THE NATURE OF AUTONOMOUS AGENTS
AND THE WORLDS THEY MUTUALLY CREATE
STUART A. KAUFFMAN
SEPTEMBER 13, 1996
INFORMATION, RECORDS, WORK, PROPAGATABLE
4.1) Maxwell's Demon and Measurement Costs in Equilibrium and Non-Equilibrium Settings.
This work derives from R. Landauer and W. Zurek.
i. Consider Maxwell's Demon operating a gas engine with a single gas molecule in it.
ii. Demon's aim is to measure the position of the single gas molecule, then use that information to extract work.
iii. Let the Demon successively partition the volume containing the gas molecule into half, quarter, ...etc.
iv. Each time the Demon determines which side of the next partition the gas molecule is in, he reduces the total entropy of the gas system by a constant amount, since entropy scales as the logarithm of the number of states where the gas molecule can be located, while successive halvings of that volume reduce the volume correspondingly.
v. As the Demon makes these measurements, he is to write down a "record" of the successive positions with respect to the successive partitions, leading to a binary string of length N after N successive halvings.
vi. The Demon will later use the record to extract work from the ever more precisely located gas molecule.
(Note it seems to take a DEMON (an autonomous agent) to use the record to extract work.).
vii. One now wants to consider the relation between the reduction in entropy gained by the successive measurements, the resulting work that can be extracted, and the ENERGETIC COST OF ERASING THE BITS AFTER THE WORK IS EXTRACTED, so that the memory tape is returned to its initial state and can be written on again.
viii. Landauer showed that the erasure cost per bit of the binary string in the RECORD is equal to the work extractable per 1 log unit reduction in entropy.
ix. The RECORD might, or might not, be MAXIMALLY COMPRESSED. If the record is NOT maximally compressed, then the cost of erasure of the record will be greater than were the record maximally compressed.
x. Zurek showed that, at equilibrium, the reduction in entropy achieved by the successive measurements by the Demon on average equal the cost erasure of the maximally compressed RECORD.
xi. Thus, Zurek showed that, at equilibrium, the reduction in entropy achievable by the Demon, hence the work extractable using that knowledge, on average, IS EXACTLY OFFSET BY THE COST OF ERASURE OF THE MAXIMALLY COMPRESSED RECORD OF THE MEASUREMENTS. This preserves the 2nd Law. Work cannot be extracted, on average, at equilibrium.
4.2) IN A NON-EQUILIBRIUM SETTING, MEASUREMENT "PAYS."
i. Zurek also showed that, in a non-equilibrium setting, the cost of erasure of the maximally compressed RECORD is less than the entropy reduction achieved, hence net work can be extracted by carrying out measurement, then using the information to extract work.
ii. The basic reason measurement pays in the non-equilibrium setting is that a simple description can be very useful. Suppose one considers a fluid flow. A single measurement: The flow is to the left, not the right, allows a single bit to reduce the entropy a great deal, hence allows us to place a water wheel in a position to maximize work extracted.
iii. The implication of this second result is that measurement only pays in a non-equilibrium setting.
iv. But further, the DISPLACEMENT REQUIRED FROM EQUILIBRIUM FOR MEASUREMENT TO BE USEFUL IN THE NET EXTRACTION OF WORK DEPENDS UPON HOW COMPRESSED THE RECORD IS. IF THE RECORD OF THE MEASUREMENTS IS NOT MAXIMALLY COMPRESSED, THEN A CORRESPONDINGLY GREATER DISPLACEMENT FROM EQUILIBRIUM IS NEEDED TO PAY THE EXTRA ERASURE COSTS.
v. Thus, IF THE RECORD IS "FINITELY" REDUNDANT, THEN THE SYSTEM MUST BE FINITELY DISPLACED FROM EQUILIBRIUM IN ORDER TO PAY THE EXCESS ERASURE COSTS AND STILL MAKE MEASUREMENT YIELD NET WORK.
