Autoteleology is a signature of life; living things are virtually always autoteleological,[1] while nonliving things very seldom are.[2] Darwinian theory offers a good explanation for this prevalence: effectual autoteleology will always be rewarded. But in truth Darwinian evolution owes more to autoteleology than vice versa. Evolution, and life itself, depends on autoteleology in its simplest and purest form — the drive to live. Without that, or without the intrinsically autoteleological goal of reproduction, evolution as described by Darwin could not occur. This point is one of three significant claims I make concerning autoteleology: autoteleology is more fundamental than evolution, and necessarily precedes it.

The second claim is that autoteleology is physical but universal — any physical system, regardless of the material that constitutes it or the forces that drive it, can be autoteleological, if it is so arranged that it tends to further its own existence.

The third claim is that autoteleology is antientropic — it counters the statistical probabilities that lead to entropy increase, diminishing that increase or even (locally at least) reversing it. The third claim is perhaps the boldest, so I will elaborate.

To begin with, note that the second law of thermodynamics, which states that entropy always increases, is not a fundamental physical law along the lines of Einstein’s laws of relativity or Newton’s laws of motion. Rather, it’s a statistical law, like Boltzmann’s law relating temperature and pressure in gases: if follows from the aggregate properties of large collections of randomly interacting molecules. If the interactions cease to be random, the law may lose its rationale.

Consider Maxwell’s demon. Maxwell’s demon is a mechanism imagined by James Clark Maxwell around 1870 that coaxes a non-random outcome from the random interaction of molecules through a simple but clever control system: a door between two containers of gas that the demon slides open only to allow fast molecules to pass in one direction slow molecules in the other. While the consensus among physicists is that Maxwell’s demon ultimately cannot defy the second law of thermodynamics (on one widely-held account, because of the unavoidable entropy cost of the information processing the demon must do), there is also a consensus that it might avoid the second law within certain limits of space and time — the piper must ultimately be paid, but later and somewhere else, not here and now.

Efficacious autoteleology, like Maxwell’s demon, seeks a local outcome that defies the second law of thermodynamics. It seeks to increase the probability of the system’s existence by means of some combination of persistence, growth and reproduction, each of which represents a maintainance or increase in local order. Moreover, this is accomplished by non-random behavior, non-random because it is driven by knowledge, whether that knowledge be engineered, evolved, acquired or computed.

Autoteleology raises many questions, including deep ontological questions. This should not be surprising, because autoteleology is among other things a form of self reference, and self reference is the source of some of the oldest and thorniest paradoxes in philosophy.

Most such paradoxes, however, are only logically paradoxical, while being physically unproblematic. There are no physical dillemas posed by a fellow who says, “Everything I say is a lie,” only logical ones. Autoteleology is different. Its implications are quite physical.

—–
Notes:

[1] The two exceptions being (1) when altruistic behavior provides significant genomic advantages (e.g. survival of genetically close relatives) and (2) when organisms capable of choosing their own goals (human beings) choose goals that are not autoteleological (such as the goal of defending one’s country). Other cases that might appear at first glance to be exceptions are not, when fully analyzed: symbiosis, for example, is a case of organisms supporting other organism’s continued existence as a means of pursuing their own continued existence.

[2] The only known exceptions being certain systems engineered by humans, such as satellites that maintain orbital stability by firing rockets when necessary.

Yet the notion that the dynamics of the inanimate world are simple and predictable while the dynamics of life are messy, complex and chaotic is simply false. For one thing, the inanimate world is also messy, complex and chaotic — consider the weather. For another, the fact that simple principles are indeed useful in understanding and predicting many inanimate behaviors, such as using Newton’s equations to calculate the trajectory of a projectile, does not rule out the possibility that life itself follows its own simple principles. My current proposal is exactly this.

The simple principle I have in mind is one which describes the behavior of cybernetic systems, which I formulate as follows:

Φ(r) & α(q) & ε(p) & ε(m) ⇒ q

where

• Φ(r) means that r is a goal, i.e., a desired state of the world;
• α(q) means that q is a choosable action;
• ε(p) means that the state of the world is known to be p; and
• ε(m) is short for ε(p & q ⇒ r), which means that it is known that when p, doing q leads to r.

In other words, if a system has a goal, and it knows what action will achieve its goal, and it has both the opportunity and the capacity to take the action, then the system will take the action.

To be sure, this is a simplification. For example, it doesn’t describe how a system might balance competing goals that call for contradictory actions. Nor does it discriminate between broad, long term, high level goals/actions and all the smaller, more immediate goals and actions they decompose into. But it’s straightforward to accommodate these and other complications by extending and elaborating on the formula without weakening the underlying principle.

The four terms in the formula represent the components that a system must have to be cybernetically successful, that is, to respond advantageously to opportunities in the environment and thereby achieve its goals.

The first term, Φ(r), is teleological; it defines the system’s purpose or goal. Generally this is a given, though some cybernetic systems have the ability to choose some goals for themselves.

