Once found here; resurrected from Way Back Machine; and now the book.
THE NATURE OF AUTONOMOUS AGENTS
AND THE WORLDS THEY MUTUALLY CREATE
STUART A. KAUFFMAN
SEPTEMBER 13, 1996
The following document is being published as a Santa Fe Institute Preprint. The material below is not yet science. However, it is serious “protoscience” -- an attempt to formulate questions and concepts that may, in due course, become serious science. I take the step of informal publication for two major reasons: archival and collegial. It seems sensible to publish, in a restricted venue, the results of these efforts, to set them out in a rough and ready way. The topic appears to me to be very large, and is likely to require the efforts of many scholars to bring the protoscience to the stage of real science. In view of the recent exciting if tentative evidence for life on early Mars, and the first example of an autocatalytic peptide, attempts to understand the character of autonomous agents and the worlds they make seems especially appropriate.
Several years ago, Doyne Farmer remarked that the study of complexity was rather like thermodynamics before Carnot gave us its defining concepts. This comment has rung in my mind over these years. In December 1994 I began a notebook to myself, entitled “Investigations,” knowing that too many strands were present to try to state clearly at that time. Too many strands are still raveling for clear statement. “Investigations” has grown to a 450 page note. In the summer of 1995 I attempted to bring some of these strands together in a set of five lectures given to the Santa Fe Institute Summer School on Complexity. The material below is largely that set of lecture notes. I have, however, included new issues in the document and rearranged the material into seven “lectures.” I had the good fortune to present the material below in a set of five lectures at the Niels Bohr Institute in Copenhagen, August 25-31, 1996. I take this opportunity to thank Holgar Nielsen, Don Bennett, Benny Lovtrup and their colleagues for their hospitality. The efforts below have benefited from discussion and comments by a number of colleagues including: Philip Anderson, Per Bak, David Sherrington, Lee Smolin, Brian Goodwin, Dan Stein, Richard Palmer, Bill Macready, Tim Keitt, Harold Morowitz. Because this document is informal, I have kept figures and references to a minimum.
The most general aim of “Investigations” is to explore whether there might, conceivably, be some general laws governing a certain class of non-equilibrium systems - namely the class of coevolutionary self-constructing communities of autonomous agents.
In the first “lecture,” I discuss previously published results and concepts to be found in sources including my two books, “Origins of Order” and “At Home in the Universe,” both by Oxford University Press. After pointing out the grounds to suppose that no general law is likely to govern all possible non-equilibrium systems, the central aim of the first lecture is to discuss the general conditions that allow evolution and coevolution to assemble complex systems. Not all complex systems can be assembled by an evolutionary process. The first lecture discusses the general character of “fitness landscapes,” of coevolutionary processes in which the adaptive moves of one “agent” deforms the fitness landscapes of its partners, and tries to show that an endogenous coevolutionary process allows “agents,” each adapting for its own selfish “fitness,” to tune the couplings and ruggedness of their fitness landscapes such that the entire system achieves a specific self-organized critical state. Here, as if by an invisible hand, the fitness of each agent appears to be maximized, its probability of extinction appears to be minimized, yet a power law distribution of small and large avalanches of extinction and speciation events propagates through the model ecosystem.
This self-organized critical state of a coevolving system of agents emerges as one part of a candidate general law characterizing coevolutionarily constructable systems of agents. While the model I discuss is the only one I know of that includes modification of landscape structure by coevolving agents, there are at least three other models of self-organized criticality in coevolutionary settings, by Bak and Sneppen, by Sole and Miramonte, and by Newman. All show similar phenomena.
The second lecture discusses a theory of the origin of life as a phase transition in chemical reaction networks from subcritical to supracritical behavior. This material is published in the books noted above. The recent publication in Nature of the first example of an autocatalytic peptide is, I hope, a forerunner of an understanding of phase transitions in chemical reaction graphs as the basic underpinning of the emergence of collectively self reproducing systems. The heart of this view of the emergence of life lies in the fact that as the diversity of organic molecules in a system increases, the diversity of reactions by which they transform to one another increases very much faster. In turn, this means that for a wide distribution of assumptions about which molecules catalyze which reactions, at a sufficient diversity so many reactions are catalyzed that a collectively autocatalytic set of molecules is virtually certain to emerge. On this view, the emergence of collectively reproducing systems of molecules is an expected phase transition in complex chemical systems.
