parameters (in biology, genes and other regulatory molecules). While about half of metazoan genes are expressed in such processes as organ development, less than 20% of these (thus less than 10% of all genes) are substantially regulated during expression (Yi et al. 2010). A further subset of our genome, roughly 5% of DNA in human, mouse, and rat, is highly conserved across these and other metazoan species. This 5% of our genome typically cannot be changed without stopping, or causing major deleterious effects to, processes of development. The majority of this highly conserved DNA, 3.5% of our genome, is noncoding, yet presumably also constrains functional expression (Wagman and Stephens 2004). A subset of this conserved DNA is sometimes referred to as the developmental genetic toolkit (DGT), or less accurately, the evo-devo gene toolkit. These genes include the Hox genes which determine animal body plans, and they often involve initial symmetry breaking choices in spatial, dynamical, and informational form and function that commit the organism to a particular developmental path. Thus a subset of all metazoan genomes have become very finely tuned, over many past replications, for the production of complex, path dependent modularity, hierarchy and life cycle. Presumably, the other 95% of these genomes can change and generate diversity without such immediately deleterious effects.
Consider also Cope's "Law of the Unspecialized", which tells us that during catastrophes, overly specialized (overly adapted) forms of life will typically become become extinct due to their "overperfection" (essentially, overadaptation and overdevelopment). Cope's law is not the same as Cope's Rule of body size increase (for a review of both, see Raia and Fortelius 2013). The dieoffs that have occurred in our great past catastrophes (Permian, K-T, etc.) have presumably selected for significantly more developmentally unspecialized (more totipotent) forms, and this in turn has catalyzed rapid leaps (punctuations) in evolutionary innovation afterward. I call this the catalytic catastrophe hypothesis. It argues that there is a "conserved core" of developmental genes and regulatory networks (roughly speaking, "the 5%") that are, perhaps counterintuitively, reinvigorated by catastrophe. Cope's law reminds us that development's conserved core of regulatory networks must include capacities like neoteny (the ability to revert an organism back to a more juvenile form) and heterochrony (slowing down and speeding up the timing of developmental events). For example, it has long been noted that primate juveniles (apes, chimps) have a similar cranial anatomy to adult Homo sapiens. Thus one key to humanity's great leap forward, in addition to growing social complexity and selective pressures, may have been a developmental leap backward (neotenizing our neurological development) for our primate ancestors, giving our children a longer altricial period (mentally immature and dependent state) that would facilitate the more rapid and valuable innovations of cultural evolution.
Thus all genomes can be categorized into two groups, of conserved and non-conserved genes, and we can propose that all highly conserved genes which are also highly tuned (highly sensitive to change, with deleterious effect) are the core constraints on development itself. I call this observation the 95/5 rule, and have found early evidence for it in replicative systems at a wide variety of scales (Smart 2008). The rule proposes that some small subset of developmental parameters are always top-down causal, involving essentially one-way information flow (in this case, developmental genes to organism). They can no longer be easily changed, they can only be added to, as organisms get more complex. The remainder of the genome can be considered evolutionary, whether it controls evolutionary or developmental process, as all of those genes can be altered by two-way information flow with the environment, with feedback. But per the 95/5 rule, a small, and highly-tuned set of top-down constraints must always exist, in any evo devo system.
There are a variety of levels of biological hierarchy at which evo devo concepts can be applied, and evo-devo biologists believe developmental processes and genes must themselves act to constrain evolutionary processes, in ways not yet understood by traditional evolutionary theory (notably the neo-Darwinian Modern Synthesis), and that both evolutionary diversity and developmental constraint are important to understanding long range “macrobiological” change (Pigliucci 2007; Pigliucci and Müller 2010). Evo-devo genetics and epigenetics are rapidly improving fields, and they promise to improve our understanding of such complex yet central topics as biological constraint, adaptation, intelligence, and convergent evolution.
The riddle of development and the challenge to cosmologybiological development, including genetic, embryonic, organismic, and psychological development. How is it that developing organisms can reliably converge on far-future form and function (from the molecular perspective), under chaotic and variable environmental conditions? How is this done with just a small percentage of highly-conserved developmental genes? Development employs stochastic, contingent, and selectionist processes, presumably ranging from quantum to macroscopic scales, in service to statistically deterministic, modular, hierarchical and cyclic emergent change, from conception to organism, and from organism to reproduction, senescence, and death (recycling) (Salthe 2010). Our mathematical models of development are incomplete today, but they continue to make progress. Our models of evolution, of random genetic reassortment and selection in populations, are much more advanced. Development also involves teleology, or the assumption of goal-driven, end-seeking behavior, including successful replication. For these and other reasons, most scientists have focused on the idea that our universe may be evolving, while ignoring the idea that it may also be developing. This oversight, more than any other, has motivated the creation of the EDU community. The great challenge to cosmology today is to change this state of affairs, to learn from biology to better understand universal change.
Biologically-inspired hypotheses for cosmological change offer us a number of predictive models of the dynamics and large-scale properties of the universe. This is necessary for establishing the potential value of ED as an explanatory approach, but only falsifiable predictions can establish (or negate) its legitimacy. Unfortunately, falsifiability is not easy in our present level of cosmological understanding. Whenever these hypotheses appear (and we shall see some below) we may need to investigate many details before concluding that that the hypothesis is impossible or unfeasible. In such circumstances, it is best not to jump to negative conclusions on the basis of the greater familiarity that science has had to date with mechanistic, bottom-up reductionism.
Ever since Plato, scholars have occasionally compared our universe, in some ways, to a living organism. If evo devo models are correct, this organicist philosophy may be true in part, but we should also expect this analogy to be both overly and poorly applied at first. Fortunately, the rise of bio-inspired design, and the recent successes of bio-inspired approaches in deep machine learning, are showing the value of generalizing organic structure and function to other substrates. As our understanding of biological development grows, and we gain the ability to predict developmental outcomes in embryogenesis via partial dependence on top-down parameters like developmental genes, our understanding of causality will improve, and so too will our cosmology. Fortunately, a subset of scholars continue to call for more holistic and top-down approaches to understanding universal change (e.g., Vidal 2010; Ellis, Noble, and O’Connor 2012; Ellis 2015; Adams et al. 2016). The great challenge we have is in learning how to blend our best top-down and bottom-up perspectives.
Predictability, convergence, and constraint help explain our universe, but these concepts only take us so far. Consider symmetry. Discovering hidden symmetries underlying physical reality has been tremendously helpful in building our standard model of physics, allowing us to understand conservation laws, Maxwell’s equations, the electroweak interaction, and predict fundamental particles like the charm quark. Exploration of the symmetries of very high dimensional shapes, like the Lie groups, may uncover a constraining relationship between our universe’s forces and particles. But our attempts to use supersymmetry to arrive at a single “theory of everything” for our universe, and to make verifiable predictions in our particle accelerators, have stalled. I do not expect such a single theory exists, and would predict that supersymmetry, or any other fully top-down, constraint-based model, will never be enough to explain reality.
Our universe also seems to use unpredictability, divergence, and freedom just as fundamentally. Besides quantum theory, two other useful physical theories, eternal inflation and string theory, each offer a mathematics of diversity and unpredictability, in which our universe is just one in a multiverse of possible universes, and our fundamental parameters alone cannot fully specify all the features of this universe. Some scholars propose that these new multiverse models imply we live in an “accidental” universe (Lightman 2011), and that our ability to understand our universe in terms of fundamental principles, at a level below this essential randomness, is, like a fully deterministic (non-statistical) understanding of quantum theory, an objective we will never achieve.
I am sympathetic to both of these views, and expect each will continue to make progress, while each alone will remain incomplete. If we live in an evo devo universe, where universal dynamics and informatics have proceeded something like biology has, from simple to more complex, over many past cycles, then something like the 95/5 rule should apply. Our universe will increasingly be understood as both “accidental” and “purposeful.” While the vast majority of our universe’s mathematics will have this random, accidental, and evolutionary looking nature, we will also continue to discover a growing subset of top-down, constraining maths and causalities that guide our universes critical processes of development, including complexity and intelligence production, conservation, and replication. Some combination of environmental selection (for adaptation) and self-selection (for intelligence) should apply. We need to get smart enough to see both classes of process, and ask how they each relate to and support the growth of useful (intelligent, adapted) complexity.
Like living systems, our universe broadly exhibits both stochastic and deterministic components, in all historical epochs and at all levels of scale (Miller 1978). It has a definite birth and it is inevitably senescing toward heat death. The idea that we live in an “evo devo universe,” one that has self-organized over past replications both to generate multilocal evolutionary variation (preselected diversity), and to convergently develop and pass to future generations selected aspects of its accumulated complexity ("intelligence") is an obvious hypothesis. Living systems harness stochastic evolutionary processes to produce novel developments, especially under stress, in a variety of systems and scales (Noble 2017). If our universe is an adaptive replicator, it makes sense that it would do the same. Yet very few cosmologists or physicists, even in the community that theorizes universal replication and the multiverse, has entertained the idea that our universe may be both evolving and developing (engaging in both evolutionary innovation and goal-driven, teleological, directional change and a replicative life cycle).
There is a reasonable frequency of discussion, in the cosmology and astrophysics literature, of the idea of universal evolution. But none of it takes an evo devo approach. We find plenty of random, Monte Carlo models of change, applied to our universe's initial conditions (e.g. various chaotic inflationary multiverse models; Linde 1992), but no models in which adaptive complexity and modeling intelligence emerges via evolutionary development in replicating universes in the multiverse, just as it does in all living replicators, and in several nonliving ones, such as hierarchical prebiotic chemistries on the path to RNA, and hierarchical populations of increasingly chemically complex stars. Even our best current models of universal replication, like Lee Smolin's cosmological natural selection, do not yet use the concept of universal development, or refer to development literature, or to any theories of intelligence. Yet intelligence and its causal implications are an emergent property of all organic replicators, and if our universe is a replicator it is reasonable to expect universal intelligence must be accounted for in our future cosmology, as we will describe.
In the same fashion, a handful of our universe's fundamental parameters appear breathtakingly finely tuned, in their mathematical values, for producing stable, long-lived, complex universes (Barrow and Tipler 1986; Rees 1999; Smolin 2006, 2012). If our universe is an adaptive replicator, under some sort of selection (either self-selection or environmental selection), the most parsimonious explanation for our universe's incredible developmental fine tuning would be past universal replication, with both optimization and path dependency of developmental parameters (conserved inheritance) aiding in universal complexity survival and adaptation. Virtually all known or proposed intrauniversal complex adaptive systems are replicators, with the exception of galaxies, which presumably replicate as dependents on their universes (Smart 2008), so it is conceptually the simplest to infer that the universe is also such a system, in my view.
In living systems, developed properties like intelligence, immunity, and morality strongly alter previously locally contingent environmental selection processes toward organism improvement and survival. See Corning 2018 for a nuanced argument that synergy (cooperative competition, interdependence) is central to adaptive selection in all intelligent systems. If our universe is a replicating system under selection, it is a reasonable hypothesis that aspects of its internal adaptive complexity, intelligence, immunity, and morality may not only be evolutionary (stochastic, unpredictable) but also developmental (fine-tuned, predictable) as well. Yet at present, the scientists exploring the fine-tuned universe problem presently do not consider explanations in terms of universal development. Instead, we find fine-tuning research disproportionately dominated by intelligent design creationists championing the idea of fine-tuning as "evidence for God", leading to much confusion in professional and lay circles.
Perhaps as a result, the field remains professionally controversial for orthodox science, and a minority of astrophysicists, seeking to debunk theists, argue that fine-tuning, beyond the weak anthropic principle (observer-selection effects) doesn't exist (Adams 2008; Stenger 2011; Carroll 2016). But since the anthropic principle was first clearly articulated in cosmology (Carter 1974), another community of scientists have offered their own reasonable evidence that such tuning appears baked into our standard model of physics and empirically observed cosmology. In recent decades, fine-tuning explanations are commonly done via appeal to the multiverse. Among multiverse models, the hypothesis of universal evolutionary development offers a naturalistic explanation for fine-tuning that is homologous to biological fine-tuning. It deserves elucidation and critique.
The primary bias that exists in our cosmological models today is not observer selection bias, which is real but overrated. The primary bias at present is our failure to consider the concept of universal development, the idea that our universe's special initial conditions and stunning internal complexity are likely self-organized, via evolutionary development, just as our initial conditions and complexity have self-organized in all living systems. If our universe is a replicator, then evo devo self-organization is the most parsimonious explanation for the surprising levels of fine-tuning, massive parallelism, and fitness for life we find in our universe, not randomness alone, and not "design." See the fine-tuned universe hypothesis - early evidence for universal evolutionary development below for further discussion.
In the meantime, our leading theories of universal change are presently missing the concept of evolutionary development. If our universe is an evo devo system, then cosmologists, astrophysicists, geochemists, planetary scientists, astrobiologists, information theorists, philosophers, Big Historians, anthropologists, sociologists, and scholars of long-range biological, social, and technological change will have to update their models of the future. For more on this perspective, see Humanity Rising: Why Evolutionary Developmentalism Will Inherit the Future (Smart 2015).
Do we live in an evo devo universe? The EDU hypothesis
All replicating complex systems can be viewed from two fundamentally different perspectives. When we look at the system up close, whether it is a star, a prebiotic chemistry, a cell, or an organism, we see much that is locally unpredictable. Yet when we observe the same system either at a larger scale, or over a longer time frame, long enough to see its replication cycle, we see much about it that is predictable--even when we don't yet know any of the math or causal forces behind its predictability.
Think of an acorn. Once you've seen one acorn grow into an oak tree, you learn that the shape of the acorn seed tells you that it will make an oak tree, with its characteristic leaves, morphology, and behavioral proclivities. Once you've planted more than one acorn, you know, in advance, that most of the structural and molecular details of each oak tree will remain contingent, "random," and unpredictable. But you also know much about its future that is predictable. That predictability, in biology, is called development. The unpredictable, diversity-generating parts we can call evolution, in an evo devo model.
If our universe is a replicating system, it is very much like an oak tree, moving from a highly defined initial seed, to a very flexible, undifferentiated, and totipotent embryo, as we see in our universe's pregalactic era, then to an increasingly specified and constrained set of outcomes, like the increasingly terminally differentiated structure of the oak tree, or the terminally differentiated tissue types that emerge in a developing embryo. Some scholars have represented the latter as a "tree of differentiation" (picture right), a developmental counterpart to the evolutionary "tree of life". The more we learn about the shape of the seed that created our universe, its nurturing environment (multiverse), and the "organism" itself, the more we'll know about both our evolutionary futures--what will stay unpredictable, and about our developmental future--what predictable and constraining "portals" and "terminal destinies" lie ahead, both for us and for all intelligent life.
The massive scale and isotropy (parallelism) of our particular universe, and its severe migration and communication constraints, can also be suspected, given the presumably sharply limited complexity of each local intelligence (constrained by physical law), to have been self-organized by the universe to maximize the local evolutionary variety of each intelligence prior to contact (Smart 2008). If universal intelligence plays a nonrandom role in universal replication, as it does in living systems, a bio-inspired case for the emergence of our kind of massively parallel yet apparently intelligence-compartmentalized universe can be made, as well as the prediction that a mechanism must exist for all end-of-universe intelligences to eventually be able to compare and contrast their computationally incomplete yet usefully locally unique models of reality. If future intelligences can survive a black hole transition, a number of arguments can be made that black holes themselves may uniquely offer such a merger and selection mechanism, in what I call the transcension hypothesis for universal intelligence (Crane 1994; Harrison 1995; Smart 2008, 2012; Vidal 2008, 2016).
Our intelligence allows us to take these multiscale and macrotemporal views on our reality, even as we are physically stuck in one small corner of our universe. All the universe's most complex bits are curiously isolated, by astronomical distances, and thus each is constrained to follow its own unique evolutionary path toward common developmental destinations. When viewed from a cosmic perspective, we can also see that our computers are rapidly becoming the new leading local intelligence on our planet. They may soon (perhaps even this century) exceed us in their adaptiveness, immunity, and intelligence. Such intelligences may be immune to environmental catastrophe, able to exist in near space, fully independent of our planet's nurturing environment. Using nuclear fusion technology, they would not even require our sun for energy. Using quantum computation, these intelligences might even function best in the cold environment of space. This accelerating transition to a new level of hierarchical complexity (and presumably, consciousness) may predictably occur on all planets that harbor intelligent biological life.
The evo devo universe (EDU) hypothesis (Smart 2008) proposes that our universe has two fundamental drives, to evolve (vary, diverge, create, experiment) and to develop (converge on a predictable, information-conservative hierarchy and life cycle). In the VCRIS model, the adaptive intelligence of any replicating complex system lives in, and is opportunistically partitioned between, at least three physical and informational actors: the initiating Seed, the Organism, and the selective Environment.
The cosmological natural selection (CNS) hypothesis (Smolin 1992, 1997, 2004), in which our universe replicates via black holes, with random reassortment of fundamental cosmological parameters at each replication, is one such evo devo model. In CNS, black holes can be considered the "seeds", the universe the "organism," and the multiverse the "environment." CNS remains controversial (see Vaas 1998 for one review of questions to be resolved). See Gardner and Conlon (2013) for an evolutionary biological approach to CNS using the Price equation to model selection for black hole replication. In my view, CNS as a model for an evo devo universe is an auspicious start, but has at least two shortcomings (Smart 2008).