vi. In Lecture 1 I gave grounds to believe that COEVOLUTIONARILY ASSEMBLABLE COMPLEX ADAPTIVE SYSTEMS, -- NOW TO BE CALLED AUTONOMOUS AGENTS -- WOULD CONSTRUCT A WORLD WHERE FITNESS LANDSCAPES ARE OF INTERMEDIATE RUGGEDNESS. MAXIMALLY COMPRESSED COMPUTER PROGRAMS, I SUGGESTED, WERE NOT EVOLUTIONARILY CONSTRUCTABLE.
vii. THUS, I SUGGEST THAT COEVOLUTIONARILY ASSEMBLABLE "SOCIETIES" OR ECOLOGIES OF AUTONOMOUS AGENTS WILL HAVE AGENTS WHICH CANNOT HAVE MAXIMALLY COMPRESSED RECORDS OF THEIR ENVIRONMENTS. IF SO, THEN SUCH AGENTS, LADEN WITH THE REDUNDANCY THAT ALLOWS EVOLUTION TO OCCUR, MUST BE FINITELY DISPLACED FROM EQUILIBRIUM TO EVOLVE.
4.3) THE CRAECRAFT CLUE.
Geologist Craecraft suggests an interesting possibility:
Consider two cases:
i. A mixture of hydrogen and oxygen at standard temperature and pressure. Slow "cold" combustion to yield water molecules. Compare this to a spark and explosive transformation to H2O. The heat generated by the combustion is itself part of the causal loop that maintains the continued rapid conversion of substrates to products. Once initiated, the hot combustion system converts free energy faster than does the cold combustion system.
ii. Consider a fluid in a pan heated from below. At a critical temperature difference between bottom and top, the system passes from conductive flow of heat to convective flow in Baynard cells. Here some of the energy in the system is used to maintain a non-equilibrium system organization that increases the rate of free energy dissipation compared to mere conductive heat transmission.
In both cases, as Craecraft notes, the system undergoes a transition such that free energy is passed through the system more rapidly, and some of that flow of energy is "housed" within the system to sustain an "organization" of processes that self-consistently enhances the flow of free energy. In both cases, he notes, the system "chooses" the pathway that dissipates free energy faster, but does so by sustaining some form of internal organization. Craecraft proposes that non-equilibrium systems will tend to "choose" pathways that maximize free energy loss, and pay a price in terms of "using" some of the energy to sustain internal organization. He suggests a measure of efficiency that is maximized as the ratio of energy held internally compared to the dissipation of that free energy. As such, his concept has similarities to the concern ecologists have with the residence time of energy held within the biosphere.
But in addition, both of his cases are informative because each exemplifies a clear kind of autocatalysis. The heat generated in the hot combustion case, for example in an open flame, sustains a local region of high temperature against the heat loss to a cooler environment, and the high temperature itself sustains the rapid generation of heat. In the Baynard cell, the convective flow upward in a cell induces a loss of heat at the upper surface, lateral movement by conservation of fluid volume, and subsequent sinking of the cooler fluid, thus sustaining the heat flow and the organization of the convective cells.
Another way of casting Craecraft's point in a wider context, noted already in the previous lecture, is that work is the constrained release of energy, and in general work must be done to construct those constraints. In the example of the flame, the exergonic process does work to heat the constrained local region which, in turn, sustains the flame.
I will return below to the core concept of "function" in an autonomous agent, which, in the familiar Darwinian framework, is simultaneously an account of the causal consequences of a part for the whole "agent" by virtue of which it has been selected, and therefore is also an ontological account of the fact that part exists in the Universe at all. The function of a part in the Darwinian account is, in general, a subset of its causal consequences. Thus, the task performed by a part is a subset of its causal consequences. The tasks are definable in terms of their role in the functioning of the whole cell or organism.