Lecture two sets the basis not only for a general theory of the emergence of collectively autocatalytic systems of molecules, but also of a concept that then receives increased attention throughout the lectures: The Adjacent Possible. If we consider the species of organic molecules now on the earth, we can also consider those species of organic molecules that do not exist, but that are one reaction step away from molecules that currently do exist. The set of kinds of organic molecules that are one reaction step away from those that do exist constitutes the chemically “adjacent possible.” A feature of the biosphere is that the diversity of polymers and organic molecules in the biosphere has increased as the biosphere has expanded into the adjacent possible over the past 4 billion years. It is then a matter of considerable importance to understand what processes might govern and drive the flow into the molecular adjacent possible.
In pursing the biosphere’s persistent generation of novel small molecules, in the second lecture I ask whether the biosphere as a whole is supracritical, while cells must remain subcritical to avoid the lethal consequences of the uncontrolled synthesis of novel molecules at the entry of any novel molecular species. Then I broach the possibility that causally local metabolic communities, say microbial communities, are poised on the subcritical - supracritical boundary. Such poised communities would graft into themselves useful novel molecules without being destroyed by uncontrolled supracritical behavior. This suggests that the biosphere may be poised to advance, on average, into the molecular adjacent possible as rapidly as possible. Indeed this supposition becomes part of the tentative candidate law I later propose for coevolving self-constructing communities of autonomous agents.
In later lectures I return to the concept of the adjacent possible, for it appears to be an expression of a very general feature of the Universe: Since the number of possible large molecules, say with several tens of thousands of atoms per molecule, is vast, the Universe has not had sufficient time to make each such molecule at least once since the Big Bang. Thus, there are 10 raised to the 260 kinds of proteins length 200. If the universe has 10 to the 80th particles, and if a “reaction” required only a femtosecond, then the universe can have “tried” only 10 to the 193 pairwise “reactions” among the particles - a vast number, but vastly smaller than the number of proteins length 200. Thus, it is an essential feature of the universe that it is necessarily non-ergodic with respect to the generation of complex structures such as complex molecules, let alone species, legal systems and so forth.
That the Universe is non-ergodic with respect to such hierarchical complexity, and must be so on time scales vastly longer than the current lifetime of the Universe, may lie at the heart of a search for a “fourth law,” for we are entitled to ask, if this be true, whether general laws may govern the way matter-energy-information self-organizes to “flow” from “the Actual” into the Adjacent Possible. And in later lectures I return to this theme, wondering if such a general law has the form that the flow is such that the dimensionality of the adjacent possible, on average, expands as rapidly as it can. Indeed, I find myself wondering if this expansion itself constitutes “the arrow of time.”
In the third lecture I turn to a key issue: What properties must a physical system have such that it can be said to be able to “act on its own behalf.” A bacterium swimming upstream in a glucose gradient is acting on its own behalf, seeking dinner. Yet the bacterium is “just” a physical system. In attempting to answer this question, I have been led to a tentative definition of an “autonomous agent” making a living in an environment. By this definition, an autonomous agent is a collectively autocatalytic system that performs one or more thermodynamic work cycles. This definition leads to several new concepts. Among these:
i. An Autonomous Agent achieves a new kind of “closure” in a space of catalytic and work tasks such that the components of the Agent are amplified or reproduced and all the tasks are accomplished. Closure in a space of tasks such that this occurs is an objective property of real systems. Cells are just such collectively autocatalytic systems performing one or more linked work cycles. This closure, in turn, means that an Agent is a functionally coherent self sustaining system.
ii. Because a Carnot engine performs different tasks if run in a “forward” and a “reverse” direction, in general, Autonomous Agents must be displaced from equilibrium in a particular direction, and perform their work cycles in a particular direction to achieve task closure. Autonomous Agents are, necessarily, non-equilibrium systems.