First, CNS currently predicts that universes which replicate via black holes would select for a maximum of progeny (black holes), when real biological replicators always balance replication fecundity with other adaptive goals, including the resource cost to add somatic complexity to the organism (the universe, in this case). In a biologically analogous evo devo model, the qualities of the soma (universe), of the seeds (black holes) and of the environment itself (multiverse) can all be modified, both randomly (evo) and predictably (devo), and increasingly intelligently in more complex replicators, to make the system more adaptive. Some critics of CNS who state that our universe doesn’t appear to maximize black hole production have assumed this insight makes the theory invalid, when in fact, any adaptive theory would rarely argue for black hole maximization.
Second, CNS has a very incomplete selection function, which does not yet account for intelligence (modeling ability) at any level. CNS assumes a random reassortment of our universe’s fundamental parameters at the replication step, but this model is not appropriate even for the simplest biological replicators, as all living systems encode a kind of world modeling (intelligence) in both their evolutionary and developmental gene complexes. Genes reassort nonrandomly, based on developmental constraints. In higher complexity systems like human civilizations, consider the way ideas replicate in communities of brains. Idea replication is not random, but is increasingly selected by the intelligences responsible for modifying and passing them on. As internal intelligence grows in any replicator, it seems increasingly hard to neglect, in any good model of selection.
In my view, a theoretical framework we can call CNS with Intelligence (CNSI) (Smart 2008; Price 2017) will be necessary, if we are to use CNS to causally explain the roles of life and intelligence in our universe in coming years. Adding intelligence to our selection function allows us to consider to what extend the parameters of any seed have been self-selected (e.g., for greater capacity to simulate, and to engage in replication, independent of multiversal environment) by the growing evolutionary and developmental intelligence of the replicator itself. I will offer one such speculative model (Five goals of complex systems, Smart 2017b) for such self-selection later. At the same time, we must also consider if and how our universe’s parameters have been environmentally selected, in some specific multiverse context, as we would do in a conventional Darwinian view of selection. For more on CNS and CNSI, see the EDU wiki page “cosmological natural selection (fecund universes)” (Wikipedia 2008, and Smart and Vidal 2008-2017).
If our universe has these general similarities to living systems, and is subject to selection, in some fashion, either self-selection or selection in the multiversal environment, we can predict that development at all system scales (organismic, ecological, biogeographic, cultural, technological, universal, etc.) will act as a constraint on evolution at all system scales. Likewise, we can expect that evolution, via preferential replicative selection, will continually and slowly change future development, again at all scales.
I expect a future information-centric theory of adaptation will find a number of evo devo processes (goals, values, drives, abilities) that are widely shared by complex systems. I can imagine, but not validate, one such speculative evo devo model, which we will see later. If we live in a noetic (information and intelligence-accumulating) universe, we may need such normative (goal and value-based) models to understand the way the growth of information and modeling abilities change complex environments. Modern hypotheses on how top-down causal information (Walker et al. 2017) and niche-constructing intelligence (Odling-Smee 2003; Heylighen 2016; Noble and Noble 2017) constrain and direct innovation and selection in biology are an important start in this direction.
The better we understand the evo and devo roles for information-driven processes in bioadaptation (and today we often do not) the better we may understand their adaptive role for the universe as a replicator. When we discover and validate evolutionary process and structure, we can better describe innovation possibilities for complex systems in our universe. Likewise, when we find and model developmental process, we can predict or guess developmental constraints on those systems, and where they are striving to go. Most auspiciously for our moral and intellectual lives, we can better understand more of the evo and devo “purposes” or “telos” for ourselves, our societies, and the universe. We can recognize our natural drives to pursue both evolutionary goals (e.g., to create/ innovate/ experiment) and developmental goals (to conserve/ sustain/ discover), and seek to harness these two apparently fundamental processes to greater individual, organizational, and societal adaptiveness (Smart 2017a).
The fine-tuned universe hypothesis: Early evidence for universal ED
The fine-tuned universe hypothesis (Rees 1999, 2001) can be best understood as an important and early example of universal evolutionary development. In most organisms, you can change many genes and generate phenotypically different organisms, but they will still develop. We can call those “evolutionary” genes. But there is a subset of genes that are highly conserved in evolutionary history, and highly resistant to change. Nudge them just a bit, and you don’t get viable development.
In the same way, while our universe and multiverse simulation capacity are still emerging (picture left), and our physical and informational theories are not yet complete, we know that among the known 26 or so fundamental parameters of our universe, most can be changed and simulations will still produce viable universes (Smolin 1997). We can call these the universe’s “evolutionary” parameters in its initiating “seed” or “genome” in an evo devo model. At the same time, there are a special subset of parameters that seem improbably precisely tuned (one, the cosmological constant, apparently even to 120 orders of magnitude), to work with the other finely-tuned parameters to produce universes capable of rich internal complexity and longevity.
When we nudge any of these precisely-tuned parameters in our simulations, we don’t get viable universes. We can call those “developmental” parameters, in an evo devo model. If our universe replicates, they seem homologous to the small subset of developmental genes in organisms. Edit any of those parameters and you never get viable organisms. They've been self-organized, over vast numbers of previous cycles, to work together to conserve the developmental forms, functions, hierarchies, and life cycle of the organism.
The proposition that our universe's laws are finely-tuned for various evo devo outcomes can be made in a variety of ways. These claims can be considered various forms of the anthropic principle, and anthropic reasoning its field of inquiry. In its most useful variation, the anthropic principle is the idea that our universe's initial conditions and laws seem improbably biased toward the production of intelligent observers (Barrow and Tipler 1986). But there is a more fundamental bias that must be considered when evaluating fine-tuning models, the bias that results from observer-selection effects. Is the particular kind of physical and mathematical universe we live in logically necessary, if there are intelligent observers around to ask questions about it?
We will discuss anthropic selection effects in detail in a later section, but for now, let us make just one potentially useful observation. It does seem plausible that we must have a quantum universe in order to have a universe with observers. So that level of observer-selection bias does seem to exist, at least. In perhaps the best-known "weirdness" of quantum physics in the standard (“Copenhagen”) interpretation of the wave function, we observers alter physical reality (quantum states) by the manner in which we choose to observe them. The apparent necessity for quantum physics in our observable universe in no way tells us that fine-tuning doesn't exist. If anything, it could point to necessity of some version of the “co-evolution” between universe and observers, perhaps as first sketched by John Archibald Wheeler (1977, 1988). The kind of quantum physics we have, and its relation to the rest of our physics, may be tuned for the necessary emergence of not just observers, but of intelligent observers, of "mind." Quantum physics doesn't presently integrate fully with other physical and mathematical features of our universe, such as the fundamental parameters, general relativity, symmetry, information, and meaning (whatever that is). So we don't know yet what our most fundamental universal theories are. As we don't yet have the ability to definitively answer such questions at present, the fine-tuning debates will continue, and continue to be productive.
Most physicists were strongly opposed to the teleological (purposeful, directional) idea of fine-tuning when the Barrow and Tipler book emerged in 1986. Today, many leading physicists, like Leonard Susskind (2006) and Steven Weinberg (2007), now argue that multiverse models offer us the simplest explanation (principle of parsimony) for the mathematically improbable levels of fine tuning we find in several of our fundamental physical constants, and as a result, in the strengths and nature of the four forces in our Standard Model of Physics. So we have seen a shift of many leaders in the physics community toward multiverse explanations of fine-tuning, and thus an implicit recognition that our universe's apparent fine-tuning is a real problem that must be addressed.
In the most dramatic example presently known, the empirically observed value of our universe's cosmological constant appears to be tuned to one part in ten to the power of one hundred and twenty. In current models, any imperceptibly small change in that constant would lead to either near-immediate collapse or destructive inflation of the early universe. Likewise, Planck's constant, the gravitational constant, the neutron-proton mass difference, the strengths of electromagnetism, weak and strong nuclear forces, the masses of particles in early inflation after the Big Bang, and several other aspects of our physical universe appear fine-tuned for the production of long lived universes that support high levels of emergent complexity.
Assuming other values of these constants are possible and would lead to alternative universes, there are many improbable transitions and architectures to explain, including the special subatomic physical resonance (quantum "fine-tuning") we call the triple-alpha process, famously predicted by astronomer Fred Hoyle (1954), which produces an abundance of carbon and oxygen in our particular universe. Recent calculations (Meissner 2013) continue to support the hypothesis of fine-tuning in the fundamental parameters of quantum chromodynamics and quantum electrodynamics for this fortuitous result to occur. Carbon, oxygen, and a handful of other elements (HNPS, and some metal cofactors) are developmental portals (unique gateways on the molecular phase space landscape) to redox organic chemistry, which is the structural and energetic foundation of life. For another potential fine tuning example, consider the way dark matter and a smattering of older Population II stars, often presenting as globular clusters, form an elliptical "halo" galactic superstructure, one that allows newer Population I stars to precipitate into elegant planar spiral and elliptical galaxies. The newer stars' rotation and metallicity gradients are apparently created and maintained by this halo dark matter distribution across vast ranges of space and time, giving the mature form of our most chemically complex galaxies the rough appearance of a complex biological development, like an ovarian follicle. Consider also the curiously scale free and organic looking appearance of the large scale structure of matter distribution in our universe (picture right).
We also can suspect a variety of improbably life-generating conditions on the early cooling Earth, whose geochemistry may be catalytically optimized for life’s emergence, including a predictable distribution of mineral cofactors able to catalyze the rTCA cycle on metal-rich planets around Population I stars, perhaps one of several critical preconditions for life (Smith and Morowitz 2016). Life appears to have arrived on Earth almost as soon as our crust sufficiently cooled. Consider also the curiously biphasic nature of Earth's crust. It consists of a denser, continually recycling oceanic crust, which regulates CO2 and other atmospheric parameters via plate tectonics, and a lighter, continually floating continental crust, which offers a stable nursery for the growth of land-based life. Stable continents, in turn, may be a developmental portal for the first emergence of complex social mimicry, language and tool use on any Earthlike, given the many physical advantages of air over water for such forms of social intelligence growth. We can also identify, with varying degrees of controversy, several life-stabilizing biogeohomoeostatic features (e.g., the Gaia hypothesis) in various atmospheric and ocean properties on our current Earth (Volk 2003). All these tunings and others may be necessary for robust phase transitions to higher complexity, in special domains of space and time, allowing life to emerge, diversify, persist, and grow more complex and generally intelligent at its leading edge of evolutionary development.
But there's even more to explain, because not only does our universe support improbably high levels of emergent complexity and mind, it supports an even more improbable condition of continuously accelerating complexification in special environments, an acceleration that seems increasingly self-stabilizing under periodic, and often catalyzing, episodes of selective catastrophe (Smart 2000, 2008, 2012). Konrad Lorenz (1977) was an early advocate of the view that both energy transfer and information processing must work together to create the mode and tempo of biological change. If our universe is a replicator, we can expect both physical and informational causes are needed to explain its accelerative aspects as well. Chaisson (2001), Aunger (2007a-b) and others have proposed that it is the increasingly intelligent control of energy flow that drives structural-functional acceleration in our universe. Chaisson has estimated exponentially increasing energy flow density (free energy flow per gram or volume) in a special subset of complex adaptive systems over universal time (figure below left). Processes like galactic structure formation, stellar nucleosynthesis, and redox organic chemistry are themselves accelerative, in free energy flow measures, over previous complex systems, and each may be developmental portals (unique gateways) to further structural and functional complexification and intelligence growth. Life's accelerating complexification, in turn, has reliably produced a variety of social tool using species, and in humans, accelerating intelligence, immunity, and (though it is often debated) morality in recent millennia.
Curiously, our leading technology, digital computers, have a free energy density control rate that is now at least a millionfold faster than our biological neurons. This differential has grown exponentially over our "Moore's law" era of computing, and may grow by many additional orders of magnitude as we shift to future even more miniaturized, dense, and complex architectures and technologies including massive parallelism, single electron transistors and optical and quantum computing. I call this process of accelerating complexification "STEM compression" (Smart 2002, 2008, 2012), with "compression" referring to predictable growth in both physical and informational density and efficiency of critical spatial, temporal, energetic, and material (STEM) metabolic, effector, and thinking processes in our most dominantly adaptive systems over time. I consider energy flow density acceleration (e.g., Chaisson 2001) to be just one example of this apparent universal trend. Furthermore, now that our leading computers are using biologically-inspired algorithms, and are developing increasingly general forms of intelligence, the adaptive goals they learn from their environment should be similarly accelerated, particularly if we can intelligently aid this apparently natural process. Later in this paper I will propose five learnable goals (abilities, drives, ends, telos) that seem particularly universally adaptive and self-stabilizing for intelligent complex systems, if they are built from both evolutionary and developmental processes.
Empirically, this record of growing internal control and self-stabilization, and increasingly general adaptiveness of our most complex systems, which we can argue exists in the geophysical processes of Earth, in life, and in human civilization and our leading technology, seems unlikely in a randomly generated universe. Why hasn't our universe been far more disruptive to our general record of accelerating complexification? Some kind of developmental immunity may exist, tuned into the developmental components of our universe's genes, soma, and environment, if our universe replicates with inherited characteristics, and if the acceleration of complexity and/or intelligence have had some past adaptive (selective) value. Immunity is a challenging and slippery concept to define. Consider the special nature of the developmental genes, regulatory networks, cellular capacities, and lymphatic architecture that produce a scale invariant immune response, confronting pathogens nearly as fast in an elephant as in a mouse (Banerjee & Moses 2010). Most other physiologic systems do not show this scale invariance.
In intelligent collectives, we also need to understand developmental morality (ethical and empathic interdependence) to stabilize growing individual intelligence. If we come to understand the inevitability of these processes in biological systems, we may come to understand them in replicating cultures and their technologies as well. Early developmentalist models of social immunity (stability) and morality (virtue) were championed by such 19th theorists as August Comte (1844) and Herbert Spencer (1864). In the 20th century, the priest-paleontologist Pierre Teilhard de Chardin (1955) was perhaps their most famous advocate. In recent years, a few psychologists have offered us statistical arguments that both the average severity and average frequency of global social violence have substantially declined over human history, even as our potential for committing acts of violence at scale, via science and technology, has steadily grown (Pinker 2010). Causal models for this decline are still lacking, but Pinker is clearly arguing for both an evolutionary and developmental morality, and I believe our collective morality must grow predictably in nuance, force and scale on all Earth-like planets if intelligence is to be stabilized as complexity accelerates.
We may also need to explain the impressive simplicity and comprehensibility of (most of) the mathematics that underlies nature. According to Leslie and Kuhn (2013), Gottfried Leibniz (1686) was perhaps the first to argue that while some mathematical equation could be found to fit any curve one might draw, the vast majority of the set of possible curves and equations would be exceedingly complex. Similarly to Leibniz, Vilenkin (2006) argued that one would expect "horrendously large and cumbersome" mathematics underlying a typical randomly derived universe in a multiverse ensemble. Yet the applied mathematics and physics that our minds can understand seems unreasonably effective for both scientific modeling and technological development (Wigner 1960).
We should also explain why our universe appears to use massive parallelism in its production of intelligent civilizations, and keeps them spatially separated for the majority of their evolutionary development. This is related to the Fermi paradox, namely the observed absence of extraterrestrial intelligent beings and their artefacts in our past light-cone (Brin 1983; Webb 2015), even though we are likely to have emerged one to three billion years later than other Earth-like planets further in on our galactic habitable zone (Lineweaver et al. 2004). Although many hypotheses have been suggested for this curious absence, a large subset of these hypotheses require some form of fine-tuning (Ćirković 2009). It's as if the universe seeks to promote maximum evolutionary diversity in each civilization, while developmentally guiding each of them to a future in which their evolutionary learning might be instantly shared (Smart 2012). In short, there's a lot of apparent tuning that needs explaining.
Of course, there is also a lot of waste and danger and randomness in our universe as well. The vast majority of physical systems in our universe are simple and dead, not complex or adaptive, and there are catastrophes and danger everywhere. Intelligent life may be so hard to produce that our universe may need to evolve an entire galaxy of stars to develop one intelligent planet, on average. Evolution on Earth has seemed equally wasteful and violent, if we focus on all the species that have disappeared, rather than the intelligence, morality, and immunity that has survived and grown. In the far future, our Milky Way galaxy and Andromeda are destined to crash into each other, obliterating their beautiful spiral structures. Observing all this apparent waste, danger, and chaos has led many astrophysicists, including Neal DeGrasse Tyson (2006), to argue that fine-tuning doesn't exist.
Besides the desire to avoid the idea of a purposeful universe, and its historically theistic implications, scientists commonly seem to reject the idea of fine-tuning via two ways: misperception and mischaracterization. Let us consider each of these latter issues now, and propose an alternative description, the partially-fine-tuned universe hypothesis, to try to reduce these problems. In my view, modeling fine-tuning in evo devo terms, in both physical and informational dimensions, and simulation testing it in both organisms and universes, is a core challenge science must address if it is to properly critique universal fine-tuning models.
The partially fine-tuned universe hypothesis: Intelligence is a weak selector, not a designer
Misperceptions can commonly cause us to reject fine-tuning, if we examine complex systems from inappropriate perspectives and scales. Consider a few examples:
- If it turns out to be true that it takes a galaxy of stellar "experiments" to produce just one (or a few) intelligent civilizations per galaxy, on average, that looks extremely wasteful and random (evolutionary) at the solar system scale, but simultaneously convergent and predictable (developmental) when we view the same process at the universal scale. A system that reliably produces hundreds of billions of something may very well be fine-tuned for that end.