All of this matters because we seem to need to develop a new concept, presumably based on the concept of Autonomous Agents and Coevolutionarily Constructable communities of autonomous Agents, that goes beyond the familiar concepts of work and energy in physics. Familiar physics has work, energy, entropy, and information. But physics is lacking a set of concepts concerning the functional closure seen in an Autonomous Agent by virtue of which it is an organization of matter and energy such that it does work and carries out a set of tasks to sustain itself. And communities of such agents can and have proliferated functionally coherent and integrated ways of making livings with one another for 4 billion years in this corner of the Universe. Since 4 billion is a reasonable fraction of the estimated 14 billion years of the Universe, all this fancy proliferation, mere physics mind you, is pretty impressive. But there is no concept of functional closure in physics as we know it. Consequently there is no framework to discuss the propagating consequences of functional closure.
4.4) WHAT IS WORK? WHAT IS "PROPAGATABLE" WORK?
i. The standard definition of work is force acting through a distance. An example is a spring accelerating a frictionless mass.
ii. Atkins, in his book on the second law, comments on the cylinder and piston plus the working gas in the standard Carnot cycle. He points out that the boundary conditions created by the cylinder and piston allow the kinetic energy of the hot gas to be released in a coordinated way to achieve macroscopic work. He tries to define work as the coordinated release of energy. Then, he notes, the boundary conditions themselves participate in creating the work, and work is itself a "thing."
But, in general, as noted above, work is required to create the constraints themselves. Work begets constraints begets work. Either this confronts us with an infinite regress, or work and constraints self-consistently and jointly coemerge in the non-equilibrium Universe.
Note that Autonomous Agents do work to construct their own boundary conditions, hence create the ordered release of energy that constitutes work. More powerfully, not only do autonomous Agents evolve their own boundary conditions over evolutionary time, but during their life cycle, Autonomous Agents actually do real work to create their own boundary conditions. This work shows up, in contemporary cells, as synthesis of polysaccharide, lipid and protein molecules to form membranes, form structures and organelles such as mitochondria which provide the constraints on the release of energy that constitutes work. Thus, it takes thermodynamic work to construct enzymes that lower potential barriers along some reaction pathways and speed those reactions. And it takes work to construct membranes that allow proton gradients to form and do work to synthesize ATP, or to bury parts of organic molecules, thereby constraining their motion and increasing the potential barriers to some reaction pathways.
In short, Agents perform work to create the constraints on the release of energy which itself constitutes work. Work creating constraints is itself requisite for the occurrence of work. Work and constraints, therefore, must jointly and self consistently arise in Agents and be part of their ongoing coherent organization.
iii. Propagatable Work.
Let us consider two different examples.
a. In the first example, consider passing a rope over a pulley, tying one end of the rope to a heavy stone, pulling the other end of the rope, hoisting the stone to a ledge and storing it there. Then untie the stone, trip a lever allowing the stone to fall from the ledge to hit the ground, largely giving up its kinetic energy as heat.
b. In the second example, the initial set up is the same, but once the heavy stone is placed on the ledge, the other end of the rope is tied to a slightly lighter stone. Then the heavy stone is allowed to fall, raising the slightly lighter stone to the ledge where it is placed. In turn, the heavy stone is freed from its end of the rope, a still lighter stone is tied to the now free end. The stone on the ledge is allowed to fall, raising the still lighter stone. This procedure is carried out until frictional losses into heat of the initial potential energy of first heavy stone raised to the ledge arrests the series.
iv. What are the differences between these two examples? In the first, a single falling of the heavy stone is followed by some long term shifting of the atoms in the ground, and some generation of heat in vibratory motions which soon dissipate.
In the second example, a long series of macroscopic events occurs. The potential energy attained on first raising the heavy stone is released in a coordinated way that propagates a series of these macroscopic events. This propagation of a series of macroscopic events constitutes the propagation of work - the sequential linked constrained release of energy such that force acts through a distance, rather that the dissipation of the kinetic energy of the falling stone in the ground in the first example above.
v. But in addition, the second example takes a lot of actual work and coordination by an Agent. The Agent must recognize when to tie and untie stones, when to trip levers to allow stones to fall, etc. and actually carry out this actual work, thereby changing the constraints within the system. So the propagation of the series of macroscopic events requires work and some type of informed control to coordinate together with the high potential energy of the heavy stone raised to the ledge, enabling work to propagate.