iii. Autonomous agents such as cells coordinate “propagating work” - linked sequences of events many or all of which require that free energy be used to drive the process. Following Atkins, I note that work is the constrained release of energy - as in a steam engine, where the cylinder and piston provide constraints on the expansion of the gas that drives the motion of the piston. But autonomous agents themselves do work to construct the very constraints that channel the release of energy and allow it to constitute work. Constraint begets work begets constraint. Part of the collectively self-consistent coherence of an autonomous agent lies in the fact that each such constraint and the work that is engendered carries out a “task,” and the work tasks are linked such that all tasks are performed, all constraints constructed, and the autonomous agent maintains or reproduces itself.
iv. The emergence of collectively self-consistent linked webs of work and tasks is, like the emergence of collectively autocatalytic reactions networks where the number of reactions increases faster than the number of molecular species, abetted by the fact that any work done creates low energy structures, coherent motions, or flows whose own manifold causal consequences can be “jury rigged” in a wide variety of ways. As this variety and diversity of causal consequences increases, it is easier to achieve collective coherence. A flying cannon ball can be “used” to smash a house, turn a sturdy paddle wheel, create a warning whistling noise, pull an object by itself hitting a slack net attached to the object, etc. Since each bit of “work” done creates an ordered macroscopic state with many causal consequences, as diversity of work done increases, jury rigging at least one coherent web becomes easier. Simultaneously, the “glitches” in the self organizing jury rigged system, if they do not entirely block coordinated function, can become control points to tune timing and levels of work flows to optimize performance.
v. It is essential that only some, not all, of the causal consequences of any part or work constitutes the “task” carried out. Once the self-consistent coherent linked web of propagating work is in place, the “task” carried out by each bit of work among its many causal consequences. is definable in the context of the whole Autonomous Agent.
vi. In order to act on their worlds, Autonomous Agents must not use all the propagating work merely to construct the constraints that allow work to propagate. Some of the work must be done on the outside world, including other agents. Thus, a distinction between low energy signaling and constraint construction work that builds and coordinates the Agent, and work done on the world can be drawn.
vii. The fact that Autonomous Agents are necessarily displaced from equilibrium and perform work cycles means that agents can, and do “ratchet” themselves further from equilibrium.
viii. In turn, as displacement from equilibrium increases, the number of possible alternative accessible microstates of the system increases, thereby increasing the “adjacent possible.”
ix. In turn, the increase in the adjacent possible means that autonomous agents can more readily happen upon and incorporate novel means to coordinate the activities and work cycles and constraint construction that correspond to their evolving organization and reproduction.
x. Autonomous Agents play “natural games,” namely, ways of making a living. As the agents coevolve, so do their ways of making a living. Thus, the “winning games are the games the winners play.” This near tautology packs deeper meaning however. William Macready and David Wolpert at SFI have proven the “No Free Lunch” theorem showing that, averaged over all possible fitness landscapes, no search algorithm outperforms any other search algorithm. Thus, on average, evolution by mutation and selection is no better than random search.
The “No Free Lunch” theorem seems to imply that coevolutionary constructable systems of autonomous agents must simultaneously and self-consistently bring forth both the Agents and the Natural Games they play. Job and Jobholder arise as “duals” of one another. Good jobs proliferate. But good jobs are just the activities -- niches -- that allow a living to be made and are well searched and exploited by the current search procedures of the Agents. Thus, this returns to the themes of lecture 1. Agents persistently cocreate the worlds they “inhabit,” bringing forth ever changing webs of agents and their niches. Finally, I note that nothing in the definition of an Autonomous Agent demands that it be constructed of polymers and organic molecules. Perhaps spiral galaxies are also communities of autonomous agents involved in formation of complex molecular clouds and star formation. In the final lecture I broach the radical question of whether the universe as a whole might be a community of Plank-scale autonomous agents.