- Intelligence in living systems looks very fragile and endangered (evolutionary) at the species scale, but very robust and accelerative (developmental) when viewed from the ecosystem or planetary scale (Heylighen 2008). A type of developmental immunity appears to exist. For example, we can presume that very little of the "conserved core" of developmental genetic intelligence (Gerhart and Kirschner 2005) in Earth's species pool was eliminated by any of the major past catastrophes and extinctions that Earth's ecosystem has experienced. Instead, those catastrophes appear to have pruned back the evolutionary variety, created new exploration space, and catalyzed powerful new punctuations of evolutionary innovation (new phenotypic or sociotechnological morphology and function), while increasing immunity to further disruptions of the same type, shortly after each major catastrophe (K-T event, Permian extinction, Ice Ages, Toba event, many others). I call this process developmental immunity, in general terms, or the catalytic catastrophe hypothesis, in relation to specific catastrophic events (Smart 2008, 2018). This hypothesis has been explored in biological systems by Gerhart and Kirschner 2005, Bhullar 2017 and many others. In social systems, economist Nick Taleb calls it antifragility (Taleb 2012). A general kind of immune learning appears to have operated throughout life's long history on our planet, as a central stabilizer of accelerating change.
- We can see chaos and randomness in galaxies colliding billions of years from now, as Tyson (2006) emphasizes. But if both low-intelligence universes (via cosmological natural selection) and higher-intelligence civilizations (via the transcension hypothesis) use black holes (either to "randomly" produce new universes in CNS, or to do intelligence-guided replication in the transcension hypothesis), then future galaxy collisions long after many of the universe's black holes are created looks like normal aging and recycling of an evo devo system after it has aged past replicative maturity. All complex living systems are developmentally fated to senesce and be recycled. What looks fine-tuned, from that life cycle perspective, is that galaxies are stable for the billions of years necessary to produce complex life, and that mechanisms for universal replication and civilization communication (e.g., black holes, in the CNS and transcension hypotheses) have self-organized to be fecund in our universe.
- We can focus on how easy it is for planets to be outside a galactic or stellar habitable zone, and (in our solar system) become greenhouse hells like Venus, or lose their plate-tectonics and atmospheres and dry up like Mars, (taking an evolutionary perspective on planetary science) or we can consider the marvel of the apparently robust (developmental) existence of both habitable zones and Earthlike planets in our galaxy, and the unique features of Earthlikes as a cradle for life (see Barrow 2008). Water-bearing Earthlike planets and yellow-white suns may be universally unique developmental portals (accelerative gateways) for life. Yellow-white stars like our Sun have their peak irradiance in the visible light range, optimal for water-based, photosynthetic life (figure right). Hotter (blue) stars radiate much more in the dangerous, high energy range, and colder (red stars) radiate more in the infrared range, and with far lower specificity (their peak irradiance curve is much flatter). Our Sun's particular spectral type, and our Earth's plentiful water vapor, water, and strong magnetic field efficiently shield life from radiation arriving outside our Sun's most useful range, a "lock and key" fit that makes most sense as a coadaptation, arising in past cycles of evolutionary development. Our plate tectonics, oceans and clouds, carbon and nitrogen cycles, and ecosystem itself stabilize many other features of our nurturing environment (Volk 2003). Why is our Sun-Earth energy transfer and geophysics so apparently co-adapted for the generation and buffering of life processes? Either of two kinds of observation selection would seem to be involved. Either the law of large numbers explains these local conditions (an evolutionary observer selection explanation), or some of the physical parameters in our universe have become biased toward the production and protection of life (an evolutionary developmental observer selection explanation). We must predict that the vast majority of planets would still be expected to be barren in an evo devo universe, as biological evolution always requires massive and "wasteful" stochastic variety to find new developmental optima. But we should also expect some improbable fine tuning, in a small subset of parameters (the “5%”), for the robust emergence of a special class of life-supporting planets. Without taking this evo devo perspective, we might predict that self-aware life would typically emerge, on average, in both a much less efficient and less safe ecogeophysical environment.
Mischaracterizations, arising from incorrect models, can also cause us to reject fine-tuning. Perhaps the most common mischaracterization comes from focusing only on the evolutionary processes of adaptation, to the exclusion of the developmental processes. That can happen when we view a system from only one scale or perspective, as we have just described. But there is another mischaracterization that comes with the assumption, surprisingly common among fine-tuning critics, that fine-tuning must be extensive if intelligence is involved in universe replication. But as we'll argue now, if replicating universes are anything like replicating organisms, then extensive fine-tuning by any finite intelligence, whether internal or external to our universe, seems an unsupportable and non-naturalistic assumption.
If we live in an evo devo universe, it can only be a partially fine-tuned universe, as evolutionary parameters may always be far more plentiful than developmental ones (per the 95/5 rule), and as any real physical intelligence must always be a computationally incomplete aid to selection in evo devo systems, never an omniscient or omnipotent designer.
What I call CNS with Intelligence (CNSI) (Smart 2008, 2011), or what my EDU colleague Clement Vidal calls Cosmological Artificial Selection (CAS) (Vidal 2008, 2010) is the hypothesis that intelligence, and its ability to simulate more and less adaptive futures, must play some useful, nonrandom role in the replication of universes. These hypotheses do not argue that intelligence can rationally design future universes, but rather that universes that self-organize intelligence are somehow more adaptive, in a nonrandom fashion, than universes that don't. In other words, some kind of multiversal selection occurs, in which intelligence, at both the very fundamental "genetic" level of self-replicating universal parameters, and at its higher levels, which includes conscious beings able to develop science and engineering, plays a nonrandomly beneficial role. That's the core hypothesis.
Even without the math it is easy to induce, in any biological replicator, that there is likely to be a nonrandom adaptive value to the emergence of general intelligence, of immunity (defensive intelligence), and of various forms of interdependence (collective intelligence, social morality, positive sum games), the latter starting with kin. We can imagine many circumstances when each of these computational systems that encode models of self-, others, and environment have adaptive value. These intelligence systems (general intelligence, immunity, and interdependence) may be present, in some fashion, in all complex adaptive systems. If we also consider evolutionary innovation and developmental sustainability as forms of intelligence, including the mix of stochastic and predictable genetic processes that generate our minds, we can imagine at least five potentially universal processes of intelligence, as we will discuss later (Smart 2017b). We can also identify multiple forms of intelligence (genetic, cellular, collective, neurological, societal, technological, etc.) on Earth, in living systems and their creations. Why should intelligence not also be a central property of the universe as a complex system, if it in fact is a self-replicator, existing in some larger environment (the multiverse)?
Yet we must also recognize that all simulations that any intelligence can do, either within our outside our universe, must be sharply finite and constrained, rapidly unable to predict most of the multivariate nonlinear dynamics and informatics of any complex adaptive system, the farther we extrapolate it to the future, or the more we include its evolutionary (vs. developmental) mechanisms. All such systems quickly become combinatorially explosive in their potential complexity, and all real intelligences are limited in their physical and computational complexity. With respect to the special case of logical-mathematical provability, this concept is as old as Gödel’s Incompleteness Theorem (1931), and incompleteness seems intrinsic to the nature of informational complexity itself (Chaitin 1992; Calude and Jürgensen 2005).
The informational incompleteness of all intelligence, along with the inability to have perfect knowledge (simulation capacity) of initial conditions and all the relevant laws, may also explain why all complex actors with mind have "free will" (unpredictability to self), even under the most informationally ideal conditions. Philosophers since Lucas (1961) have tried to relate free will to Gödel’s theorem, but such a relation starts by preassuming the physical universe conforms to Gödel’s conditions for mathematics (it may not). At present, the full informational and physical nature of (our self-experienced) free will, and of conscious decisionmaking, remain a mystery to be solved.
Nevertheless, what we know and can guess so far about intelligence argues that any "design" that real intelligences can do, of their future selves and environments, will be highly limited. It does not seem defensible to imagine that end-of-universe or extrauniversal intelligences might be omniscient or Godlike, if they are physically real and we live in an evo devo universe, and they also seem unlikely to have the capacity to create "anything" out of "nothing." It is illuminating that “Why there is anything rather than nothing?” or alternatively, “Why did anything begin?”, is sometimes called “Question Zero” in physics. It may always remain a metaphysical question to real intelligences. Perhaps one of the most useful clues to its metaphysical nature is that the concept of nothing itself, just like the concept of infinity, while very useful in our mathematics, may be only an informational, not a physical concept (Aguirre 2016). Real intelligences may be forever stuck within the (evo devo) system (supporting universal environment) that they find themselves emerging within, a system that has its own physical and informational laws and constraints, only some of which are likely to be modifiable.
Discovering our universe's parameters and laws, and learning how to manipulate them to improve adaptation under selection, but always in finite and limited ways, seems to me to be the central benefit of intelligence in living systems. All living systems, while they possess some level of intelligence, still have many vestigial systems and errors and maladaptations in them which are beyond their control or even understanding. This should be true for our universe as well, if it is an evo devo system.
My current intuition for what an end-of-universe intelligence might be able to do with respect to "design" of future universes, would be to alter some of the coupling constants which influence the developmental characteristics of the next universe, presumably to raise the probability that it will be complexity, life and intelligence friendly. They might do this, for example, if intelligences in this universe can use their intelligence, and the laws of physics, to produce black holes which create other universes, and if the coupling constants can pass through the singularity of a black hole into another universe, as some physicists have postulated (Smolin 1992; Crane 1994). But notice that this kind of "universe engineering", though it is simulation-guided, is not the rationally-engineered universes idea of James Gardner (2003, 2007). Gardner assumes that end of universe intelligences could change any of the constants, and might have extensive foreknowledge and control of the consequences of those changes. I would equate this view with intelligent design, which we shall discuss later. It sounds like non-naturalistic theology, not science.
Instead, tinkering with the values of our universe's coupling constants, in a way that might produce even more life- and intelligence-friendly universes, seems likely to be analogous to what human genetic "engineers" do today when we alter genes in "designer" organisms. What we are actually doing is making intelligence-guided engineering guesses at what will be more adaptive, and some of the most critical conserved genes are beyond our ability to tweak, without killing the organism. Our foreknowledge of these complex evolutionary systems must always be limited the further ahead we look. An honest assessment would be that we are not really "engineering" or "designing" new organisms, but are instead making our best experimental guesses, based on our finite simulation capacity and knowledge, working within the evo devo framework we have inherited, at what might be more adaptive.
Any hypothetical universe "design" would have to work the same way, in an evo devo universe. It would be a process of partly intelligence-guided selection, and partly unknowable experiment. It is not accurate to call such an undertaking by the word design. When we are talking about bottles and bridges, and other nonautonomous systems, it makes sense to use the word design. But the more complex adaptive and internally intelligent the system gets, the more the unpredictable evolutionary aspects of the system overwhelm the predictable developmental parts. Once we get to the "design" of things like living organisms, or new deep learning computers, or future universes, it makes more sense to call this process selection than it does design.
Gardening future universes using our own best science and intelligence would be directly analogous to the artificial selection we humans do on our domestic plants and animals, a process Darwin discussed at length in Origin of Species (1859). This is why Clement Vidal prefers the term Cosmological Artificial Selection to describe what this process might look like to any future intelligences that become competent enough to "engineer" intelligence-influenced black holes (if those are the seeds of new universes, per CNS), or to otherwise aid in the production of future universes.
In sum, there are at least five important levels of evo devo-related partial fine-tuning models that should be critiqued in future fine-tuning debates:
- Level I. Our universe appears fine-tuned (self-organized) for the emergence of complex, long-lived universes and black holes (Smolin 1997; Rees 1999; Gardner and Conlon 2013).
- Level II. Our universe appears fine-tuned for the fecund emergence of G-, K-, and M-class stars and biological life (Henderson 1913; Barrow et al. 2008; Lewis and Barnes 2016).
- Level III. Our universe may be fine-tuned for the fecund and accelerating emergence of intelligent life (Piel 1972; Sagan 1977; Moravec 1979; Dick 1996; Kurzweil 1999, 2005).
- Level IV. Our universe may be fine-tuned for the fecund emergence of intelligence life, which can then produce new universes (Crane 1994; Harrison 1995).
- Level V. Our universe may be fine-tuned for the fecund and accelerating emergence of increasingly innovative, intelligent, immune, interdependent (defending evo devo values) and sustainable forms of complex life (Smart 2008, 2012, 2017b).
If this analogy between replicating organisms and universes holds up, models like Smolin’s CNS, in some variation that also includes intelligence (CNSI), will continue to gain theoretical and empirical support. The better we understand and can simulate the operation of evolutionary and developmental parameters in living systems, the better we should be able to understand and simulate them in universes as well. Both look like finite and replicating systems, in an evo devo model.
The riddle of convergent evolution: Limited forms most beautiful
Convergent evolution is evidence or argument for physical attractors in the phase space of dynamical possibility which guide and constrain contingently adaptive evolutionary processes into statistically predictable future-specific structure or function, in a variety of physical and informational environments. When we look at evolutionary history, species morphology or function is often seen to converge to particular "archetypal forms and functions" in a variety of environments.
Such attractors have been called deep structure, guiding evolutionary process in predictable ways, regardless of local environmental differences. We have described how deep homology, and accretive developmental regulation, must be central to understanding physical and informational evolutionary convergence. Deep homology and deep structure in turn must depend on specific initial conditions (developmental genes in the “seed”), the emergence of hierarchies of modular structure and function in the unfolding organism, and persistent constancies (physical and chemical laws, stable biomes) in the environment. Likewise, some examples of convergent evolution may be best characterized as ecological, biogeographical, stellar-planetary, or universal evolutionary development (ED) if their emergence can be modeled, after adjusting for observer selection bias, to depend on specific universal initial conditions, emergent hierarchies, and environmental constancies.
A famous example of convergence is found in eyes, which appear to have evolved on Earth from different genetic lineages to work similarly (function as sensors for nervous systems) in all species possessing sight, and in the case of camera eyes, to also look very similar (form) in both vertebrate and invertebrate species, like humans and octopi (humans have a blind spot, however, as our eyes evolved via a different evolutionary developmental history than invertebrates, see picture left). One can easily advance the argument that, in universes of our type, eyes, though first created by a process of evolutionary contingency, become a developmental archetype, an adaptive optimization for the great majority of multicellular species in Earth-like environments.
Presumably, the previously rapidly-changing "evolutionary" gene groups that led to eye creation become part of an increasingly conserved "developmental" genetic toolkit for all eye-possessing species in environments where eyes are adaptive. Eventually, due to both path dependency and emergent hierarchies, some subset of these gene groups should be incapable of being changed without preventing development itself. Proving such genetic convergence arguments with evidence and theory is of course more difficult, yet it is a fertile area of investigation today.
Charles Darwin ended his foundational text on evolution, On the Origin of Species (Darwin 1859) with a well-known phrase, predicting "endless forms most beautiful" continuing to evolve. But as George McGhee describes in a well-titled book, Convergent Evolution: Limited Forms Most Beautiful (McGhee 2011), preexisting physical and informational optima in our particular universe mandate that only a limited subset of forms and functions will ever emerge in biological evolution. Evolutionary development always grows morphological and functional diversity, and especially rapidly under stress, but developmental control and optimization makes it a net subtractive and constraining process, relative to its theoretical potential. Creative evolutionary process is continually reconverging to developmental optima, driven there by functional (environmental) and developmental (genetic) constraints. Better understanding and modeling convergence is one of the great and underappreciated challenges of modern evolutionary biology.
Less-optimizing convergence (LOC) versus optimizing convergence (OC)
In our mostly chaotic, contingent, and deeply nonlinear universe, we can predict that the vast majority of examples of convergent evolution will not be driven by the evolving system's discovery of some hidden general optimization function in parameter space, like the eye archetype, but rather, the discovery of many less-valuable and less-permanent optima that do not lead to higher complexity, yet may still be required for the universe’s evolutionary development. To understand convergence, we will need some kind of evo devo-guided general optimization theory. Let's consider two necessary features of that theory now.
- We can predict that any optimization that occurs must be on a continuum, from highly-optimizing convergence, which we will refer to simply as optimizing convergence (OC), conferring advantage in all the most competitive and complex environments, to a wide variety of other cases, which we can refer to collectively as less-optimizing convergence (LOC). LOC cases would include convergence that offers only some temporary or local adaptive advantage, to just a few specific species, or in some subset of specialized or less-complex environments, convergence that offers no advantage, or convergence that is deleterious but not fatal. Names for a few general classes of LOC cases have been offered by scholars, including passive convergence, parallel evolution, etc.
- Optimizing convergence can occur via both physical and informational processes. Physically, we might see greater efficiency of employment of physical resources, as in Bejan's constructal law, or greater density of employment of physical resources for offense or defense, the escalation hypothesis (Vermeij 1987). Informationally, we might see efficiency or density gains via informational substitution for physical processes, what Fuller called ephemeralization, or greater general intelligence (modeling ability), greater immunity, or a more useful collective morality, offering more general and persistent adaptation to a wider range of environments than previous strategies. Intelligence also offers the ability to modify environments to suit the organism, what biologists call niche-construction or stigmergy, as humans, social insects, and many other species do either in limited forms or extensively today. To understand OC, we will need a theory of optimization that tells us when a physical or informational advantage is likely to be more generally adaptive, particularly in the most complex, competitive and rapidly-changing environments. We also need to know whether there are any other paths that can lead, in a competitive timeframe, toward a competitively superior new form of adaptiveness. If not, then we may have discovered a developmental portal, a global optimum that represents a bottleneck, a singular pathway toward greater adaptation at the leading edge of local complexity. Organic chemistry, RNA, photosynthesis, and oxidative phosphorylation are all potential examples of portals that all universal life must pass through first, on the way toward greater adaptive complexity. They may be the only global optima on their landscapes, at the relevant timeframes, that will allow the creation of vastly greater adaptive complexity.