vi. Why must the Carnot cycle be a "cycle"? Suppose it were not. Suppose that a cannon fired a ball which is to be used to propagate work. Perhaps the cannon ball hits a paddle wheel, turning it, thereby rotating a shaft which raises a bucket of water which is finally tilted by hitting the shaft, dumping the water into a sluiceway flowing downward to a second paddle wheel.... Even if energy were added at each step to overcome dissipation, notice that the coordination of odd bits of this and that to create the Rube Goldberg machine is highly improbable. Why a cycle in the Carnot cycle? The organization of the cycle, including gears and escapements to move the working gas into and out of contact with the two reservoirs, constitutes the simple way to achieve a system that can receive repeated energy inputs and perform sustained work.
vii. This leads me to suggest that "organization" is that arrangement of matter and flows which creates constraints that allow work - the constrained release of energy - to propagate.
viii. Notice that "organization," so defined, has nothing to do with the concept of entropy, which, in any case is an equilibrium concept.
4.4.1) Records as correlated, coherent macroscopic states usable to extract work, and Propagatable Records.
i. Try thinking of RECORDS as "coherent macroscopic correlations and constraints that can be used to extract work."
a. In the second example above with rope, pulley and series of ever lighter stones, the work done to raise the first stone creates a macroscopic condition, with correlations of mass points of the macroscopic stone located at a position high on the ledge.
This position of correlated masses is a COHERENT MACROSCOPIC STATE that can be USED BY AN AGENT, who must typically do work to create appropriate constraints, TO EXTRACT WORK, namely by tying the free end of the rope to another lighter stone, and allowing the heavy stone to fall from the ledge by tripping a release. But the tying of the free end of the rope to another lighter stone constitutes doing work which creates a constraint that allows work to propagate.
While the macroscopic state of the rock on the ledge is NOT our normal notion of Record, it IS A MACROSCOPIC STATE THAT CAN BE NOTICED AND USED BY AN AUTONOMOUS AGENT TO EXTRACT WORK. Thus, LET ME GENERALIZE THE CONCEPT OF RECORD TO INCLUDE CORRELATED MACROSCOPIC STATES OF THE WORLD - that is, states of matter and flow - THAT CAN BE NOTICED AND USED BY AN AGENT TO EXTRACT WORK.
Each successive position of the stone(s) constitutes not only work propagating, but also a State-Record that can be used by an Agent to propagate further work. For example, a paddle wheel can be placed in the path of the falling stone and thereby caused to rotate.
b. Consider a table with perfectly elastic borders backed by springs. Line up perfectly elastic pool balls with one border which has previously been compressed. Release the border, so the pool balls roll rapidly and in coordinated motion towards the opposite border of the table.
The coordinated positions and momenta of the pool balls is simultaneously a result of work done and a correlated state, or record, which can be used to extract further work. For example, collisions by the rolling balls with an object on the table may rotate the object in a prespecified direction. Think of the correlated motions of the balls as "high quality" energy that can be tapped to extract work.
(Note again, as with the Maxwell Demon in the non-equilibrium gas engine, an Autonomous Agent is typically needed TO USE THE RECORD TO EXTRACT THE WORK. But see "Natural Machines").
WORK THEREFORE CAN CREATE RECORDS IN THE "HIGH QUALITY CORRELATED STATES OF MOTION OF MASSES, OR OTHER CORRELATED STATES THAT ARE USABLE (BY AGENTS DOING WORK TO CREATE THE APPROPRIATE CONSTRAINTS) TO EXTRACT OR PROPAGATE FURTHER WORK.
Then, if Record is defined as above, Record appears to be at the same time part of "organization," the constrained arrangement of matter and flow that allows work to propagate.
If work is the constrained release of energy from a "record" - a macroscopic state of the world, and the constraints required to limit the release of energy at each point in the system typically require work to be performed for their creation and maintenance, then we need to consider ORGANIZATION as the union of matter, energy and constraint construction which allows work to propagate.