In the forth lecture I attempt to consider autonomous agents as non-equilibrium Maxwell Demons. To extract work from their environments such demons must make records of their environments, then pay an erasure cost. That cost would be minimized were the autonomous agents, in some sense, able to construct maximally compressed records of their environment. But almost certainly, autonomous agents cannot do so, for that would require that they themselves be, in some sense, maximally compressed. If so, they would not themselves live on smooth enough fitness landscapes to be coevolutionarily constructable. Thus, communities of autonomous agents must pay the additional erasure cost arising from redundancy, hence must be finitely displaced from equilibrium not only to act with sufficient velocity in real time, but to pay the added erasure cost.
But lecture four emphasizes very puzzling questions partially noted and developed in lecture three, concerning constraints and propagating work in Agents. We appear to have, as yet, no physical theory that comfortably unites matter, energy and information in a single dynamical framework. Yet Maxwell’s Demon is one place in physics where matter, energy and information come into intimate contact. A collectively autocatalytic, autonomous Agent performing linked propagating work, however, is characterized by a new and objective property: It is a physical system that achieves a collectively self-consistent functional “closure” in a space of catalytic and work tasks. An autonomous agent, a non-equilibrium Maxwell Demon, coevolving in its community has embodied know-how. It knows how to make a living in the context in which it lives and carries out real construction work in doing so. Its self consistent structure and dynamical logic constitute the embodied “record” of its environment, its reproduction and proliferation carries out linked work cycles and simultaneously, via mutation and selection, updates its record. A growing microbial community constitutes something like “propagating coconstructing organization of propagating constraints - work - record.”
The concept of a union of propagating constraint - work-record housed in propagating functionally whole agents seems to be missing in physics, but then current physics lacks the concept of an Autonomous Agent as an organization of matter and energy able to act on its own behalf. One insight into this issue is gained by asking why a “Carnot cycle” must be a “cycle” in the first place. Were it not, then the propagation of work, the constrained release of energy, would require a highly improbable concatenation of coordinated macroscopic states of entities able to receive propagated work as input, and, perhaps with some added work, propagate the work further. The cyclic arrangement of the Carnot cycle returns the machine to its initial state such that it can receive another input of energy and perform another work cycle. This leads me to consider that “organization” is fundamentally related to that coordination of matter and energy which enables and controls the constrained release of energy - work - to be propagated. Such coordination is achieved, fundamentally, by doing work to create structures or flows that alter the potential barriers involved in the release of energy from components in the system. As noted above, constraint begets work begets constraint. In cells, enzymes lower potential barriers along some reaction coordinates, but work was done to construct the enzymes. Work is done to construct the molecules that then self organize into a membrane. Thereafter, the spontaneous burying of substrates partially within a membrane may restrict degrees of freedom in the substrates, raising and blocking potential barriers to reactions in some but not other directions.
Organization, so defined to include work done to construct constraints, the resultant structure and dynamical logic, and propagating work is necessarily a non-equilibrium concept and is clearly not the concept of entropy - itself an equilibrium concept concerned with ergodic flow to macrostates having the largest number of microstates.
The life cycle of autonomous agents is precisely the cyclic self construction of systems controlling their own boundaries, hence constraining the release of energy such that it can constitute propagating work whereby the agents sustains itself or reproduces, while the organization itself modified by heritable variations and natural selection is the propagating updating of the “record.”
Outside of an autonomous agent, or a coevolving community of autonomous agents, the chance occurrence of intricate “Rube Goldberg” machines channeling the constrained release of energy are vastly improbable spontaneously. Nor would such linked systems be likely to sustain themselves. While channeling energy flows, the general Rube Goldberg does not construct itself. Thus, the cyclic life cycles of Autonomous agents coevolving with one another to create ecosystems and finding and incorporating new functionalities in the adjacent possible begin to appear to be the major way that organization arises and proliferates in the Universe. If building order requires degradation of free energy, then autonomous agents ratchetting themselves far from equilibrium, thereby storing energy and “recorded embodied know how” in structure and flow to control their own constrained release of energy, construction of themselves, and exploration of the adjacent possible, appear the paramount way to build up complexity.