Another complication of optimizing convergence at the leading edge of complexification is that over time, it must occur within an increasingly limited set of evolutionary possibilities, as increasing developmental genes and processes at the leading edge will progressively limit the evolutionary possibility space within any particular inheritance system. Processes like heterochrony, neoteny, and gene duplication (Wagner 2003) can temporarily reverse generally growing genetic constraint, but only the invention of a new class of inheritance system, in a metasystem inheritance transition (e.g., self-replicating genes in organisms inventing self-replicating ideas in brains, inventing self-replicating algorithms in technology) seem able to lead to large new regimes of evolutionary innovation (Turchin 1977).
If we live in a noetic (information accumulating and intelligence-centric) universe, nervous systems would surely qualify as OC. Based on neurotransmitter and genomic differences, Flores-Martinez (2017) argues that nervous systems were convergently invented three different times, by comb jellies, jellyfish, and bilaterians. But only in a small subset of prosocial, tool-using, land-based vertebrate bilaterians do we see a strong trend toward runaway brain size. OC is clearly multifactorial for developmental transitions to more rapid, more stable, and more complex evolutionary regimes (e.g., cultural evolution).
Consider eyes again. As with nervous systems, which are particularly helpful in complex environments, we can make a plausible case that eyes, at one point in time, became a necessary functional adaptation in the most complex environments. Andrew Parker's light switch theory (In the Blink of an Eye, 2003) proposes that the development of vision in Precambrian animals directly caused the Cambrian explosion. Critics have observed problems with the timing, and that complex eyes may instead be a consequence of rapid body plan complexification, rather than a generator of new selection pressure for complexification. Either way, this is a fascinating theory, as it implies a necessary coevolution of intelligence and morphological and functional complexity. Once they emerged, it is easy to argue eyes were an evolutionary ratchet, and that all visible animals in the most complex environments would soon need them, or a handful of other uniquely effective defensive strategies, to survive.
Many other examples of OC can be proposed, in the most physically and informationally complex, and rapidly changing, environments on Earth, including the necessary emergence of eukaryotes, oxidative phosphorylation, multicellularity, bioluminescence, nervous systems, bilateral symmetry, jointed limbs, opposable thumbs, tool and language use on land (much faster-improving than aqueous environments), culture, and technology, including machine intelligence.
To make a few intelligence-related predictions in OC, I suspect that grasping limbs and tree niches on land are an early developmental portal (optimized convergence and phase transition in collective intelligence) leading to complex tool use on Earthlike planets, as tree swinging and grasping limbs offer an ideal training ground for complex, predictive brains, and as tool use and construction in air offer far greater mechanical advantage than in water. I am also a fan of Dale Russell's Dinosauroid hypothesis (Russell and Séguin 1982), which argues that the bilaterian tetrapodal humanoid form, which includes two locomotion and two prehensile (grasping) appendages, may be an optimizing convergence (minimum viable solution set) for becoming the most generally intelligent (and largest brain to body weight) land-dwelling bilaterian. I have also predicted that competitive-cooperative tool use on land, in the manner employed by early humans with Oldowan axes, is likely to be a universal developmental portal to runaway collective intelligence in bilaterians, as that environment seems to offer such strong selection pressures for generally adaptive defensive and offensive capacities, by contrast to animals that cannot collectively employ such "game-changing" early offensive and defensive tools as stone axes, clubs, and fire (Smart 2015).
Future science will need better theories of complexity, complexification, and optimization, to deeply understand such candidates for evolutionary convergence, and to distinguish the much greater variety of examples of less-optimized convergence from the most highly optimized forms.
Optimizing convergence as accelerating and stabilizing evolutionary development, on many scales
When convergence is viewed from the perspective not of the evolving species, but from some larger system scale (the biogeography, the planet, the universe) we can view optimizing convergent evolution as a process of not simply evolution, but of evolutionary development (ED), an ED that continually accelerates and stabilizes its complexity in special domains of space and time.
When we claim a convergence process is an example of ED, we are not only claiming that some kind of general optimization is occurring. We are also claiming that some kind of evolutionary developmental process, with both "random" and creative evolutionary search, and predictable convergence, directionality, hierarchy, modularity, life cycle, and perhaps other features found in biological development, is being followed, at some larger systems level. Consider embryogenesis. Viewed from the perspective of the individual actors (molecules), we see mostly stochastic, divergent and contingent processes. As we zoom outward to larger and longer spatitemporal scales, we can also see a few convergent, hierarchical, and life cycle processes. To view optimizing convergence as not simply evolution, but as evolutionary development, we often must take these wider scale views, as in the following examples:
- Galactic and Universal change offer many potential examples of not only evolutionary but apparent developmental change, as we have discussed. Curiously, the evolutionary development of complexity seems strongly accelerative, with increasingly rapid complexity transitions in increasingly local spatial domains (Smart 2008). As a high school student contemplating this trend in 1972, I recognized the logical limit of that process was the black hole. The first of these puzzling objects, Cygnus X-1, had been discovered just the year previously, in 1971.
- Stellar-Planetary-Astrobiological change offers more examples. When we look down from early universal change to the stelliferous era, and the genesis of our life-permissive planet and its star, astrophysical theory tells us that the way stars have replicated, and chemically complexified, through three different populations over billions of years, has been not only evolutionary (a variety of randomly arrived at star and planet types and distributions) but evolutionary developmental, involving a progressive drive to complexification in a predictable subset of types. Many astrobiologists and planetologists argue that a subset of chaotic and nonlinear (“evolutionary”) stellar-planetary change has reliably led, with high probability and massive parallelism, to G-(and perhaps some K- and M-) class stars and Earth-like planets ideal for the development of archaeal (geothermal vent) life, and from there, to prokaryotes and eukaryotes. See Nick Lane's The Vital Question (2016) and Smith and Morowitz's The Origin and Nature of Life on Earth (2016) for two such stories.
- Biogeography and Ecology offer more examples of not only evolutionary but apparent developmental change. In biogeography, we find scaling laws, like Copes rule, and biogeographic laws like Foster’s rule and Bergmann’s rule, with their predictable processes of optimizing convergent evolution, or evolutionary development. The famous convergence of form seen in placental and marsupial mammals, on separate continents, offers another example of not just evolution, but biogeographic ED. For many more examples, including intelligence traits, see Conway-Morris (2004, 2015), McGhee (2011), Losos (2017) and our list of examples of convergent evolution (Wikipedia 2012, and Smart and Chattergee 2012-2017) in species morphology and function. In ecologies, we can identify predictable patterns of ecological change, including ecological succession, ascendancy, and panarchy.
- Culture, Science, and Technology change offers yet more examples. When we look above individual cultures and do cross-cultural comparisons, we find many examples of developmental features at the leading edge of competitiveness, including parallel invention and/or convergent development of archetypal scientific and technological inventions like fire, language, stone tools, clubs, sticks, levers, written language, mathematics, hydraulic empires for our first great cities, wheels, electricity, computers, artificial neural networks, etc. In each of these cases, a high-order convergence has occurred. These and other specific examples of cultural change look not only evolutionary, but evolutionary developmental (ED). Once these archetypes and algorithms exist, there's no going back, for any culture seeking to stay on the leading edge of physical and informational complexification, and general adaptiveness. We also find many examples of developmental constraint laws that operate in social and economic systems, like scale laws (West 2017; Bejan and Zane 2013) and more generally, the least action principle (Georgiev et al. 2015). We also see emergent constraint in Karl Friston's concepts of Markov blankets, a method that creates cognitive boundaries in machine learning, allowing emergent logic and hierarchies, and in his model of active inference, a form of Bayesian learning that seeks to minimize the discrepancy between predictions and sensory input, an intelligence-centric formulation of the free energy principle (Friston 2010). In active inference, our brains, bodies, and niches help us to get better either at predicting in general, irrespective of our desires, or at actualizing the particular and unique predictive visions we imagine. The first of these mental processes can be called development (from the neuron's perspective), and the second, evolution, making active inference a model of evo devo foresight. How these two potentially conflicting neural goals are mediated is one of many questions for future research.
In each of these rough hierarchies of complexity, our universe is not only generating local variation, creativity, and difference, it is also developing toward a small set (in our present understanding) of currently-predictable destinations. While there is much about cosmogony that remains unclear, we know that dark energy is accelerating complex galactic groups away from each other, that total entropy increases, and that an increasing fraction of the mass-energy of our universe will end up in black holes. The better we understand the conservative and predictable features of our universe, and can distinguish them from creative and unpredictable ones, the better we may understand evolutionary and developmental processes at all scales.
There are two more curious features of this predictable developmental trajectory, across all of these environments, that should now be mentioned.
- The first is the ever-faster complexification rates seen in the historical record of the most physically and informationally complex locations in our universe, since the emergence of G-class stars, Earth-like planets, and almost simultaneously, on our planet, life. This acceleration was famously summarized in Carl Sagan's metaphor of the Cosmic Calendar. Ever since August, on this calendar metaphor, leading-edge complexity environments have become exponentially faster, more complex, and more intelligent, on average, on Earth. Sagan said this phenomenon, which we can call acceleration studies, was an understudied area of science, in need of better understanding. See Sagan's The Dragons of Eden (1977) for his original account, and Heylighen 2008 for more recent work. It is my hope that better models of early universe, astrophysical, chemical, biological, psychological, social, economic, technological, and other evolutionary development will help us understand our universe's emergence record of ever faster and more physically- and informationally-complex local environments.
- The second is the increasingly informationally stable (developmentally immune, antifragile) nature of complexity in our most complex environments. In prehistory, species could easily be destroyed by environmental change. But once we began recording and simulating our world in nonbiological substrates, human-technological culture has gotten better every year at recording, simulating, and recreating both biological and cultural information (Malone 2012). As a result, such information has become far more resilient to catastrophe (Smart 2008; Taleb 2012; Dartnell 2014). There is something about mind, culture, science, and technology that makes the information it produces more stable to destruction via environmental fluctuations. Perhaps a growing intelligence typically provides increasingly useful sets of adaptive strategies for survival. Some kind of nonlinear input-to-state stability (a form of Lyapunov stability) may emerge as intelligence's potential to moderate environmental inputs grows. Perhaps the most intelligent collectives develop not only greater immunity but greater morality (both have been proposed as subtypes of intelligence). This latter view is controversial, given recent human history with advanced technology, but there are good early arguments for it as well (Pinker 2010). Perhaps it is simply that increasing intelligence allows progressively more durable (informationally immune) forms of cultural memory to be developed (Malone 2012). The best descriptor of local intelligence’s ever-growing immunity may be niche construction (environmental engineering), of which memory is just one form. Niche construction has afforded humanity the ability to move our core complexity to increasingly time-stable architectural environments (books, villages, cities, computers), but these are nothing compared to what may soon come. Several scholars have argued that humanity appears just a few decades away from being able to port its essential complexity, in body and mind, into a technological substrate (substrate shift). Such postbiological entities seem likely to be vastly more stable to destruction via any imaginable universal process, and far more redundant, than today's bio-dependent culture, science, and technology. Such entities should be able to harness (and do intelligence-guided experiments with) molecular nanotech, fusion energy, and perhaps even subatomic processes (femtotech), and should no longer require either planets or functioning stars to maintain their existence (Forward 1980; Smart 2008; Davies 2010; Rees 2015). Due to accelerating change, such stable new entities also seem likely to arrive much earlier, in cosmic time, than most of us would presently expect or predict.
Both of these features, our potentially developmentally-guided acceleration and our progressive forms of informational stability, suggest that today's current models of existential risks are likely to be overstating the near term risk to our species of many apparent species-threatening events (for a detailed review of such risks, see Bostrom and Ćirković 2008). The time for which such risks actually threaten our informational complexity seems to be rapidly decaying. We appear to be on the edge of entering a far more stable substrate for life and intelligence, in a cosmologically insignificant fraction of future time.
“Tape of life” (“identical Earths”) experiments: Simulating ecogeophysical ED
If life emerges on two similar Earth-like planets, in either in reality or in a sufficiently accurate simulation test, then by definition we can predict that the evolutionary aspects will almost always turn out differently in the two environments, and the developmental aspects will turn out the same. This is called the “Tape of Life” experiment, and it is commonly discussed in the philosophy of biology and by some of the more systems-oriented evolutionary (developmental) biologists.
Beginning in the 1970s, leading evolutionary theorist Stephen Jay Gould (1977, 2002) famously predicted that little of life’s functions and morphologies on another similar Earth would turn out the same as those presently found on our Earth. He expected a few broad similarities, in kingdoms and some phyla, but most species would turn out very differently, in his view. Beginning in the 1990's, Simon Conway-Morris (1998, 2004, 2015) famously argued the opposite, that most functions and many morphologies would turn out the same, optimized for replication and adaptation in this particular Earth environment. We may aptly call this an evolutionary developmental perspective on Earth’s history (picture left). In the decades since, some biologists and most astrobiologists have migrated from Gould's to Conway-Morris’s camp, though the dividing line between predictable and unpredictable processes of change remains a productive and contentious debate.
In recent years, there has been a surge of studies of evolutionary convergence, motivated by such wide-ranging questions as the structure of the protein space to experimental evolution to evolutionary genetics to ergodicity in biophysics to the attempted “neo-Gouldian” developmental account of homology versus homoplasy (Dryden, Thomson, and White 2008; Turner 2011; Lobkovsky and Koonin 2012; Pearce 2012; Powell 2012; Orgogozo 2015; Powell and Mariscal 2015; McLeish 2015; O’Malley and Powell 2016; Louis 2016). Roughly speaking, most of these new results are strongly supportive of convergence – in more or less radical form – as the key feature of macroevolution. For instance, Dryden et al. (2008) and McLeish (2015) argue that the accessible part of the genomic space is much smaller than conventional combinatorial wisdom suggests, and that evolution may have actually explored most of it by now. This is a powerful idea. Consider that a fully explored (statistically repetitive and no longer creative) phase space may be a necessary but not sufficient condition of all developmental portals (complexity transitions), to make such transitions appropriate guides (checks, funnels, gateways) to evolutionary exploration.
Convergent evolution, at all universal scales, can be productively modeled as a pull of attractors, and if those attractors are subject to replication and selection, as a process of evolutionary development. Such modeling should work, to varying degrees, whether we are describing physical, chemical, genetic, organismic, species, ecosystem, organizational, cultural, or technological evolutionary development. Perhaps the simplest phrase to encompass all these and other types is “universal evolutionary development”. Applied to the universe, evo devo theory argues that both universal evolution (useful diversity) and universal development (useful similarity) must be aspects of any universal biology that some scientists and systems theorists (Mariscal 2016) are seeking. Though we seek simplicity in our models, discussing either unpredictable or predictable processes alone will lead to insufficient views of how adaptation actually occurs. We must learn how they blend, and when those blends are adaptive.
We must also recognize that just as in biological evo devo, our science and simulation skills will be insufficiently advanced to predict many of the developmental similarities (“convergent evolutionary developments”) that emerge between two parametrically identical universes, two Earth-like planets, two similar but biogeographically separated continents, two highly-similar cities or organizations, two genetically identical twins, or even two dividing cells.
Fortunately, the latter examples, and others, have happened many times on Earth. So we can look to these “natural experiments” to better understand processes of development, at all scales. As our science and simulation capacity gets better, we can also develop better and more predictive models of how our physical universe evolved and developed.
In a few of our more advanced biotechnological prosthetics (e.g., cochlear and vision implants, even hippocampal "chips"), our software and hardware models are good enough to substitute for the biological system without significant loss of function. We can hope that this intelligence substitution will also serve us as we learn to simulate universes in our future computers as well.
If so, we will increasingly be able to predict and validate ED hypotheses in at least two major ways. By discovering more natural experiments, at all scales, and by simulating the emergence of those experiments, at a level sufficient for the simulation to substitute for the physical process.
“Tape of the cosmos” (“identical Universes”) experiments: Simulating Universal ED
Let’s look now at convergent evolution on the largest scale we can presently imagine: our universe. In Carl Sagan’s famous Cosmic Calendar metaphor of change (1977, 1980; picture right), we see that earlier stages of hierarchical evolutionary development, involving the emergence of large scale structure, galaxies, and stelliferous and planetary change, are highly isomorphic and convergent, across the universe. Simply looking at the night sky shows us these amazing levels of convergence. In the last century, physicists have worked out many of the reasons this convergence is evolutionary developmental. It is written into the initial conditions and emergent laws of our particular universe.
Are the observable morphological, functional, and informational features of our universe that have clearly accelerated on Earth since the emergence of life, as depicted from August afterward in the Cosmic Calendar metaphor, also found convergently throughout the universe? Is this convergence on multilocal complexity acceleration in our universe strong, happening with high frequency, as a developmental process, or is it random and happening weakly, as an evolutionary process? In other words, should we expect Earthlike acceleration in a multitude of special environments, such as those found on habitable planets around G-class stars? These are questions of universal ED. Astrophysicists and astrobiologists hope to answer such questions, by theory and observation, in coming years.
Today we can conduct primitive simulation tests ("simulation experiments") to explore the divergences and convergences we see in two model universes, but our science remains incomplete, and our cosmological simulations, both in their physical and informational dimensions, still do not capture all the reality they attempt to model. Fortunately, our experiments in simulating evo-devo phylogenetics in biology (picture left) may lead the way to better simulations of all kinds of evo devo systems. If we live in an evo devo universe, our simulation experiments will get ever more predictive in their developmental components, and they’ll eventually convince even the most die-hard believers in contingency that we have a set of highly constrained futures ahead of us.