If all the work performed within the system were used to construct the very constraints that allow work to propagate, then the system could not do work on the external world. Thus, it seems useful to attempt to distinguish between work generating and timing the construction and "position" of constraints, and the work created by flow according to those constraints. In a sense, the union of these two is the "know-how" the Organization has - and for an Autonomous Agent, this is the know-how that, together with autocatalytic closure, constitutes making a living it its world.
Notice that as the system is displaced further from equilibrium, the number of possible arrangements of matter, small flows and large flows increases, so the "Adjacent Possible" space of organizations increases. Simultaneously, as displacement from equilibrium increases, the total stored energy increases. There are similarities to the Craecraft clue: "Systems choose organizations which dissipate free energy faster but house some of the energy internally in organization."
The fact that alternative options in the Adjacent Possible expand as the system displaces further into the non-equilibrium regime, suggests that Agents will find enhanced ways to make a living by ratchetting themselves and their communities collectively further from equilibrium.
Note the increased use of energy in technological society over the past 10 centuries, and the evolution of increased efficiency in trapping energy in the biosphere, from sulfur-iron to photo-synthesis, to use of oxygen for ATP production.
iv. Use of records by agents to extract work may fail:
Consider a standard pool table, modified to be certain that the lengths of the ends and sides were irrational with respect to one another.
Consider a "macroscopic Maxwell Demon" an Autonomous Agent, who wishes to place the 13 ball on the table in a position such that the moving 8 ball will knock it into the side pocket.
The Autonomous Agent, or Demon may fail:
* The Agent may fail to notice the positions and momenta of the other balls on the table.
* He may mismeasure their positions and momenta.
* He may misRECORD such positions and momenta.
* The RECORD may be degraded by "mutations" prior to its use.
* He may wait too long before using the record, so that chaotic motion of balls caroming off walls before the 13 ball is hit means the RECORD IS OUT OF DATE.
* The calculation about where to place the 13 ball may be chaotic due to accumulation of round off errors. The prediction may be incorrect.
* Finally, the Agent may suffer from TREMBLING HANDS when trying to place the 13 ball sufficiently accurately so it will be knocked by the 8 ball into the side pocket. (So the Agent must be able to ACT reliably!)
Notice that the Agent could place the RECORD at 0 degrees K before using it to predict where to place the 13 ball. Thus, the RECORD itself could be inscribed in a solid and settle to a state of no molecular motions.
At 0 degrees K Record is a correlated macroscopic state that carries no high quality energy.
Unless the Agent USES the RECORD to do something, including placing the 13 ball on the table to be hit by the 8 ball, NO WORK WILL BE EXTRACTED.
v. Propagatable RECORDS appear to require that the RECORDS BE UPDATED.
The updating appears to be capable of being performed by throwing away an earlier record and measuring a successor configuration of position and momenta of the balls.
Alternatively, the RECORD COULD BE UPDATED by Using the Record in a Timely Fashion, noting the consequences for the next configuration of the Table with the 13 ball added as a PROBE of the system and its predicted behavior, then using the measured behavior of the 13 BALL PROBE to try to correct and update the RECORD.
In short, a record can be propagated by being used (by an Agent) to help create a next state of the world which, when measured constitutes the next written" record.
(Here I am distinguishing between "written" records able to be placed at 0 degrees K and used by an Agent to extract work, and the high quality energy correlated state of the Balls on the Table itself which also constitutes a Record usable by an Agent to extract work.)
Note that the Autonomous Agent appears to be necessary to the propagation, and thus must be INCLUDED IN THE DESCRIPTION OF THE PROPAGATION OF WORK-CUM-RECORD
4.5) TOWARD THE UNION OF PROPAGATABLE WORK-CUM-CONSTRAINT - CUM - PROPAGATABLE RECORD: PROPAGATABLE FUNCTIONALLY WHOLE ORGANIZATION.
i. I seek here a new concept. I think I can formulate the concept in a rough way.
* Consider a colony of bacteria as Autonomous Agents undergoing growth by cell proliferation.
* Each bacterial cell CONSTITUTES THE EMBODIED KNOWLEDGE - "KNOW-HOW" - TO MAKE A LIVING IN ITS WORLD.. THIS EMBODIED KNOWLEDGE CONSTITUTES "THE RECORD."