Indeed, once such autonomous agents exist and create a world in which they proliferate, the Darwinian categories of “function” come into play. A given molecule is “food” or “poison” to a situated agent. Hence, once Agents exist, a genuine semantics, with a physical interpretation, appears to arise. A coevolving community of non-equilibrium Maxwell Demons is a union of matter-energy-information into an organization that proliferates and constructs hierarchical complexity.
Lecture 5 focuses on a formulation of a working hypothesis suggesting that communities of autonomous agents will come to lie in the vicinity of three apparently different phase transitions:
1) As parallel processing systems, such Agents will lie within the ordered regime or at a phase transition between dynamical order and chaos. In addition, communities of Agents will tune the couplings among the Agents such that the community as a whole is within the ordered regime, very near the phase transition to chaos. This position on an internal order-chaos axis should maximize the number of discriminations that Agents can make yet act without trembling hands as they play every fancier Natural Games. Thus autonomous agent Maxwell demons endogenously create their own relevant coarse graining of their worlds.
Recent results, developed since last summer and based on analysis of biases in the behavior of regulated eukaryotic genes, strongly suggest that eukaryotic genetic regulatory networks do, in fact, lie in the ordered regime close to the “edge of chaos.” This material is currently being written up as a separate article. If this proves true, then 1.6 billion years of eukaryotic evolution have achieved cells poised between convergent and divergent flow in their respective state spaces as, presumably, an optimized way to carry out the complex coordination of sensing, constructing, and acting that is the life cycle of a cell. If so, the result will be powerful support for this, the first part of the three part working hypothesis.
2) As the members of the community tune their positions on their internal order-chaos axis and their couplings to one another, these evolutionary adaptations simultaneously tune the ruggedness of fitness landscapes the web structure of the ecosystem, and the amount any Agent’s landscape is deformed by the mutational alterations of other agents. The coevolutionary system will approach a self-organized critical state with mean maximized fitness, mean minimized probability of extinction, but a power-law distribution of speciation and extinction events. Data on the size distribution of extinction events in the Record and artificial worlds, as well as life-time distributions at the level of Genera and Families, and the fractal character of higher taxa supports this part of the working hypothesis.
3) As parallel processing systems, Autonomous Agents will tune their internal molecular diversity and the species diversity of communities of Agents, such that each Agent is subcritical, perhaps near the subcritical - supracritical boundary, and such that the causally local community is just at the subcritical - supracritical boundary. Here the probability than a novel input molecule to the system undergoes reactions generating “descendant novel molecules” is exactly 1.0. Thus, novel molecules in the effectively local community should be generated in power-law bursts. Useful novel molecules will be grafted into the community by natural selection. Hence the biosphere will take a step into the chemical Adjacent Possible. If this view is correct, then the biosphere as a whole is composed of many effectively local communities, each at the subcritical-supracritical boundary, having the property that the sustained mean rate of propagation into the chemical adjacent possible is maximized.
These three aspects of this working hypothesis are fully testable. I discuss supporting evidence. More broadly, they suggest that the biosphere is non-ergodic, as is the Universe as a whole. The three component laws claim that any biosphere will self-organize to three phase transitions: The dynamical “edge of chaos” among a community of agents; a self-organized critical state among those agents considered as a coevolutionarily constructable ever changing system; a system poised on a subcritical-supracritical boundary at which the total chemical and perhaps functional diversity of the biosphere expands, on average, as fast as it can.
Given that the Universe is non-ergodic, as noted above, these candidate laws may be aspects of a general theory of self-constructing non-equilibrium systems within the Universe, and perhaps even of the Universe itself. The latter possibilities are the focus of chapters 6 and 7.
In the sixth lecture I raise new issues then try to bring several strands somewhat together. I note that there is an indefinite hierarchy of Agents, for Agents can be comprised of Agents. The hierarchy reflects a hierarchy of combinatorially complex entities. I discuss the emergence of true novelty in evolution - what biologists call “exaptations,” and raise the question as to whether any basement language of “functions” could prestate all possible functionalities that can arise in coevolving hierarchies of agents. I think not. I point out that “downward causation” is utterly non-mysterious. The last trilobite jumped the wrong way when some starfish caught it for dinner. But with the extinction of the organism - the trilobite - the earth lost its unique molecular species. Hence the extinction event, due to actions and effects among whole organisms, has changed the molecular unfolding of the biosphere. Causes run upward and downward seemlessly. This, plus the difficulty in finding a basement language for all possible functionalities in the Darwinian sense, raises the strongest doubts about the capacity of any fundamental theory to account and explain, let alone predict, all phenomena at “higher levels.” Hierarchical Agents are parts of the furniture of the Universe. As they arise, the way the universe behaves changes.