Consider genetically identical twins. Most molecular and tissue-level aspects of two genetically identical twins look different when you view them up close (different fingerprints, organ microstructure, ideas, etc.). Those are "evolutionary" differences in an evo devo model. They are locally unique in myriad ways, either because the twins genetic systems aren’t capable of ensuring perfect identicalness, or because there are adaptive (e.g., immunity) advantages to this local diversity. Genes are not a blueprint, but a recipe for building local complexity in a way that allows contingent local diversity, yet is also robust enough to local molecular chaos that each twin is reliably guided toward a set of useful far-future destinations in structure and function. All the aspects of the two genetically identical twins that turn out the same, we call “developmental.”
Now consider that if our universe replicates, and its emergent features and intelligence undergo some form of self-selection or selection in the multiverse, this twin model helps us to define evo devo terms like universal evolution (variation between universes) and universal development (similarity between universes). Cosmology models typically assume that if our multiverse had two parametrically identical universes (universes with the same fundamental parameters and initial and boundary conditions), some aspects of those universes would turn out the same and some would turn out differently. Astrophysics guides our universe toward future-varying (evolutionary) and future-determined (developmental) form and function, at the same time. Both evo and devo processes, and a recognition of the adaptive value of both evolutionary variation and developmental conservation, would seem to be necessary to any accurate simulation.
Physical and informational adaptation: Autopoesis and intelligence
Autopoesis is a term introduced by Chilean biologists Humberto Maturana and Francisco Varela (1973/1980) to describe the chemistry of living cells. It became popular with a few systems theorists in the late 20th century to describe the capacity of some complex systems to self-reproduce and self-maintain. Autopoesis scholars seek to find general systems rules applicable to any stably self-reproducing complex systems, including not only living systems, but stars, the chemical origin of life, and ideas, behaviors, algorithms, organizational rulesets, and technologies in culture. Implicit to autopoetic models is the idea that a better information theory, including a theory of cumulative embodied and adaptive cognition (intelligence) in the replicator, its inheritance system, and its environment, will be necessary to understand dynamical change in complex systems. See Varela et al. 1974; Maturana and Varela 1987; Mingers 1995; Luhmann 2003/2013; Luisi 2003; Bourgine and Stewart 2004 for some autopoetic models. Mingers (1995) offers a particularly good introduction to rules, drivers, and research questions regarding autopoetic chemical, biological, social, and technological systems, though even this excellent work does not consider the universe as an autopoetic system.
While they have made little scientific progress to date, autopoetic models are focused on what we might call the right questions: the physical and informational sources of adaptation in self-producing, autocatalytic systems, and the ways adaptation changes over time in environments which are, in the most likely presumption, replicating, autocatalytic complex systems as well. At the least complex end of the spectrum, all stable replicating systems depend on the emergence of some set of predictable action-reaction couplings to their environment. Stars are autopoetic systems, dependent on physical action-reaction processes. Moving up the chain of information-production rate density (a form of complexity), a variety of self-reproducing prebiotic systems (clays, RNA, protein polymers) are dependent on not only physical but also chemical action-reaction processes afforded by the complexifying Earth environment. On the path to life, certain self-replicating chemical systems developed autocatalytic protometabolic networks (Kauffman 1993), and some developed sensory-motor cognitive Bayesian (predictive) chemical networks, including memory networks, observable in single-celled organisms like Paramecium, Amoeba or Stentor (Bray 2011). At some point, gene-protein regulatory networks also emerged, and lipid cellularization. A subset of cells developed multicellularity, another subset developed specialized neural networks, and a subset of those, self-awareness. Understanding these increasingly complex set of adaptations, and the necessary emergence of life and mind as adaptations in certain environments, is one of the main challenges of modern science.
Scholars in such complementary fields as the origin of life (Smith and Morowitz 2016; Pross 2014), computational astrobiology (Pohorille 2012; Forgan et al. 2017), artificial life and information theory (Adami 2016), evolutionary escalation (Vermeij 1987), top-down causation (Walker et al. 2017) and evo-devo theory (see above) have all made progress in recent decades in understanding how successful cumulative replication changes the replicator, its seeds, and its local environment. Continued progress in such fields, especially in intelligence and information theory, will be critical to developing better models of adaptation.
Neo-Darwinian models of evolutionary adaptation, such as the adaptive landscape theory of Sewall Wright (1932) seek to model adaptation as phenotypic fitness to the environment, in some genotypic, morphological, or functional parameter space. Such models, while they have been tremendously useful, are also deeply incomplete, as they do not allow that the environment itself may change in predictable and highly-nonrandom ways over time, as the growth of intelligence influences local environments, as described in the phenomenon of stigmergy/niche construction (Odling-Smee et al. 2003). More precisely, the selective environment itself may be both evolving and developing over time, which in turn changes the nature of selection and adaptation.
If our universe itself is a replicator, as the evo devo universe hypothesis proposes, then it too is a selective environment that is not just evolving (experimenting, diversifying), it is also developing (complexifying, and engaged in a life cycle). From the perspective of biological evo-devo theory, much of this environmental complexification is both constraining, directional, progressive, and in-principle predictable, just as biological development is in-principle statistically predictable (though not always so in practice). If the universe is an evo devo replicator, at least some kinds of local environmental complexification will function to protect the replication and self-maintenance of the system (the universe). Any evolutionary (experimental, creative, contingent) activity that occurs within a developing organism must be increasingly constrained as that organism develops, in service to the organism's self-replication and self-maintenance. If our universe is an evolutionary developmental system, the local adaptive landscape must constantly be shifting toward certain developmental attractors, as evo devo complexity grows in certain local environments.
As with intrauniversal replicators, mechanisms that guide and protect universal replication may be very simple action-reaction and maximally energy-dissipative physical and informational processes, such as those that statistically guarantee stellar replication via star-formation feedback in the nebular hypothesis (Krumholz and McKee 2005), or the reductive tricarboxylic acid (rTCA) cycle, a proposed universal intermediary metabolism (Smith and Morowitz 2004). The rTCA cycle generates the five fundamental precursors to all biosynthesis (acetate, pyruvate, oxaloacetate, succinate, and α-ketoglutarate), and may be a maximal free energy dissipator in high energy flow environments, like geothermal vents. The rTCA cycle can be catalyzed by inorganic mineral cofactors. When run in reverse, the rTCA cycle is the oxidative Kreb's cycle, central to all life. After the rTCA phase transition occurred in a local environment, presumably further phase transitions allowed combination with another cycle, oxidative phosphorylation, and an energy harvester, photosynthesis, to store energy for the first cells. It is possible energy storage wasn't part of the first metabolism, as some photosynthetic bacteria use the rTCA cycle (Kreb's in reverse) to produce carbon compounds (Smith and Morowitz). Note that we've still left out DNA-guided protein synthesis, which is an information producing and environmental simulation system, and the transition that would merge it with metabolism, if we are going to describe the origin of life in evo devo terms. No wonder it is such a complex puzzle at present.
How advanced universal developmental processes may be, and how deeply they structure cascades of nonequilibrium phase transitions, may depend on the prior degree of universal replication, and the strength and nature of intra- and extrauniversal selection. At some point, the analogy with developmental genes in living organisms may apply, in which tuned parameters guide the emergence of planetary-scale social and technological processes that are functionally similar to biological intelligence, immunity, and morality. Such top-down causal informational mechanisms could be an integral part of our universe's self-maintaining processes. From a functional perspective, mind might inevitably emerge in a universal replicator, just as it has in biological replicators, if intrauniversal intelligence plays any usefully nonrandom role in universal replication and selection.
In summary, evo-devo biology may offer us the most complex and rigorous model for understanding not only convergent evolution in universal evolutionary development, but how adaptation itself must change in a universe that is itself a replicator. Once certain critical biological advantages, like eyes, emerge and are strongly adaptive in an environment, the majority of the most successful complex replicators in those environments may have to employ that advantage. Once certain critical technological advantages, like digital computers and machine intelligence, emerge and are adaptive in an environment, a subset of replicators (individuals, organizations, societies) must use those technologies if they wish to remain at the leading edge of adaptation. In this view only evo-devo biology, and its successive processes of molecular, genetic, physiological, and psychological evolutionary development, offer us a sufficiently complex analogy for understanding how adaptation may change in the universe, if it too is a self-reproducing, self-maintaining evo devo system.
Evo devo models require advances in a variety of theories, especially our theories of intelligence
If universal evolutionary development is occurring, future science must show that each successive environment in the developmental hierarchy inherits certain initial conditions and physical constancies from the environment that preceded it, back to the birth of the universe, and that some of these initial conditions and constancies act to predictably constrain the future dynamics of each successive environment, to some degree. Such constraints have been called developmental attractors (or more commonly, just attractors) by a variety of scholars. If they are the only such attractors on the adaptive landscape at that level of complexity and timescale, in a universe where accelerating complexification is possible, and such acceleration results in local dominance of the most rapidly improving systems, then I think it is clarifying to call them developmental portals (gateways, checks, bottlenecks) as well (Smart 2016a). For specific examples, G-, K- and M-class stars and organic chemistry may be necessary portals to planets capable of generating life. Fats, proteins, and nucleic acids may be necessary portals to cells. Eyes may be necessary portals to higher nervous systems. Tree niches (which support complex prediction), and animals with grasping appendages, language, and technology use may be necessary portals to human civilization acceleration, etc.
From an adaptive landscape (phase space) perspective, if ED is occurring, as the evolutionary “search” landscape gets more diverse and locally complex, certain portions must convert into developmental funnels, then portals. These portals must also work together to periodically produce a metasystem transition (a higher level of order or control), a new level of ED hierarchy. Both the landscape’s tendency to produce funnels/portals as complexity emerges, and the number of portals (lower is generally better) are two obvious ways to maintain developmental control in any evolutionary (chaotic, creative, locally unpredictable) system.
It is widely agreed that physical complexification, and such riddles as the origin of life, must be described by non-equilibrium thermodynamics as a coupled cascade of phase transitions in energy degradation and information production. As Smith and Morowitz (2016) state, each emergence (phase transition) in the development of hierarchy creates new simplifying constraints and logic, and there is a floor and a ceiling of environmental complexity for which those constraints apply. Reductionism can be very successfully applied at each level to discover its internal constraints (laws of chemistry, biology, etc.) The holism problem comes at the transitions (portals) between hierarchies. Some combination of bottom-up (evo, atomistic) and top-down (devo, holistic) parameters are involved, but how this works in any transition still remains unclear.
Evo-devo genetic and epigenetic models, as they seek to differentiate developmental and "evolutionary" gene fitness landscapes, will have to incorporate phase space models and (wherever there is high dimensional reduction) landscape models as our theory, tools, and data advance. Unfortunately, there are many problems with current adaptive and fitness landscape models in depicting the hyperspace of structure and function, and as critics of the adaptive landscape metaphor point out (Kaplan 2008) few models incorporate any concept of probability of movement across the landscape. In useful landscape models, potential portals would have to emerge as persistent, and theoretically globally optimal peaks (or in a more thermodynamically useful depiction, valleys) on adaptive fitness landscapes. These models will eventually have to evolve into network-based models with search basins and portal paths, which include both "evolutionary" tangles of similar-fitness landscapes and portals (Crutchfield and van Nimwegen 2002, picture right) as well as regions that use portals to predictably transition to globally optimal, hierarchical and developmental forms, landscape locations offering the greatest resource (space, time, energy, matter) efficiency or density of adaptation.
Another field that will help evo devo models advance will be protein folding, which already use funnel (portal) landscapes to depict 2D to 3D transformation of protein structure, involving both energy minimization and information production or conservation, a key example of biomolecular evolutionary development. In evo devo models, alternative chemistries for life, periodically sought by astrobiologists (see Goodwin et al. 2014) if they continue to be undiscovered by observation or simulation, would be more evidence indicating a universe with a high level of ED (self-organizing) constraint on the life transition. Science fiction authors and origin of life theorists have been imagining them for decades, but so far we haven't found any particularly credible alternatives, in my view. Such constraint (only one physico-chemical portal for the life transition being accessible in reasonable astronomical time, see Koonin 2007), if it exists, might be due to strong or weak multiversal selection for life and intelligence with both evo and devo properties, over many past cyclings of our universe.
In addition to better simulation capacity, progress in any theory of evolutionary development will require better:
- Complex systems theory - Seeing the appropriate system and scale at which ED is occurring, and any information-dependent processes (goals, drives) that may operate all in complex adaptive systems. I offer one such speculative model (Five goals of complex systems) in Smart 2017b.
- Evo-devo theory - Better understanding organismic ED, modularity, reaction-diffusion systems, gene-protein regulatory networks, intelligence, immunity, morality, and other ED features of living systems, both individually and as collectives. This will require advances in evo-devo genetics and epigenetics, theoretical morphology, paleontology, evolutionary (developmental) biology and psychology, anthropology, sociology, economics, and many other fields.
- Adaptation theory - Moving beyond the MES (modern evolutionary synthesis) to evo devo models, including self-selection (intelligence) and self-organization (development) as sources of adapted order.
- Optimization theory - Reliably differentiating less-optimized convergence (LOC) and optimized convergence (OC) in the emerging study of convergent evolution, via better definitions, tools, data, models and optimization functions.
- Acceleration theory - Understanding accelerating change, in ED terms. When it happens as a physical process, acceleration always seems to involve both densification and miniaturization of critical adaptive processes in complex systems. Speculative proposals like the transcension hypothesis (Smart 2012) and the stellivore hypothesis (Vidal 2016) extrapolate accelerating densification trends in adaptive systems to their universal limit, a black hole. Acceleration also happens via informational or computational processes as well. For that we may need a better intelligence theory.
- Intelligence theory (cog evo devo) - The Baldwin effect is the recognition, beginning with James M. Baldwin in 1896, that learned behavior affects an organism’s reproductive success. It is a modest start in understanding learning and intelligence in evo devo systems, but we must go much farther. The better we understand the evo and devo roles for cognitive processes in adaptation (and today we often do not) the better we may understand the roles of intelligence for the universe as a replicator. I can imagine (figure right, Smart 2017b) at least five goals of complex systems: innovation, intelligence, interdependence, immunity, and sustainability, each of which may be considered a form of intelligence. All of these goals may be self-selecting in evo devo systems, and their interaction a primary driver of adaptation, as follows:
- Intelligence as innovation (exploratory intelligence) – Evolutionary process is the hallmark of this type of intelligence. As Shapiro 2011 and others propose, living systems harness stochasticity to generate selectable variety (experiments, possible futures), particularly under stress or after catastrophe. They seek to do this in increasingly clever (“good bet”) ways, the smarter they become. Evolutionary innovation is nonrandomly guided by intelligence, particularly in the “next adjacent” action and feedback cycle. At the same time, the complexity generated becomes rapidly unpredictable the farther ahead any intelligence looks.
- Intelligence as intelligence (representation intelligence) - Most fundamentally, intelligence is a process of informational representation of environmental reality (Fischler and Firschein 1987). Informational representation (modeling) can be argued to be a dominantly divergent, evolutionary process. Our neural models conform to regularities of their environments, but they also generate astounding numbers of exploratory representations, only a fraction of which are universal (predictable) or adaptive. Think of imagination, fiction, or theoretical math, most of which has no known application. Being “intelligent” is also no guarantee of being adaptive. Indeed, those with too much of this single ability may be maladaptive. The finite nature of all intelligence (its computational incompleteness) also strongly argues that massive parallelism is a fundamental adaptive evolutionary strategy. All models are wrong, but some are useful. Massive parallelism, and regular selection on that parallel variety, seems key to how genes, neural nets, populations, and civilizations get more adaptive.
- Intelligence as interdependence (empathic-ethical intelligence) – Organisms engage in positive sum games, rules and algorithms (morality, ethics), involving not just self- and world-modeling but collective competition and cooperation, coordinated by other-modeling and other-feeling (empathy). Complex interdependent organisms develop culture, which evolves and develops independently from the individual, both faster and more resiliently, and allows them to view and optimize outcomes from either personal or group perspectives (which may conflict). A variety of universal evolutionary and developmental ethics (algorithms that protect collective adaptation and intelligence) may apply to all complex cultures. For more on how emergent synergies (interdependences) may have driven major transitions in evolutionary development, see Corning and Szathmáry 2015.
- Intelligence as immunity (security intelligence) – Organisms use many strategies for differentiating self from other, and passively and actively countering degradation and predation. Chronic stress and stress avoidance both weaken immunity, but right-sized cyclic stress and catastrophes both build immune system capacity and accelerate evolutionary innovation. Taleb’s concept of antifragility argues this for organizations, as does the catalytic catastrophe hypothesis. If there are universally discoverable security architectures and strategies (many ways to fail, only a few ways to survive), as I suspect, then immunity can be classed as a dominantly convergent and developmental process.
- Intelligence as sustainability (predictive intelligence) - Developmental process itself is the hallmark of this type of intelligence. Organisms use their intelligence not just to explore possible (innovation, intelligence) and preferable (interdependent, immune) futures, but to build predictive, and presumably Bayesian, models of probable futures. A subset of these predictive models are encoded in an organisms developmental genes, in emergent properties of their soma, in their environment, and in more complex organisms, culture. The growth of knowledge, common sense, science, and all the processes of development that predict, but do not protect (immunity) can all be considered sustainability. These processes grow “truth” and understanding. This form of intelligence is in constant tension with innovation, which can rapidly cause both poorly understood and dangerous forms of complexity to emerge.