* The know how is embodied, in part, in the structure and dynamics of the cell such that flow of energy at each point is constrained. The cell does work to construct and control its boundary conditions. The cell uses work to constrain the very flow of energy which constitutes carrying out the work and performing the tasks whereby it makes a living. To do so, much of the control work must not take large energy loss; the signal amplification must be high.
* THE VERY PROLIFERATION OF THE CELLS CONSTITUTES PROPAGATABLE WORK -- LITERALLY VIA LINKING OF EXERGONIC AND ENDERGONIC REACTIONS.
* THE VERY PROPAGATION OF WORK IS THE CONSTRAINED RELEASE OF ENERGY, BUT THE CONSTRUCTION OF THE CONSTRAINTS BY THE AGENT REQUIRES WORK. WORK BEGETS WORK. THE WORK TO CREATE THE CONSTRAINTS ALLOWS THE CONSTRAINED RELEASE OF ENERGY WHICH CARRIES OUT A COHERENT SET OF TASKS WHEREBY THE AGENT REPRODUCES, DOES WORK ON ITS OWN BEHALF INTERNALLY AND WITH RESPECT TO MAKING A LIVING IN ITS ENVIRONMENT, AND PROPAGATES. THE COHERENCE OF CONSTRAINT, WORK, TASK AND SELF REPRODUCTION CONSTITUTES THE AUTONOMOUS ORGANIZATION OF THE AGENT.
* THE PROCESSES OF MUTATION AND SELECTION OPERATING IN THE COLONY CONSTITUTES THE VERY UPDATING OF THE RECORD. THIS UPDATING RENDERS THE RECORD PROPAGATABLE.
* The process of mutation and selection implies a cost of search in the Adjacent Possible. Many mutations fail. The agents die with no progeny. This mutation load appears to be related to the costs for erasure of the Record for the Maxwell Demon. The Demon must erase to record anew. The growing microbial community undergoes death among its members for novel superior functionalities to be discovered.
ii. Thus, the very proliferation, mutation and selection of the colony of bacterial Autonomous Agents CONSTITUTES the union of Propagatable Work-cum Constraint-cum-Record via the Autonomous Agent.
iii. Among the properties present in the community of autonomous agents is that collectively, the community achieves functional closure such that the tasks that must be done and the flow of matter and energy constituting the work to carry out the tasks whereby members of the community, on average are sustained or "amplified" in abundance, is self-consistently achieved.
Tasks, matter, and energy are jointly self-sustaining, self-defining, and whole. This "whole" is not mystical, it is given by the conditions defining an Agent, and more broadly, a collection of Agents and Niches or Natural Games.
It is precisely this property of functional closure of propagating constraints, work, and self construction that is missing in an account based merely in terms of atoms, energy, and even entropically unexpected patterns of atoms.
iv. In short, in the present case propagatable work-cum-constraint-cum-record in the bacterial colony IS A UNION OF MATTER-ENERGY-INFORMATION in a propagating functionally coherent and self sustaining ORGANIZATION: The Coevolving Community of Autonomous Agents. But the concept of information used here is not the entropy-Shannon information concept. It is the embodied causal functional linkages whereby the agent makes its living in its world.
Autonomous Agents appear to constitute sufficient conditions for such propagating organization of matter-energy-information. They may or may not be necessary.
4.5.1) The Spontaneous Formation of Collectively Closed Sets of Tasks and Propagating Work in Autonomous Agents.
i. In the theory of the emergence of autocatalytic sets, I noted that as the diversity of molecular species increases, the diversity of reactions by which they transform to one another increases even more rapidly. This makes the emergence of at least one collectively autocatalytic set easier as diversity increases.
ii. The function of a part is one or a subset of the causal consequences of a part. Each macroscopic state of nature, a low energy structure, or a flow of some kind, has manifold causal consequences. (The cannonball whistling in flight for example). These causal consequences can be functional in the proper context.