In the sixth lecture I return to the theme of the Adjacent Possible and try to give it a formal definition in terms of a classical 6N dimensional phase space. Let such a system of micro states be partitioned into small volumes. In systems so vast that flow along trajectories will always be non-ergodic, the number of adjacent volumes that are accessible from the current volume is the “adjacent possible.” Otherwise stated, the volumes next accessible are the current dimensionality of the “work space” of the non-ergodic system. The hypothesis that the biosphere is expanding as rapidly as is sustainably possible into the adjacent possible is the same as stating that the dimensionality of its work space is expanding as fast as is sustainably possible. This hypothesis implies that the phase volume within successive boxes along the non-ergodic trajectory in the 6N phase space are being progressively subdivided into finer volumes, each leading to a different successor box in the adjacent possible. Hence more internal symmetries are being broken. Just such symmetry breaking is occurring in Agents as they optimize their internal organization such that small fluctuations cross internal threshold between different pathways of self construction, categorizations, and actions in order to win the Darwinian race.
But there is a minimum set of dimensions concerning the subvolumes of a box in 6N which flow to different adjacent boxes set, in the quantum mechanics, by the Heisenberg Uncertainty Principle. Thus, if we pass from a classical to a quantum description of Agents, a task not yet even begun, then the symmetry breaking within each “box” that sets all the internal thresholds leading the system to flow in different directions, in enhancing the dimensionality of the adjacent possible, must hit the Uncertainty Principle limit. The expansion of the adjacent possible by coevolving “classical agents” seems to drive towards quantum behavior within the agents in a phase space box to maximize the flow into the adjacent possible. (A bacterium “making up its mind” converts quantum to classical behavior? A whiff of consciousness?)
The fact that higher order autonomous agents such as multicelled and colonial organisms, social insects colonies, and even economic entities such as firms, can, and have emerged in the biosphere suggests that not only at the molecular level, but at higher organizational and functional levels, the system will tend to advance into the adjacent possible as fast as is sustainably possible. Data supporting these concepts from the patterns of speciation and extinction events in the Record as well as the behavior of technologically evolving economic systems confronted with “future shock” and Schumpeterian gales of creative destruction, are considered. Evidence that the biosphere as a whole may have tuned its mean resident energy to maximize chemical diversity, hence its “work space,” is also considered.
In the seventh lecture I try to sketch some possible extensions of the ideas about a biosphere to the universe itself. These attempts are the most tentative in the set of lectures. They derive from the realization that nothing in the definition of an Autonomous Agent demands that it be made of organic molecules. The functional closure within an autonomous Agent such that it can be collectively autocatalytic and carry out work cycles yields the new concept that such a functionally “whole” agent is the locus of a union of matter, energy and information - propagating embodied know-how and work task cycles, that can do honest construction work. The ultimate question is whether a true “fourth law” might govern the non-ergodic behavior of the Universe such that, in some sense, it expands its “working” dimensionality as rapidly as is sustainably possible.
The most radical possibility that I broach, based on the work of Lee Smolin with respect to quantizing gravity in terms of spinnor networks is that such networks can form combinatorially complex “knotted” structures. Then, as molecules are combinatorially complex objects and at sufficient diversity undergo a phase transition to collectively autocatalytic systems, the same may be true of a knotted ravel of a spinnor-net. Thus, space (or space and mass-energy) might conceivably be comprised of autocatalytic autonomous Plank scale agents coevolving with one another. One interpretation of such agents is that they are the “crystallization” of seeds of classicity and its propagation.