- Intelligence substitution - Understanding precisely when information, or a computational process, can substitute for a physical process, and either retain or improve adaptiveness for the system under study. Some scholars call this dematerialization, or ephemeralization. Along with densification, dematerialization seems to be a central driver of accelerating complexification (Smart 2016b)
- Intelligence partitioning - Adaptation and intelligence always exist in three interacting subsystems: seeds (with evo and devo initial conditions), organisms (which engage in a life cycle), and the selective Environment (some scholars call this ambient intelligence). Because of niche-construction or stigmergy (intelligence always alters its local environment, in minor or major ways, changing adaptive pressures), environments essentially replicate along with seeds and organisms (think of the replication we see in city structure and function) and are a full partner with organisms and seeds in the ED of intelligence.
- Hierarchy theory - Seeing the ED trajectory for the system as a whole. Stan Salthe’s work on subsumptive hierarchies is an excellent example. Hierarchy theory (Salthe 1985, 1993, 2012) tells us how each new hierarchy is in some sense more free and in another more constrained than the latter. While we traditionally think of intelligence in an evolutionary role (increasing diversity and options), hierarchies tell us the ways that new “higher” systems are also more developmentally constrained than the ones from which they emerged. Using Volk's concept of "combogenesis" (Volk 2017), we can think of chemistry as both a set of new freedoms (to space- and time- efficiently create new adaptive structure and function) and new constraints on the local dynamics of physical laws. Biology locally enables and constrains chemistry, society locally enables and constrains biology, and so on. In a physical universe, such nesting and accelerating hierarchies must have a limit, a point at which further evolutionary development cannot proceed within this universe (Smart 2008, 2012).
- Information theory - Convergent evolution in biology can be modeled as the result of networks made up by biomolecules or other agents that are organized and structured by information hierarchies emerging via top-down causation. The emergence of modularity and of functional equivalence classes in subroutines – both in biological and technological systems – can be explained via such information hierarchies. Top-down causation describes the process whereby higher levels of emergent informational organization in structural hierarchies constrain the dynamics of lower levels of organization. In a typical reductionist paradigm it is assumed that purely physical effects entirely determine the dynamics of lower levels of organization and, by extension, at higher levels as well. But an emerging school of investigators hypothesize that the transition from non-life to life, abiogenesis, requires a top-down transition in causation and information flow (Flack 2017; Walker et al. 2017).
- Life cycle theory - Seeing the full replicative cycle of the developing system. If we can predict the remaining stages of the life cycle in any system, aided by comparisons with other evo devo systems, we can see its developmental futures, in broad outline at least. Its evolutionary futures, of course, remain intrinsically unpredictable at the same time. Both predictable and unpredictable process are perennially found in complex systems, whether an organism, a culture, a star, a galaxy, or a universe.
Building better hypotheses and theory of evolutionary and developmental processes will be an immense amount of work. But this path may be the only viable way forward (a conceptual developmental portal) to fully understanding such scientific challenges as convergent evolution, galactic development, and the origin of life. If validated, the benefits we stand to gain, via better collective foresight, also seem comparatively immense.
Observation selection effects: The challenge of assessing them for fine-tuning and convergence
Any form of reasoning about traits or properties which constitute observers, or that are logically or physically necessary for the existence of observers, is subject to observation-selection effects and biases. The importance of these selection effects and biases has only recently begun to be fully appreciated. For example, the number of small bodies’ (asteroids and comets) impacts in Earth’s history is constrained by our existence at the present time through the “anthropic shadow” effect (Ćirković, Sandberg, and Bostrom 2010).
Several detailed reviews of observation selection effects exist (for example, see Bostrom 2002). Observer selection arguments and models are often used to critique the fine-tuned universe hypothesis. Unfortunately, our cosmological models remain quite primitive, so it is easy to argue that either fine tuning or observer selection bias are more important in such models. But even more fundamentally, as I have argued before (Smart 2008) all observer selection models in common use depend on a random observer self-sampling assumption, a random distribution of possible universes, or some other random, Monte Carlo-style mathematical framework in their evaluations. In other words, they assume we live in an essentially random, evolutionary universe, and this is a major limit to their utility.
Consider that if we actually live in an evo devo universe, such math must itself be incorrect. If our universe is not just randomly (contingently) evolving, but it also nonrandomly developing, then some subset of its critical probability distributions (informational and dynamical) will continually be skewed in the direction of the universe's developmental trajectory, given its special initial conditions and constraints. As complexity and hierarchy grow in local environments, those environments will further bias a special subset of locally constraining, nonrandom developmental processes to occur. Such developmental biases may be why accelerating complexification occurs in special environments (acceleration bias) and why spontaneous abortions (miscarriages) are so frequent early in gestational development, but so rare late in development, with miscarriage frequency in humans declining from 40% of pregnancies at conception to 0.1% of pregnancies at 42 weeks of gestation (Rosenstein et al. 2012). Presumably, the more developmentally complex both the fetus and the gestational environment become, the less often that any survivable random perturbations of a standard size or duration are disruptive (stability bias).
Math that describes an increasing developmental bias toward both acceleration and toward growing informational stability during complexification in special environments is the kind of math we may need to properly model developmental processes, and to properly understand complex observers. Such observers are not random, they are privileged, in some proportion to their complexity. We surely do not live in an anthropic universe, if by that we mean one self-organized for the end purpose of producing biological humans. But we may well live in a noetic (intelligence-centric) universe, self-organized to produce accelerating and increasingly stable intelligent observers, as a central adaptive strategy for the universe itself. Biological humans may well be one of a long chain of developmental purposes, both an important and a transitory intelligence substrate (Hoyle 1983; Gardner 2007).
Even the frequency of evolutionary convergence, versus presumably much more commonplace evolutionary contingency in biological change, is a complex issue we don't yet agree upon. In the Signor-Lipps Effect, because rarer species are much less common in the fossil record, and the record itself is so sparsely sampled, rarer species will seem to disappear long before their actual time of extinction, simply due to chance. This makes the timing and speed of mass extinctions, and the ancestry of specific genera both much harder to determine from paleontology alone. As one consequence, quick and dramatic extinctions can go undetected, as they may look gradual due to selective sampling of a poor record, a classic observation selection effect (Signor and Lipps 1982).
Unfortunately, even the consequences of known biases like this are not clear. On one hand, if extreme ecological perturbations have overturned entire faunas via mass extinction events at much greater rates than we presently appreciate, and more rapidly switching Earth's macroevolutionary regimes (Jablonski 1986), then the contingency that life is subject to due to intrinsic evolutionary (diversity-generating) mechanisms may be less than is currently theorized. Our environment, not evolution itself, may be our greatest source of contingency. On the other hand, it has long been argued that environmental catastrophes themselves act as a major (and perhaps even the primary) catalyst for evolutionary innovation (Gerhart and Kirschner 2005; Bhullar 2017). We know that immune systems depend on catastrophe and hormesis to get stronger, and evolutionary gene complexes may also depend on catastrophe and chaotic stress to innovate. Without periodic catastrophe, some biologists have argued that evolutionary contingency must steadily reduce with time (Salthe 1993; Shapiro 2011). In stable niches, stabilizing genes presumably gain the upper hand over innovating genes. At other times, evolutionary innovation itself has provided the catastrophic environmental change, spurring further evolutionary innovation, as in the Great Oxygenation Event, causing massive dieoffs due to oxygen-excreting cyanobacteria, the End-Ediacaran extinction of large sessile organisms due to the emergence of mobile sighted animals during the Cambrian Explosion, and the Permian-Triassic extinction, perhaps precipitated by the emergence of Methanosarcina, a methane-synthesizing archaea (Ray 2017).
Geerat Vermeij has offered a particularly interesting and relevant argument supporting high frequency of convergence in evolution based on the logical asymmetry between singular and multiple events in the incomplete terrestrial fossil record (Vermeij 2006). If some specific character, call it Z, demonstrably evolved multiple times in different lineages in different ecological conditions, this is clearly an argument for the convergent nature of Z. However, if Z evolved only once to the best of our knowledge, it is still not an argument for uniqueness, since the incompleteness of our record may hide repeated instances of the independent evolution of Z. This logical asymmetry has intriguing consequences when a wider set of various important innovative characters scattered over all of the history of life are analyzed from a Bayesian point of view. Vermeij was able to show that the alleged singular innovations tend to be either more ancient or to appear in small clades. Small clades, in turn, are more invisible in the fossil record if located in the more distant past. In other words, purportedly unique innovations in small clades, or in the distant past, may be only the latest or the dominant instances of convergent evolutionary events, with most past convergences hidden from our view.
Vermeij concludes that few innovations are ever truly unique: “Purportedly unique innovations either arose from the union and integration of previously independent components or belong to classes of functionally similar innovations. Claims of singularity are therefore not well supported by the available evidence. Details of initial conditions, evolutionary pathways, phenotypes, and timing are contingent, but important ecological, functional, and directional aspects of the history of life are replicable and predictable.” (Vermeij 2006, p. 1804) Insofar as key evolutionary innovations are largely determined by universal principles of physics and economics, they will lead to widely-to-universally useful designs. This is the classic view of environment-driven convergence.
Ambiguity of the word “evolution”, and the modern evolutionary synthesis
In the scientific literature, the term “evolution” is used to describe any process of growth or change that involves the accumulation of historical information, in either living or nonliving complex systems (Myers 2009). When we restrict the term to refer to biology, and modern forms of Darwinian evolution, it is used to describe cumulative inherited change, via descent with modification from preexisting organisms. A classic conceptual model of Darwinian evolution, often taught in undergraduate classes, is the acronym VIST. Evolutionary change is proposed to happen via Variation, with Inheritance, and (Natural) Selection, over long amounts of Time (Russell 2006).
While it is a good start, there are three basic problems with the VIST model:
- VIST does not explicitly consider the concept of development, and of developmental genes and processes, which act in opposition to processes of variation within the organism. Developmental genes and processes are those that keep the organism on a convergent, conservative life and reproduction cycle. Their fundamental role is Convergence, funneling the organism toward a series of future-specific states. Variation, within the organism or within the environment, is the “enemy of development.” It must be overcome by Convergence, if the organism is to develop in a predictable way. Unfortunately, both classical Darwinism and modern evolutionary theory deprioritize the influence of organismic development on macrobiological change.
- VIST does not explicitly consider cumulative Replication, and its growing informational constraints on organizational change, in any substrate, over cumulative life cycles. Replication is implicitly considered as the factor of “Time” in the classic VIST model. But it is not Time that causes biological change. Organic change occurs via cumulative and increasingly ergodic cycles of Replication (of the organism), within any substrate, as guided by Inheritance factors (genes, brains, and other information carriers, or “seeds”), and Selection (in the environment). In all three of these interacting systems (organisms, seeds, and the environment) we find processes of Variation (evolutionary processes) and Convergence (developmental processes), working together in service to adaptation. Considered together, these five factors give us the VCRIS model of change. After variation and convergence themselves (what changes, and what doesn’t), replication is the next most fundamental process we should acknowledge in any model of the self-organization of complex adaptive systems. Whether we are discussing replicating suns creating organic chemistry, replicating chemicals creating cells, replicating cells creating organisms, replicating organisms creating ideas, or replicating ideas creating self-replicating machines, we must recognize that the most complex forms of adaptation, learning, and intelligence always require replication, inheritance, and life cycle, in some kind of "organism" (system).
- VIST doesn't recognize that the natural environment may itself be not only evolving (creating unpredictable experimental variety, by our definition above) but also developing, if our universe is itself a replicator. As a result, the Modern Evolutionary Synthesis, our current standard in biological investigations, is biased toward the idea of an “accidental” universe, and “random” experimentation and diversity as the primary (or in some views, exclusive) cause of macrobiological change.
Evo devo models, whether in biology or in other replicating systems, help us correct the biases of both the original Darwinian VIST view of evolution (white oval, figure right), and of modern evolutionary theory (light gray oval, figure right), both of which view diversification as the primary source of adaptiveness. Each of these views ignores or minimize the converging, conserving role of development, and the possibility of development on scales far larger than the organism. An evo devo-centric perspective (dark grey oval, figure right, for the case of living systems) will allow us to see that complex adaptive systems must harness both unpredictable, divergent evolutionary stochasticity and predictable, convergent developmental destiny and life cycle in search of greater adaptiveness, and that these two sets of mechanisms act in productive opposition to and tension with each other, at every scale. Evo devo models allow us to see evo devo self-organization as the natural source of adapted complexity and causal order in all successfully replicating systems, which we must come to understand from both physical and informational perspectives.
As we'll discuss shortly, understanding self-organization also shows us why challenges to Darwinism that have been launched by groups like the “intelligent design” community are more in line with supernatural belief, not science. They are typically motivated by belief in an “intelligent designer.” But if the universe replicates, as several cosmologists propose, parsimony and evidence both argue that evo devo self-organization, via many past replications in a selective environment, not intelligent design, is the source of the intelligence we see.
After we have done our best to adjust for observer-selection effects, we still see many highly unreasonable examples of adaptedness for complexification, in the laws and processes of our universe as a system. The phenomenon of accelerating change, evolutionary constraint laws (like the constructal law and various scaling laws), terminal differentiation of morphological complexity, the fine-tuned universe hypothesis, the presumed fecundity of Earth-like planets, the collective morality of social animals, and the Gaia hypothesis (in a more rigorous form) all come to mind. To explain such unreasonable adaptedness for complexification in our universe we should think first of replicative self-organization under selection, not design. After all, such self-organization is our best model for the source of the intelligence that is reading this page, right now.
As we come to understand the complex phenomenon of convergent evolution, on myriad system levels (physical, chemical, genetic, morphological, functional, algorithmic, cognitive, technological, etc.), we will rectify the historical biases that the Modern Evolutionary Synthesis (MES) have perpetuated with respect to our presumably living in a “random”, “directionless” and “purposeless” universe. To do this, we will need what Pigliucci and Müller (2010) and in a particularly comprehensive review, Laland et al. (2015) call an Extended Evolutionary Synthesis (EES). I expect this synthesis must include evo-devo and evo devo perspectives, a better theory of intelligence, better science and simulations, and more.
A large and well-funded group exploring an EES, led by evolutionary biologists Kevin Laland and Tobias Uller, can be found at ExtendedEvolutionarySynthesis.com. Another group working on an EES, led by biochemist and molecular biologist James Shapiro and physiologist Denis Noble, can be found at The Third Way of Evolution. This latter website is admirable, but not entirely error-free. As biologist Jerry Coyne points out in a post at the Richard Dawkins Foundation, the web editor of the Third Way website, Raju Pookottil, who does not have biology training, once argued that life "carries the hallmarks of design." That is a useful critique (see my section on the Fallacy of Intelligent Design below) but Coyne's post also ignores the scientific contributions of the many eminent meta-Darwinist scholars listed at the website. In exploring an EES, both poor evolutionary thinking and ultraorthodoxy with respect to the modern synthesis must be avoided.
"Ultra-Darwinists" like Coyne and Dawkins are any scholars who advocate, with a confident certainty, the position that contingent evolutionary selection (neo-Darwinism) can be the only force driving macrobiological change. Though Darwinism has deep evidence behind it, and its models appear to aptly describe the vast majority of organic change, they also seem insufficient to explain a small subset of phenomena that appear developmental, not evolutionary. That subset includes the accelerating development of intelligence, and the increasingly nonrandom guidance of evolutionary innovation in intelligent systems (Smart 2008). In a similar way, the vast majority of change we can sample at the molecular scale in biology seems locally stochastic, randomly selectionist, and diversity generating ("evolutionary"), yet a small subset, as we have proposed, also seems deterministic, convergent, and predictively selectionist ("developmental"), particularly when viewed from larger or longer-range spatial, temporal, energetic, and material (STEM) scales. Better defining and empirically discovering that subset seems a reasonable next step in evo devo inquiries.
The value of religion, and the fallacy of intelligent design as an explanation for adapted complexity
Religious belief is deeply valuable to many of us, and religious communities are globally important social institutions. Religion is humanity's first effort at universal moral philosophy, greatly predating Greek natural philosophy, and religion often ventures first into areas of moral prescription where science cannot yet easily tread. History shows that religious community has provided invaluable guidance and public benefit for millennia, and that all of our most socially successful religions are continuously reforming their beliefs and practices to be congruent with accelerating scientific knowledge.
Yet a key insight in the philosophy of science is that every intelligence is woefully finite and incomplete relative to both the current and future complexity of physical and informational reality. Thus we all must live with our own pragmatic sets of unproven beliefs, and many of us will seek communities that share those beliefs. It seems inevitable that self-aware artificial intelligences, if and when they eventually emerge, must also evolve and develop their own set of religious beliefs (read: philosophies of universal purpose, meaning and value), as there will remain many areas of reality about which they will know little. Fortunately, freedom of religious practice and freedom from religious discrimination are bedrocks of all modern democracies. Besides the traditional religions, atheism, agnosticism, possibilianism, universism (my own belief), and others are all belief systems that offer valuable, unproven beliefs about metaphysical reality.
As good practitioners of science and natural philosophy, all of us should attempt to make our unproven beliefs explicit and public, and seek to test them with evidence and experiment in whatever partial ways we can. When we feel we cannot separate our unproven beliefs from the practice of science, we should declare our influences. Unfortunately a number of scholars in the intelligent design community do not do this, and their religious belief has led some members of these communities to a variety of objectionable political acts, like seeking "equal treatment" for their evidence-poor hypotheses in the science classrooms of our public high schools. Given the intelligent design community's position on mixing religion and science, and not declaring their supernatural beliefs, we do not welcome scholars affiliated with the Discovery Institute or other intelligent design or creation science communities within the Evo Devo Universe (EDU) research community, and non-naturalistic discussion of religious belief is outside the scope of our community.