iii. As the diversity of objects and flows, constructed by bits and pieces of work - the constrained release of energy, increases, the ratio of causal consequences to parts increases. In general, as this diversity increases, the chances of there existing at least one functionally closed set of causal consequences by which work propagates and the system is also able to construct its parts and the constraints among those parts, hence itself, increases. Thus, this diversity enhances the chance of an autonomous agent performing propagating work cycles. This should be formalizable in concrete models in parallel to models of the emergence of collectively autocatalytic systems.
iv. "Glitches" in such systems typically partially degrade propagating work. Such glitches can be "fixed" to provide control of the amount and timing of work done in the propagating work - constraint - record that constitutes an Agent.
v. Displacement from equilibrium enhances the variety of alternative ways matter and energy flows can be arranged, hence enhances the ways causal consequences can propagate, hence enhances the set of "tasks" that can be performed.
vi. Coevolving autonomous Agents propagating work and maintaining a displacement from equilibrium, advancing into the adjacent possible, build up connected webs of linked work, constraint construction, signaling, function, complexity.
4.5.2) Preliminary remarks on quantitative account of propagating work. Consider a node in a system with 100 potential pathways down which energy could flow. Represent the total "potential energy" as 1.0, and the total flux as 1.0. Using the non-equilibrium thermodynamics formula that work is the force times the conjugate flux, if all the force and flux pass down only one pathway, the total work is 1.0 x 1.0 = 1.0. If, instead, the forces and fluxes are partitioned into 100 equal parts, the total work down all 100 pathways is 100 x .01 x. 01 = .01. Thus, due to the quadratic nature of force x flux, partitioning both along branched pathways lowers total work done.
If one takes the Darwinian view that we will see the winners among coevolving autonomous agents in a coevolving ecosystem, and if total work must be divided between signaling, constraint construction, work done on the external world, and total work including all of the above, then those auto-organizations which best allocate resources to reproduce themselves in their environment will win. In the next lecture I present a linked set of three hypotheses suggesting that autonomous agents will coevolve to the edge of chaos on a dynamical order-chaos axis, to a self-organized coevolutionary critical state, and to the subcritical-supracritical boundary with respect to advance into the adjacent possible at the level of molecules and novel functionalities. If that is roughly correct, it may, in due course, help characterize the allocation of work into signaling, constraint construction, work on the world, etc. that optimizes the capacity to coevolve successfully. If the answer is unknown, at least this presents a tentative framework to state the questions. Indeed, the issues of units and levels of selection will arise in Lecture 6, for the entire biosphere is coevolving. When one Agent explores the adjacent possible and some die and others live, the gains in organizational know-how and losses due to the "mutational load" induced by this exploration bear not only on each agent, but, on the whole propagating linked ecosystem of agents. In a way that requires much deeper elucidation, one guesses that well tuned coevolving ecosystems with power-law distributions of speciation and extinction events, optimize the average secular growth of the adjacent possible.
4.6) NON-AGENTS MAY HAVE LIMITED CAPACITY TO PROPAGATE WORK, PROPAGATE RECORD, OR PROPAGATE THE UNION OF WORK-CUM-RECORD.
i. Consider a Carnot engine. Imagine we found one floating in intergalactic space. Suppose "nobody" made it. Then such a Carnot engine, a machine, might be called a Natural Machine. Such a machine could couple exergonic and endergonic processes, hence could propagate work. The record is in its structure.
ii. But Natural Machines, unable to reproduce, evolve, or coevolve as Agents can, are very unlikely to:
a. Maintain themselves against degradation and rusting failure.
b. Are unlikely to assemble highly complex linkings between exergonic and endergonic processes propagating work-cum-record, as in an evolved tangled bank of an ecosystem.