The seventh chapter presents five themes. First, I attempt to suggest that we explore a thermodynamics in which the observer is an endogenous part of the system. Such an observer creates its own coarse graining. If one takes a microbial community governed by the hoped for three part candidate law above, then each agent optimizes its position on an internal order-chaos axis and its couplings to other agents such that the community is at the dynamical edge of chaos, and poised with respect to coevolution and the advance into the adjacent possible. At that poised state, each agent has optimized its coupling, hence its channel capacity, to its neighbors. The hope is to represent entropy as the information one agent can have about each of its neighbors or its environment. Since the agents are self constructing and coevolving, this hope suggests a formulation in which entropy is not just a measure of ignorance per se, but reflects the shared know-how enabling the system to coevolve and literally construct coordinated properties.
It is at least suggestive that agents share information over their boundaries, hence there are parallels to the fact that the entropy of a Black hole is proportional to its surface area, and to ideas about entropy being related to the surfaces between different volumes of space.
Second, I explore the question of whether a large quasi-closed thermodynamic system such as an isolated spiral galaxy with its giant molecular clouds and their complex chemistry forming carbon rich dust grains with a power-law distribution of sizes can be chemically supracritical and hence not reach equilibrium for much longer time scales than the lifetime of the universe. Since “glasses” can take very long times to equilibrate, this is perhaps not such an odd notion. Since life on our planet seems to exhibit this property and we are members of a spiral galaxy, perhaps this concept is correct. Further, since agents need not be made of organic molecules, it is not impossible that a galaxy itself might be comprised of a coevolving community of autonomous agents.
Third, these wonderings are not entirely unconnected to the concerns of Lee Smolin, as expressed in his soon to be published “The Life of the Cosmos.” Smolin points out that General Relativity seems to require that each point in space be different from every other point such that the “night sky” looks different at each point. And Smolin points out that the constants demanded by the Standard Model must be very finely tuned to obtain a Universe as complex as ours, with stars, carbon, chemistry, and life. Either an ultimate theory will derive the constants of the standard model, string theory, or some other theory - or we will be left with adjustable parameters. Smolin argues that some process of self-organization may be at work in the Universe as a whole that tunes the parameters. It may not be entirely foolish to wonder if the self-organization of a biosphere of self-constructing, coevolving autonomous agents in an expanding effective phase space may not be a version of the Universal puzzle.
Smolin explores the possibility that spinnor nets may constitute a kind of prespace. Ravels of such nets would constitute space itself, and the distance between two points in space would be a function of the similarity of the local “weave” structure.
As noted above, I sketch the possibility that spinnor nets might create combinatorially complex knotted structures which can become autocatalytic functional Agents that coevolve with one another. Among the possible implications, is this: Just as the biosphere is organized into higher order taxa, species, genera, families, classes, orders, and phyla and just as no new phyla have emerged since the Cambrian, any such evolutionary process will accumulate total variety, but ever more slowly. After an initial burst of production of variety, most differences will be minor variations. Thus, if variety is distance in space, space will expand, but ever more slowly. One might hope that this feature would relate naturally to the apparently critical character of the ever slower expansion of the Universe.
Furthermore, the supracritical autocatalytic ravel knot system might be expected to break some underlying time symmetry such that the expansion of the web constituted an arrow of time. If space has a chiral knotted structure, its “handedness” might break other symmetries, such as P, with respect to neutrinos.
Spinnor networks include links representing not only space (i.e., geometry) but also spinnons: electrons, quarks, and neutrinos, as well as gauge bosons: photons, W bosons, and gluons. Further, such networks allow space and matter to interconvert. This raises the possibility that the Universe might evolve so as to maximize its total complexity, including space + simple and complex matter. Hopefully, such evolution would flow to the boundary between an expanding and a contracting Universe. Were the early universe to “make” too much matter compared to space, it would rapidly recollapse, remaining simple. Were the early universe to “make” too much space compared to matter it would expand forever but contain only subatomic particles, hydrogen and helium, hence remaining simple. Being complex is tied to being near the boundary between expansion and collapse.
The image upon which this hope is based is that maximum growth of the Adjacent Possible in the flow of a non-ergodic Universe maximizes the rate of quantum “decoherence” and the emergence of classicity.