Given that there are fully naturalistic, evolutionary developmental explanations for the sources of "design" we see in living systems, and given the sharply limited value of all known physical intelligence, the concept of intelligent design, as it is generally proposed, appears fallacious. We must recognize that adapted intelligence has always had a useful but very minor influence on processes of selection in VCRIS systems. A crucial insight is that no physical intelligence ever becomes "Godlike" in its ability to predict either its own or its environmental future. We must acknowledge that all our present attempts to "rationally design" our own environment, including our genetically modified organisms, must be more accurately characterized as intelligence-guided guesses at more adaptive forms and functions. Human science and engineering are always evo devo process, like all other natural processes. They might be 95% bottom-up creativity/ experiment/ serendipity, and 5% top-down discovery/ optimization/ prediction, if we were to guess a very rough ratio. They are never a fully top-down or future-omniscient "design". The universe, and all evo devo systems, are far too chaotic and contingent to allow such omniscient foresight.
If we live in an evo devo universe, it is easy to argue that our future must continue to become rapidly computationally opaque to any finite and physical beings, the further ahead they look into their own futures. Combinatorial explosions of possibilities and contingencies, both in the universe itself and in our own mental processes, must always limit our foresight. No matter how advanced we become, any intelligences generated by this universe, or its ancestors, seem destined to remain evo devo "gardeners" as opposed to omniscient engineers, finite beings with “free will” (self-unpredictable evolutionary futures) not gods.
Supernaturalism takes many forms, some quite subtle. Even otherwise deeply insightful works, like EDU scholar and complexity theorist James Gardner's Biocosm (2003), which some critics read as an attempt to "split the difference" between a God-created and self-organized universe, run into trouble when they speculate that our universe may have been rationally constructed (read: intelligently designed) by "godlike" entities in a previous cycle. Such models simply don't fit with all materialist experience to date with respect to intelligence's role in replicating systems within our own universe.
Just as life’s incredibly adapted complexity self-organized under selection, over many evo devo cycles, and just as everything that is complex and adaptive inside our universe is a replicating system, it is most parsimonious to assume that our universe is a replicating evo devo system as well. If it is, its evo devo intelligence will always remain a limited and incomplete aid to selection, not a "godlike" designer. We may think a highly adapted design offers evidence of a designer (Paley 1802) but this argument has been exhaustively refuted by the rise of evolutionary theory with respect to biological systems (Darwin 1859), and we should expect it to be defeated for an evo devo universe as well.
If our universe replicates, either in isolation or as part of a fractally replicating multiverse, as some cosmologists propose, evolutionary developmental self-organization under selection seems the simplest explanation for such curious universal features as our improbably fine-tuned initial conditions, the robust emergence of adaptive complexity and intelligence, our improbably self-correcting geophysical environment, our continually accelerating complexity on Earth (Sagan’s “Cosmic calendar” metaphor), even under periodic catastrophe, and other puzzling aspects of our complexity emergence story so far. We have no need to invoke supernatural entities to explain such phenomena, and we have found no credible evidence, in our five hundred year epic of science advancement, for an intelligent designer.
- How do we best improve our physical and informational theories of unpredictable evolutionary and predictable developmental process?
- What improvements to complex systems theory, evo-devo theory, adaptation theory, optimization theory, acceleration theory, intelligence theory, hierarchy theory, life cycle theory, and other topics will help us better define, delineate, and compare evo and devo process in all replicating complex systems?
- How can we better define evolutionary and developmental process as sources of intelligence, in seeds (containing initiating evo and devo parameters), organisms, and environments?
- What evo and devo goals (purposes, telos) can we discover for intelligent complex systems?
- To what extent can we find modularity, reaction-diffusion systems, immunity, intelligence, and other features of organismic ED in ecosystem ED? In biogeographical ED? In stellar-planetary ED? In galactic ED?
- What empirical and statistical tools and tests can help us to infer developmental processes in biology, based on past experience with other organisms, when we do not have the capacity to simulate development causally? Can we use those tools and tests to help us infer hierarchy and life cycle in the universe as well?
- How do we best improve our models, simulations, and tests, especially falsification tests, for the universe as an evo devo system?
The author thanks Evo Devo Universe co-directors Clement Vidal, Georgi Georgiev, Michael Price, and Claudio Flores-Martinez for helpful critiques. Special thanks to EDU member Milan Ćirković who offered extensive constructive feedback on the earliest version of this paper. Thanks also to Anthony Aguirre, Yaneer Bar-Yam, John Leslie, Denis Noble, Reudiger Vaas and Tyler Volk for key insights, and to Carlos Gershenson and the CCS2018 committee for approving our satellite on Evolution, Development, and Complexity at CCS2018, where a version of these ideas were discussed.
- Adami, Chris (2016) What is Information? Phil. Trans. Royal Soc. A 374(2063)20150230.
- Adams, Fred C. (2008) Stars in Other Universes: Stellar structure with different fundamental constants, arXiv:0807.3697 [astro-ph].
- Adams, A.M., Zenil, H., Davies, P.C.W., and Walker, S.I. (2016) Formal Definitions of Unbounded Evolution and Innovation Reveal Universal Mechanisms for Open-Ended Evolution in Dynamical Systems (preprint, ArXiv.org).
- Aguirre, Anthony (2016) Why There is "Something" rather than "Nothing", Interview by Lawrence Kuhn for Closer to Truth (YouTube, 4 min)
- Arthur, Wallace (2000) The Origin of Animal Body Plans: A Study in Evolutionary Developmental Biology, Cambridge U. Press.
- Aunger, Robert (2007a) Major transitions in 'big' history, Technological Forecasting and Social Change 74(8):1137-1163.
- —————— (2007b) A rigorous periodization of 'big' history, Technological Forecasting and Social Change 74(8):1164-1178.
- Banerjee, S. and Moses, M. (2010) Scale Invariance of Immune System Response Rates and Times, Swarm Intelligence 4(4):301-318.
- Barrow, John D. and Tipler, Frank (1986) The Anthropic Cosmological Principle, Oxford U. Press.
- Barrow, John D. et al. (2008) Fitness of the Cosmos for Life: Biochemistry and Fine-Tuning, Cambridge U. Press.
- Bejan, A. and Errera M.R. (2017) Wealth inequality: The physics basis. J. Applied Physics 121(12):124903.
- Bejan A. and Zane J.P. (2013) Design in Nature: How the Constructal Law Governs Evolution in Biology, Physics, Technology, and Social Organizations, Anchor.
- Bhullar, B-A.S. (2017) Evolution: Catastrophe triggers diversification, Nature 542:304-305.
- Bostrom, N. (2002) Anthropic Bias: Observation Selection Effects in Science and Philosophy, Routledge.
- Bostrom N. and Ćirković, M. M. (Eds.) (2008) Global Catastrophic Risks, 1st Edition, Oxford U. Press.
- Bourgine P. and Stewart J. (2004) Autopoiesis and Cognition, Artificial Life 10: 327–345.
- Bray, Dennis (2011) Wetware: A Computer in Every Living Cell, Yale U. Press.
- Brin, David (1983) The ‘Great Silence’: the Controversy Concerning Extraterrestrial Intelligent Life, Q.J.R. Astr. Soc. 24:283-309.
- Callebaut, Werner and Rasskin-Gutman, Diego (2005) Modularity: Understanding the Development and Evolution of Natural Complex Systems, MIT Press.
- Calude, Cristian S. and Jürgensen, Helmut (2005) Is complexity a source of incompleteness? Advances in Applied Math 35(2005)1-15.
- Carroll, Sean B. (2005) Endless Forms Most Beautiful: The New Science of Evo Devo, Norton.
- Carroll, Sean M. (2016) The Big Picture: On the Origins of Life, Meaning, and the Universe Itself, Dutton.
- Carter, Brandon (1974) Large Number Coincidences and the Anthropic Principle in Cosmology. IAU Symposium 63, Reidel. pp. 291-298.
- Chaitin, Gregory J. (1992) Information-Theoretic Incompleteness, World Scientific.
- Chaisson, Eric (2001) Cosmic Evolution: The Rise of Complexity in Nature, Harvard U. Press.
- Ćirković, Milan M. (2009) Fermi's Paradox – The Last Challenge for Copernicanism? Serbian Astronomical Journal 178, 1-20.
- Ćirković, M.M., Sandberg, A., and Bostrom, N. (2010) Anthropic Shadow: Observation Selection Effects and Human Extinction Risks, Risk Analysis 30:1495-1506.
- Clausius, Rudolf (1851) On the Moving Force of Heat, and the Laws regarding the Nature of Heat itself which are deducible therefrom, London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science. 4th. 2 (VIII): 1–21; 102–119.
- Comte, August (1844) Discours sur l'Esprit Positif [A General View of Positivism], Paris.
- Conway-Morris, Simon (1998) The Crucible of Creation: The Burgess Shale and the Rise of Animals, Oxford U. Press.
- —————— (2004) Life's Solution: Inevitable Humans in a Lonely Universe, Cambridge U. Press.
- —————— (2015) The Runes of Evolution: How the Universe Became Self-Aware, Templeton Press.
- Corning, Peter A. and Szathmáry, Eörs (2015) "Synergistic selection": A Darwinian frame for the evolution of complexity. J. Theoretical Biology 371(2015)45-58.
- Corning, Peter A. (2018) Synergistic Selection: How Cooperation has Shaped Evolution and the Rise of Humankind, World Scientific.
- Crane, L. (1994) Possible Implications of the Quantum Theory of Gravity: An Introduction to the Meduso-Anthropic Principle (PDF). Foundations of Science Preprint 1994, no. Special Issue of the First Conference on the Evolution and Development of the Universe (EDU-2008).
- Crutchfield J.P. and van Nimwegen E. (2002) The Evolutionary Unfolding of Complexity. In: Evolution as Computation, Landweber L.F., Winfree E. (Eds.), Natural Computing Series, Springer.
- Dartnell, Lewis (2014) The Knowledge: How to Rebuild Our World from Scratch, Penguin Press.
- Darwin, Charles (1859) (On) The Origin of Species (by Means of Natural Selection), John Murray Press.
- Davies, Paul (2010) The Eerie Silence: Renewing Our Search for Alien Intelligence, Houghton Mifflin Harcourt.
- Dick, Steven J. (1996) The Biological Universe: The Extraterrestrial Life Debate and the Limits of Science, Cambridge U. Press.
- Dryden, D.T.F., Thomson, A.R., and White, J.H. (2008), How much of protein sequence space has been explored by life on Earth?, J. R. Soc. Interface 5:953-956.
- Einstein, Albert (1915) Die Feldgleichungen der Gravitation (The Field Equations of Gravitation), Sitzungsberichte der Preussischen Akademie der Wissenschaften zu Berlin, 844–847.
- Eldredge, N. and Gould, S.J. (1972) “Punctuated equilibria: an alternative to phyletic gradualism,” In: Models in Paleobiology, T.M. Schopf (ed.), Freeman & Cooper, pp. 82-115.
- Ellis, George F.R. (2015) Recognizing Top-Down Causation, in A. Aguirre et al. (Eds.) Questioning the Foundations of Physics, Springer, pp. 17-44.
- Ellis, G.F.R., Noble, D. and O’Connor, T. (2012) Top-down causation: an integrating theme within and across the sciences?, Interface Focus 2:1-3.
- Fischler, Martin A. and Firschein, Oscar (1987) Intelligence: The Eye, the Brain, and the Computer, Addison-Wesley.
- Flack, Jessica (2017) Life's Information Hierarchy. In: From Matter to Life: Information and Causality, Walker, Sara I. et al. (Eds.), Cambridge U. Press.
- Flores-Martinez, Claudio L. (2014) SETI in the light of cosmic convergent evolution. Acta Astronautica 104(1):341-349.
- —————— (2017) Introducing Biomimomics: Combining Biomimetics and Comparative Genomics for Constraining Organismal and Technological Complexity. In: Mangan M. et al. (Eds.) Biomimetic and Biohybrid Systems: Proceedings of Living Machines 2017, Springer.
- Forgan, D., Dayal, P., Cockell, C., and Libeskind, N. (2017) Evaluating galactic habitability using high-resolution cosmological simulations of galaxy formation, International Journal of Astrobiology 16:60-73.
- Forward, Robert L. (1980) Dragon’s Egg, Del Rey.
- Friston, Karl (2010) The free-energy principle: a unified brain theory?, Nature Reviews Neuroscience 11(2010):127-138.
- Gardner, Andy and Conlon, Joseph P. (2013) Cosmological natural selection and the purpose of the universe, Complexity 18:48-56.
- Gardner, James (2003) Biocosm: A New Scientific Theory of Evolution, Inner Ocean Publishing.
- —————— (2007) The Intelligent Universe: AI, ET, and the Emerging Mind of the Cosmos, New Page.
- Georgiev, Georgi Y. et al. (2015) Mechanism of organization increase in complex systems, Complexity 21(2)18-28.
- Gerhart, John C. and Kirschner, Marc W. (2005) The Plausibility of Life: Resolving Darwin's Dilemma, Yale U. Press.
- —————— (2007) The theory of facilitated variation, PNAS 104 Suppl 1:8582-9.
- Gilbert S.F., Bosch, T.C.G. and Ledón-Rettig, C. (2015) Eco-Evo-Devo: Developmental symbiosis and developmental plasticity as evolutionary agents, Nature Reviews Genetics 16(10):611-622.
- Gödel, Kurt (1931) On formally undecidable propositions of Principia. Mathematica and related systems (German), Monatshefte für Mathematik und Physik, 38(1)173–198.
- Goodwin, Jay T. et al. (2014) Alternative Chemistries of Life: Empirical Approaches, NASA/NSF Workshop.
- Gould, Stephen J. (1977) Ontogeny and Phylogeny, Harvard U. Press.
- —————— (2002) The Structure of Evolutionary Theory, Harvard U. Press.
- Haken, Hermann (1984) The Science of Structure: Synergetics, Van Nostrand Reinhold.
- Hall, Brian K. (ed.). (2003) Environment, Development, and Evolution: Toward a Synthesis, MIT Press.
- Harrison, Edward R. (1995) The Natural Selection of Universes Containing Intelligent Life (PDF). Quarterly Journal of the Royal Astronomical Society 36(3)193-203.
- Held, Lewis I. (2009) Quirks of Human Anatomy: An Evo-Devo Look at the Human Body, Cambridge U. Press.
- —————— (2014) How the Snake Lost its Legs: Curious Tales from the Frontier of Evo-Devo, Cambridge U. Press.
- —————— (2017) Deep Homology?: Uncanny Similarities of Humans and Flies Uncovered by Evo-Devo, Cambridge U. Press.
- Henderson, Lawrence J. (1913) The Fitness of the Environment: An Inquiry into the Biological Significance of the Properties of Matter, Macmillan.
- Heylighen, Francis (2007) The Global Superorganism: an evolutionary-cybernetic model of the emerging network society. Web published manuscript. Retrieved from: http://pespmc1.vub.ac.be/Papers/Superorganism.pdf
- —————— (2008) Accelerating socio-technological evolution: from ephemeralization and stigmergy to the Global Brain. In: Globalization as Evolutionary Process: Modeling Global Change, George Modelski, Tessaleno Devezas, and William R. Thompson (eds.), Routledge.
- —————— (2016) Stigmergy as a Universal Coordination Mechanism. Cognitive Systems Research 38:4-13.
- Hoyle, Fred (1954) On nuclear reactions occurring in very hot stars: the synthesis of the elements from carbon to nickel, Astrophysics Journal Suppl. 1:121-146.
- —————— (1983) The Intelligent Universe: A New View of Creation and Evolution, Holt, Rinehart and Winston.
- Jablonka, Eva and Lamb, Marion J. (1995) Epigenetic Inheritance and Evolution: the Lamarckian Dimension, Oxford U. Press.
- —————— (2005) Evolution in Four Dimensions: Genetic, Epigenetic, Behavioral, and Symbolic Variation in the History of Life, MIT Press.
- Jablonski, D. (1986) Background and Mass Extinctions: The Alternation of Macroevolutionary Regimes, Science 231:129-133.
- Jacobs, Lucia (2012) From Chemotaxis to the Cognitive Map: The function of olfaction. PNAS, 109(1):10693-10700.
- Jantsch, Erich (1980) The Self-Organizing Universe: Scientific and Human Implications, Pergamon.
- Johnson, Norman L. (2011) What a Developmental View Can Do for You. In: Thought Leader Forum, Michael J. Mouboussin (Ed.), Credit Suisse/First Boston.
- Kaplan, Jonathan (2008) The end of the adaptive landscape metaphor? Biol Philos 23:625-638.
- Kauffman, Stuart (1993) The Origins of Order: Self-Organization and Selection in Evolution, Oxford U. Press.
- Koonin, Eugene V. (2007) The cosmological model of eternal inflation and the transition from chance to biological evolution in the history of life. Biology Direct 2:15-36.
- Krumholz, M.R. and McKee, C.F. (2005) A general theory of turbulence-regulated star formation, from spirals to ultraluminous infrared galaxies. The Astrophysical Journal 630:250-268.
- Kurzweil, Ray (1999) The Age of Spiritual Machines: When Computers Exceed Human Intelligence, Penguin.
- —————— (2005) The Singularity is Near, Penguin.