4.7) AUTONOMOUS AGENTS, CAPABLE OF COEVOLVING WITH ONE ANOTHER AND GRAFTING NOVEL METABOLIC AND OTHER FUNCTIONS INTO THEIR ONGOING HISTORY TO CREATE ECOSYSTEMS AND ECONOSYSTEMS, TEND TO ASSEMBLE VASTLY COMPLEX AND IMPROBABLE INTERWOVEN LINKAGES OF EXERGONIC AND ENDERGONIC PROCESSES ABLE TO PERFORM VERY COMPLEX INTEGRATED WORK CYCLES FOR THE BENEFIT OF EACH AGENT, AND BY INVISIBLE HANDS, FOR THE BENEFIT OF THE SOCIETY OF AGENTS.
i. Thus, Autonomous Agents coevolving with one another may be the most reliable source of sustained propagatable work-cum-record in the Universe.
ii. In propagating, such Agents INVADE the Adjacent Possible at the molecular and other levels.
4.8) AGENTS, SEMANTICS, AND EMBODIED KNOW-HOW.
i. Darwin tells us the meaning of inputs to an Agent: Food or poison.
Meaning is for "my" reproductive success. Darwin tells us we will see the winners of natural selection making a living in their worlds. Hence, from the perspective of an Agent, information has meaning and requires practical action in order to survive and proliferate.
* Given a physical definition of an Autonomous Agent, semantics acquires a physical meaning. Pierce's triad of signs is present once there are Agents.
* Record as embodied knowledge "model" of Agent's world. AGENT has capacity to sense a signal, record, and mobilize work to carry out a work cycle in response for the benefit of the Autonomous Agent.
* Semantic information is distributed among coevolving Agents that mutually create a coevolutionarily constructable "ecosystem."
* Implicit prediction: Practical action requires predictions: Food over there.
* Implicit prediction of "intent" of other Agents once Agents coevolve with one another.
4.7.1) PHYSICS OF SEMANTICS AND MEANING.
i. Carnot defines "machine" and "entropy." Boltzman interprets entropy in terms of micro states and macrostates plus the ergodic hypothesis and flow from micro states corresponding to rare macrostates - having only a few micro states - to macrostates corresponding to vastly many micro states, hence common = equilibrium partition of degrees of freedom. Boltzman, in turn, gives birth to "information" via Shannon.
ii. Information, however, remains purely syntactic - an unexpected pattern.
iii. Definition of an Autonomous Agent as Closure on a set of Devices, Work cycles, and closure on a set of Tasks, given inputs, outputs and devices is a new notion.
The new notion creates physical meaning of a functional whole: An Autonomous Agent able to act on its own behalf. EMBODIED PROPAGATING ORGANIZATION OF WORK-CUM-RECORD-CUM KNOW-HOW.
iv. Given this new notion, the very physics of coevolving Agents implies the semantics of the requisite practical action of each agent it its world. Darwin and Adam Smith tell us we will see the winners. Selection and Invisible Hands are just a form of stability analysis of such unfolding non-equilibrium systems when kinetics dominates the outcome.
v. The example of the microbial community points to the following: Darwin tells us we will see the winners. The winners in such a community are the duality of Agents and Niches jointly having come into existence as kinds of things in the Universe, and jointly propagating via the collective organization of the community embedded in the larger environment. The propagating organization is not merely processing information in the contemporary sense of information theory as if it were a pattern of bits on a universal computer. The community is physically coming into being, and proliferating, making more of itself, or sustaining itself via "know-how" patterned-matter-energy flow through it. The work cycles are the "grunting" of honest construction work.
Information alone, free of semantics, free of agents, is an inadequate concept. So too is mere work. A coevolving community of non-equilibrium Maxwell Demons uniting matter-energy-know-how drives the assembly of diversifying propagating organization. Since Agency is a new concept with physical meaning, we must seek to formalize this new concept capturing propagating organization as functionally coherent matter-work-know-how.
But more is hinted: The coevolving microbial communities in the biosphere exhibit the Universe propagating and building up complex functionally coherent organization. The autocatalytic emergence of an increasing DIVERSITY of niches and agents - in the biosphere or "econosphere," as noted in the previous lecture, raises the possibility that the cocreation of diversifying Agents and Natural Games might possibly follow a maximizing principle. In the sixth lecture I return to ask whether such propagating functional organizations of work-record-know-how might tend on average to maximize the sustained growth of diversity.
In the next lecture I sketch a candidate attractor for coevolving communities of Autonomous Agents.