The fourth theme concerns this emergence of classicity in quantum mechanics. It is generally accepted that interaction of a classical and a quantum system causes the quantum system to decohere and become classic. This suggests that classicity is “autocatalytic.” Using the Bohm Hiley interpretation of quantum mechanics, I suggest that complex quantum entities interacting with one another have more possible outcomes and predecessor possibilities than simple entities, hence will decohere more readily. If so, perhaps complexity begets classicity which locks-in the complexity in a cycle in which the lock-in tends to drive a system of interacting particles towards higher complexity. If this might be true, then it may be an aspect of a non-ergodic flow into an expanding “work space,” and might tend to pick among possible quantum histories of the universe. That flow which maximizes complexity will emerge into classicity and propagate.
Lee Smolin’s version of this vision is that among the decohering quantum histories of the Universe, those in which the Adjacent Possible expands the fastest will decohere the most readily. Thus, the Universe might tend to flow towards maximum complexity.
Other workers seek a view under which the Universe self tunes the values of its constants or even its laws. Examples include Smolin’s cosmological natural selection, Lindemann’s budding Universe approach, Wheeler’s “It from bit,” and Holgar Nielsen’s “Random Laws” approach.
Thus, in the fifth theme, I raise the question of whether the Universe as a whole might self tune the constants of the Standard Model, or any other fundamental theory with free parameters, such that the set of values associated with the fastest amplifying complexity universe “wins.” Part of the mechanism driving the winning of the fastest complexifying Universe might be the lock-in due to the hoped for autocatalytic process linking complexity and classicity. At the far reaches, one can try to imagine that there is a persistent local competition among all possible laws, seen as rules of transformations of spinnor networks, such that, locally, the set of laws yielding the fastest expansion of the local adjacent possible always “wins.” Here the concept is that “nearby laws” imply nearby particles able to interact with one another, while “distant laws” would entail particles too different from one another to interact. Coevolution of neighboring sets of laws and the particles expressed by them would, hopefully, select out and sharpen upon that set of laws and associated particles which expands the total effective dimensionality of the Universe, geometry, simple matter and complex matter, the most rapidly. The hope is that a Universe constructed by coevolving spinnor-net-laws will self construct our Universe with classical large scale behavior. A fourth law, writ large, would have at its core a law governing the non-ergodic flow of a self constructing universe expanding its effective dimensionality as rapidly as is sustainably possible. Indeed, such maximal expansion and symmetry breaking would emerge as the central law governing the coevolution of the random laws of the spinnor network-law system. Here time itself would be nothing but this expansion.
In the Epilogue, I make a tentative suggestion concerning the physical basis of consciousness in molecular autonomous agents. The hypothesis appears at least to be consistent, not obviously false, and, I hope, testable.
The fundamental ideas derive from the suggestion that, in order to expand their capacity to make discriminations and refined actions, autonomous agents will partition a bounded volume in a 6N classical phase space representing the state of an agent or collection of agents, into ever finer volumes which ultimately encounter the Heisenberg Uncertainty Limit. Further partitioning would presumably involve quantum effects. This suggests that molecular autonomous agents would be expected to span the classical and quantum realms. Indeed, cells do. A single photon can cause rhodopsin in the retina to respond. A possible image of consciousness is linked to the simultaneous propagation of quantum coherence in percolating webs within living systems, for example, neural tissue, that is simultaneously decohering along side channels as phase information is lost. This persistent propagation of quantum coherent behavior and its decoherence into specific fixed states of matter is then interpreted as “mind acting on matter”. The percolation of quantum coherences and persistent decoherence is interpreted as consciousness itself in the experience of the agent. There are some similarities between this hypothesis and Penrose’s work, but suggested on a different basis. Any such theory has the difficulty that quantum coherence seems unlikely at body temperature. Conversely, propagating work in cells may imply percolating webs of “frozen” degrees of freedom constituting the constraints on the release of energy that itself constitutes work. Thus, such frozen webs might have effectively lower local temperatures hence support percolating propagating quantum behavior that also persistently decoheres.