- Laland, Kevin et al. (2015) The extended evolutionary synthesis: Its structure, assumptions and predictions, Proceedings of the Royal Society B282:20151019
- Lane, Nick (2016) The Vital Question, W.W. Norton.
- Laplace, Pierre-Simon (1812) Theorie analytique des probabilites (A Philosophical Essay on Probabilities), Translated from the fifth French edition of 1825 with Notes by the Translator, Andrew I. Dale, Springer-Verlag, 1995.
- Leibniz, Gottfried (1686) The Discourse on Metaphysics (Fr: Discours de métaphysique).
- Lem, Stanislaw (1971/1999), The New Cosmogony, In: A Perfect Vacuum (trans. by M. Kandel), Northwestern U. Press, pp. 197-227.
- Leslie, John and Kuhn, Robert L. (2013) The Mystery of Existence: Why Is There Anything At All?, Wiley-Blackwell.
- Levin, Simon A. (1998) Ecosystems and the Biosphere as Complex Adaptive Systems, Ecosystems 1:431-436.
- Lewis, G.F. and Barnes, L.A. (2016) A Fortunate Universe: Life in a Finely Tuned Cosmos, Cambridge U. Press.
- Lightman, Alan (2011) The Accidental Universe, Harpers, Dec 2011.
- Lineweaver, Charles H.; Fenner, Y.; Gibson, B.K. (2004) The Galactic Habitable Zone and the Age Distribution of Complex Life in the Milky Way, Science 303(5654):59–62.
- Linde, Andrei D. (1992) Stochastic approach to tunneling and baby universe formation, Nuclear Physics B 372:421-442.
- Lobkovsky, A.E. and Koonin, E.V. (2012) Replaying the tape of life: quantification of the predictability of evolution, Frontiers in Genetics 3:246-254.
- Lorenz, Konrad (1977) Behind the Mirror: A Search for a Natural History of Human Knowledge, Harcourt, Brace, Jovanovich.
- Losos, Johnathan B. (2017) Improbable Destinies: Fate, Chance, and the Future of Evolution, Riverhead Books.
- Louis, A.A. (2016) Contingency, convergence and hyper-astronomical numbers in biological evolution, Studies in History and Philosophy of Biological and Biomedical Sciences 58:107-116.
- Lucas, J. R. (1961) Minds, Machines and Gödel. Philosophy 36:112-137.
- Luhmann, Niklas (2002/2013) Introduction to Systems Theory, Polity Press.
- Luisi, Pier L. (2003) Autopoiesis: a review and a reappraisal. Naturwissenschaften 90 49–59.
- Malone, Michael S. (2012) The Guardian of All Things: The Epic Story of Human Memory, St. Martin's Press.
- Mariscal, Carlos (2016) Universal biology: assessing universality from a single example, In: The Impact of Discovering Life beyond Earth, Steven J. Dick (ed.), Cambridge U. Press.
- Martin, Joseph D. (2013) Is the Contingentist/Inevitabilist Debate a Matter of Degrees? Philosophy of Science 80(5):919-930.
- Maturana, H.R. and Varela, F.J. (1973/1980) Autopoiesis and Cognition: The Realization of The Living, D. Reidel.
- —————— (1987) The Tree of Knowledge: The Biological Roots of Human Understanding, Shambhala.
- McGhee, George R. (2011) Convergent Evolution: Limited Forms Most Beautiful, MIT Press.
- McLeish, T.C.B. (2015) Are there ergodic limits to evolution? Ergodic exploration of genome space and convergence, Interface Focus 5:41-53.
- Meissner, Ulf-G. (2013) Life on earth - An accident? Chiral symmetry and the anthropic principle, Int. J. Mod. Phys. E 23.
- Miller, James G. (1978) Living Systems, McGraw Hill.
- Mingers, John (1995) Self-Producing Systems (PDF). Kluwer Academic/Plenum.
- Moravec, Hans (1979) Today's Computers, Intelligent Machines, and Our Future. Analog Science Fiction and Fact, Feb 1979.
- Müller, Gerd B. and Newman, Stuart A. (Eds.) (2003) The Origination of Organismal Form: Beyond the Gene in Developmental and Evolutionary Biology, MIT Press.
- Munitz, Milton K. (1987) Cosmic Understanding: Philosophy and Science of the Universe, Princeton U. Press.
- Meyers, Robert A. (2009) Encyclopedia of Complexity and Systems Science, Springer.
- Nazaretyan, Akop P. (2017) Psychological Background of the 21st Century Global Challenges. BAOJ Psychology 2: 029.
- Newton, Isaac (1687) The Mathematical Principles of Natural Philosophy, Daniel Adee, 1846.
- Noble, Denis (2017) Evolution viewed from physics, physiology and medicine. Interface Focus 7(5):20160159.
- Noble, Raymond and Noble, Denis (2017) Was the Watchmaker Blind? Or Was She One-Eyed? Biology 6(4):47 (Special Issue: Biology in the Early 21st Century: Evolution Beyond Selection).
- Odling-Smee, John et al. (2003) Niche Construction: The Neglected Process in Evolution, Princeton U. Press.
- Ogura A., Kazuho I., Gojobori, T. (2004) Comparative analysis of gene expression for convergent evolution of camera eye between octopus and human. Genome Research 14:1555-1561.
- O’Malley, M.A. and Powell, R. (2016) Major problems in evolutionary transitions: how a metabolic perspective can enrich our understanding of macroevolution, Biology and Philosophy 31:159–189.
- Oparin, Alexander I. (1968) Genesis and Evolutionary Development of Life, Academic Press.
- Orgogozo, V. (2015) Replaying the tape of life in the twenty-first century, Interface Focus 5:57-68.
- Paley, William (1802) Natural Theology or Evidences of the Existence and Attributes of the Deity, John Morgan Press.
- Pearce, T. (2012) Convergence and Parallelism in Evolution: A Neo-Gouldian Account, Brit. J. Phil. Sci. 63:429-448.
- Piel, Gerard (1972) The Acceleration of History, Knopf.
- Pigliucci, M (2007) Do we need an extended evolutionary synthesis?, Evolution 61:2743-2749.
- Pigliucci, M., Müller, G.B., (Eds.) (2010) Evolution: The Extended Synthesis, MIT Press.
- Pinker, Steven (2010) The Better Angels of Our Nature: Why Violence Has Declined, Penguin.
- Parker, Andrew (2003) In the Blink of an Eye, Basic Books.
- Powell, R. (2012) Convergent evolution and the limits of natural selection, Euro. J. Phil. Sci. 2:355-373.
- Powell, R. and Mariscal, C. (2015) Convergent evolution as natural experiment: the tape of life reconsidered, Interface Focus 5:40-53.
- Pohorille, Andrew (2012) Processes that Drove the Transition from Chemistry to Biology: Concepts and Evidence. Origins of Life 42(5)429-432.
- Price, Michael (2017) Entropy and selection: Life as an adaptation for universe replication (PDF). Complexity, Article ID 4745379, 4 pages, 2017. doi:10.1155/2017/4745379
- Pross, Addy (2014) What is Life? How Chemistry Becomes Biology, Oxford U. Press.
- Poundstone, William (1985) The Recursive Universe: Cosmic Complexity and the Limits of Scientific Knowledge, William Morrow & Co.
- Raff, Rudolf (1996) The Shape of Life: Genes, Development, and the Evolution of Animal Form, U. of Chicago Press.
- Raia, P. and Fortelius, M. (2013) Cope's Law of the Unspecialized, Cope's Rule, and weak directionality in evolution (PDF). Evolutionary Ecology Research 15:747-756.
- Ray, Georgia (2017) Evolutionary Innovation as a Global Catastrophic Risk, Presentation at EA Global 2017, San Francisco, CA.
- Rees, Martin (1999) Just Six Numbers: The Deep Forces that Shape Our Universe, Basic Books.
- —————— (2001) Our Cosmic Habitat, Princeton U. Press.
- —————— (2015) Post-human evolution on Earth and beyond. Presented at Exploring Exoplanets: The Search for Extraterrestrial Life and Post Biological Intelligence, John Templeton Foundation conference.
- Reid, Robert G.B. (2007) Biological Emergences: Evolution by Natural Experiment, MIT Press.
- Rosenstein, Melissa G et al. (2012) Risk of Stillbirth and Infant Death Stratified by Gestational Age, Obstet Gynecol 120(1):76-82.
- Russell, Cathy M. (2006) Retrieved from Epicofevolution.com/biological-evolution, 27 Mar 2017.
- Russell, D. A., and Séguin, R. (1982) Reconstructions of the small Cretaceous theropod Stenonychosaurus inequalis and a hypothetical dinosauroid (PDF). Ottawa: National Museums of Canada, National Museum of Natural Sciences.
- Sagan, Carl (1977) The Dragons of Eden: Speculations on the Evolution of Human Intelligence, Random House.
- —————— (1980) Cosmos, Random House.
- Salthe, Stanley M. (1985) Evolving Hierarchical Systems, Columbia U. Press.
- —————— (1993) Development and Evolution: Complexity and Change in Biology, MIT Press.
- —————— (2010) Development (and evolution) of the universe. In: The Evolution and Development of the Universe, C. Vidal (ed.), Special issue, Foundations of Science 15:357-367.
- —————— (2012) Hierarchical structures, Axiomathes 22:355-383.
- Sanderson, M. and Hufford, L. (Eds.) (1996) Homoplasy: The Recurrence of Similarity in Evolution, Academic Press.
- Schlosser, G. and Wagner, G.P. (Eds.) (2004) Modularity in Development and Evolution, U. of Chicago Press.
- Shapiro, James A. (2011) Evolution: A View From the 21st Century, FT Press Science.
- Signor, P.W. and Lipps, J.H. (1982) Sampling bias, gradual extinction patterns, and catastrophes in the fossil record, In: Geological implications of impacts of large asteroids and comets on the Earth, L.T. Silver and P.H. Schultz (Eds.), Geological Society of America Special Publication, 190:291-296.
- Smart, John M. (2000) Intro to the developmental singularity hypothesis (DSH), AccelerationWatch.com. Retrieved from: Accelerationwatch.com/developmentalsinghypothesis.html, 29 Jan 2018.
- —————— (2002) Understanding STEM compression in universal change, AccelerationWatch.com. Retrieved from: Accelerationwatch.com/mest.html, 29 Jan 2018.
- —————— (2008) Evo Devo Universe? A Framework for Speculations on Cosmic Culture. In: Cosmos and Culture: Cultural Evolution in a Cosmic Context, Steven J. Dick, Mark L. Lupisella (Eds.), NASA Press.
- —————— (2012) The transcension hypothesis: Sufficiently advanced civilizations invariably leave our universe, and implications for METI and SETI, Acta Astronautica 78:55-68.
- —————— (2015) Humanity rising: Why evolutionary developmentalism will inherit the future, World Future Review 7(2-3):116-130.
- —————— (2016a) Portals (funnels, bottlenecks) and convergent evolution. In: The Foresight Guide. Retrieved from Foresightguide.com/portals-funnels-bottlenecks-and-convergent-evolution, 20 Jan 2018.
- —————— (2016b) The great race to inner space. In: The Foresight Guide. Retrieved from: Foresightguide.com/the-great-race-to-inner-space-our-surprising-future/, 28 Mar 2017.
- —————— (2017a) The VCRIS model of natural selection. In: The Foresight Guide. Retrieved from: Foresightguide.com/the-vcris-model-of-natural-selection-evolutionary-development-of-adaptive-complexity, 6 Jun 2018.
- —————— (2017b) Five goals of complex systems. In: The Foresight Guide. Retrieved from: Foresightguide.com/five-goals-of-complex-systems, 30 Dec 2017.
- —————— (2018) Catalytic catastrophes: Advancing the five goals. In: The Foresight Guide. Retrieved from: Foresightguide.com/catalytic-catastrophes-how-right-sized-catastrophes-advance-the-five-goals/, 20 Jan 2018.
- Smith, Eric and Morowitz, Harold J. (2004) Universality in intermediary metabolism, Proc. Natl. Acad. Sci. 101:13168–13173.
- —————— (2016) The Origin and Nature of Life on Earth, Cambridge U. Press.
- Smolin, Lee (1992) Did the Universe Evolve? Classical and Quantum Gravity 9:173-191.
- —————— (1997) The Life of the Cosmos, Oxford U. Press.
- —————— (2004) Cosmological natural selection as the explanation for the complexity of the universe. Physica A 340:705-713.
- —————— (2006) The status of cosmological natural selection. arXiv:hep-th/0612185v1
- —————— (2012) Scientific Approaches to the Fine-Tuning Problem, The Nature of Reality, Nova.
- Spencer, Herbert (1864) Illustrations of Universal Progress: A Series of Discussions, D. Appleton.
- Steele, Edward J. (1981) Somatic Selection and Adaptive Evolution: On the Inheritance of Acquired Characters, 2nd Ed., U. of Chicago Press.
- —————— (1998) Lamarck's Signature: How retrogenes are changing the natural selection paradigm, Perseus.
- Stenger, Victor J. (2011) The Fallacy of Fine-Tuning: Why the Universe is Not Designed for Us, Prometheus.
- Susskind, Leonard (2006) The Cosmic Landscape, Back Bay.
- Swenson, Rod (1992) "Galileo, Babel, and Autopoiesis (It's Turtles All The Way Down)". Int. J. General Systems 21 (2): 267–269.
- Taleb, Nicholas (2012) Antifragile: Things That Gain from Disorder, Random House.
- Teilhard de Chardin, Pierre (1955) The Phenomenon of Man, Harper & Row.
- Turchin, Valentin (1977) The Phenomenon of Science, Columbia U. Press.
- Turner, D.D. (2011) Gould’s replay revisited, Biology and Philosophy 26:65-79.
- Tyson, Neal D. (2006) The Perimeter of Ignorance, Presentation at: Beyond Belief Conference, Nov 5-7, 2006, YouTube (41 mins).
- Vaas, Ruediger (1998) Is there a Darwinian evolution of the cosmos? In: Proceedings of the MicroCosmos-MacroCosmos Conference, Aachen.
- Varela F.J., Maturana, H.R. and Uribe R. (1974) Autopoiesis: the organization of living systems, its characterization and a model, Biosystems 5(4):187-196.
- Verhulst, Jos (2003) Discovering Evolutionary Principles through Comparative Morphology, Adonis Press.
- Vermeij, Geerat J. (1987) Evolution and Escalation, Princeton U. Press.
- —————— (2006) Historical contingency and the purported uniqueness of evolutionary innovations, PNAS 103:1804-1809.
- —————— (2009) Nature: an economic history, Princeton U. Press.
- Vidal, Clement (2008) The future of scientific simulations: from artificial life to artificial cosmogenesis (PDF). In: Death and Anti-Death 6: Thirty Years After Kurt Gödel (1906-1978).
- —————— (2010) Computational and biological analogies for understanding fine-tuned parameters in physics (PDF). Foundations of Science 15(4):375-393.
- —————— (2016) Stellivore Extraterrestrials? Binary Stars as Living Systems (PDF). Acta Astronautica 128:251–56.
- Vilenkin, Alexander (2006) Many Worlds In One: The Search for Other Universes, Hill and Wang.
- Volk, Tyler (2003) Gaia's Body: Toward a Physiology of Earth, MIT Press.
- —————— (2017) Quarks to Culture: How We Came to Be, Columbia U. Press.
- Wagner, Gunther (2003) Hox cluster duplications and the opportunity for evolutionary novelties, PNAS 100(25):14603–14606.
- Wagman and Stephens (2004) Surprising 'ultra-conserved regions discovered in human genome. UCSC Currents
- Walker, Sara I. et al. (Eds.) (2017) From Matter to Life: Information and Causality, Cambridge U. Press.
- Webb, Stephen (2015) Where is Everybody? Seventy-Five Solutions to the Fermi Paradox and the Problem of Extraterrestrial Life, Springer.
- Weinberg, Steven (2007) "Living in the Multiverse", In: Universe or Multiverse? B. Carr (ed.), Cambridge U. Press.
- West, Geoffrey (2017) Scale: The Universal Laws of Growth, Innovation, Sustainability, and the Pace of Life in Organisms, Cities, Economies, and Companies, Penguin.
- West-Eberhard, Mary Jane (2003) Developmental Plasticity and Evolution, Oxford U. Press.
- Wheeler, John A. (1977) Genesis and observership, In: Foundational Problems in the Special Sciences, R. E. Butts and J. Hintikka, (Eds.), D. Reidel, pp. 3-33.
- —————— (1988) World as system self-synthesized by quantum networking, IBM Journal of Research and Development 32:4-15.
- Wigner, Eugene (1960) The unreasonable effectiveness of mathematics in the natural sciences. Comm. on Pure and Applied Math., 13(1):1-14.
- Wikipedia (2008), and Smart, John M. and Vidal, Clement (2008-2017). Cosmological natural selection (fecund universes). EvoDevoUniverse.com wiki.
- Wikipedia (2012), and Smart John M. and Chattergee, Atanu (2012-2017). List of examples of convergent evolution. EvoDevoUniverse.com wiki.
- Wilkins, Adam S. (2001) The Evolution of Developmental Pathways, Sinauer Associates.
- Wright, Sewall (1932) The roles of mutation, inbreeding, crossbreeding, and selection in evolution. Proceedings of the Sixth International Congress on Genetics. pp. 355–366.
- Yi, Hong et al. (2010) Gene expression atlas for human embryogenesis. FASEB Journal 24(9):3341-3350.
- Zimmer, Carl (2015) A Planet of Viruses, 2nd Edition, U. Chicago Press.