The second and third terms, ε(p) and α(q), describe the system’s interaction with the world. They are functions of what the system is physically capable of sensing and doing. As such, they may vary widely from system to system, but the principle is always the same: to achieve its purpose, a cybernetic system must have the ability to obtain information about the state of the world, and the ability to perform useful actions.

The final term, ε(m), spells out the critical knowledge that is required for the system to achieve its goal. How this knowledge is acquired varies greatly; in my analysis, the nature of this knowledge acquisition is the primary characteristic that sets apart the main classes of cybernetic systems.

Let’s see how this applies in the case of a very simple cybernetic system: a furnace connected to a thermostat. We can describe the behavior of this system in two simple sentences:

• If the temperature is below the target level then if the furnace is off, turn the furnace on.
• If the temperature is above the target level then if the furnace is on, turn the furnace off.

Three of the four terms of the cybernetic formula are readily identifiable in these sentences:

• Φ(r), which means that r is the desired state, where r = “temperature is at the target level”
• ε(p), which means that p is known, where p has a different meaning in each of the two sentences:
  • p1 = “the temperature is below the target level”
  • p2 = “the temperature is above the target level”
• α(q) means that q is a choosable action, where q also has a different meaning in each sentence:
  • q1 = “turn the furnace on”; α(q1) is true if the furnace is off.
  • q2 = “turn the furnace off”; α(q2) is true if the furnace is on.

But what of the remaining term, ε(m)? This term refers to knowledge of the causal relationship between the action and the goal. In an engineered system such as a heating system, this knowledge is first and foremost in the mind of the engineer. From there it translates into interconnections between sensors and effectuators — ε(p) and (q if α(q)) — that the engineer builds into the system. In our heating system, the knowledge is embodied in the wiring between the thermostat and the furnace.

There exists another class of cybernetic systems where ε(m) comes to be embodied in the system without the assistance of an engineer. This is the class of systems whose designs are the product of Darwinian evolution, a category which includes all living organisms. Darwinian evolution emerges from the interplay of three processes: reproduction, mutation and selection. Selection operates on individual organisms, but reproduction and mutation are different: they depend on the existence of encoded information that persists beyond the individual and in fact accumulates and improves over generations.

In living organisms, this hard-won knowledge is encoded in nucleotide chains, that is, DNA. ε(m) is not explicitly encoded in the organism’s DNA. Rather, just as ε(m) in a heating system is embodied in the particular way the engineer connects the wires between the thermostat and the furnace, in a living creature ε(m) is embodied in the (primarily) biochemical interplay among the mechanisms that comprise it. The way these interactions are distributed and regulated is encoded in the organism’s DNA.

A third class of cybernetic systems is the class of systems capable of learning, which includes most animals to one degree or another. In such systems, a mechanism exists allowing the system to acquire and embody new knowledge, in addition to the knowledge it inherits. Such a system can potentially be much more adaptable than one whose behavior is hardcoded, whether by evolution or engineering.

Learning can take many forms. Essentially the task is to find actions (q) that succeed in a given circumstance (p). One way is to randomly generate candidates for q (perhaps by stringing together smaller, hardcoded actions), test them, and keep the ones that work. Another way is to start with q, make incremental changes, and reinforce or reject each change according to the result. Yet another way is to try a single behavior in different situations to find a beneficial combination — in other words, keep q fixed and vary p. In none of these cases does ε(m) need to be fully and explicitly represented, but at least some part of it (p or q) must be exposed in some fashion that translates into the ability to modify ε(m).

The fourth and final class of cybernetic systems are those capable of creating new ε(m) through an internal logical process of some kind, for example a mind. This class includes humans, possibly a few other species, and arguably some experimental computer systems. Because ε(m) in such a system is explicitly arrived at, it is explicitly represented. (This presumes a broad reading of the term “explicit”. For example, it includes algorithmic representations.)

The ability to synthesize ε(m) through internal processes is an enormous advantage. In contrast to acquiring information across generations, or over a lifetime of trial and error, a system with the ability to synthesize knowledge can generate new ε(m) on demand. In the case of humans, inference is so woven into our perception that we seldom acknowledge it. I walk into a home I’ve never been in before, and in the distance through a doorway I see just a sliver of a tall white object with two handles — a refrigerator. From a smidgen of evidence combined with knowledge I already possess I can now infer huge amounts of new knowledge. For example, if I get thirsty later in the evening and the host invites me to help myself to a beer, I instantly infer exactly where to look for one.

To sum up the different ways cybernetic systems get their ε(m), that is, usable knowledge of causal connections:

• burned — ε(m) is engineered into the system
• earned — ε(m) is acquired through evolution by natural selection
• learned — ε(m) is acquired through trial and error learning
• discerned — ε(m) is synthesized by an internal logical process

My contention is that the above processes, which are very different in the details of their operation, are all the same in terms of the principle that drives them and the nature of what they produce. For a single principle to underlie such varied phenomena — life, mind and engineering — it must be fundamental. And so it is: the physics of knowledge.

There has been some success, I think, in justifying that our cumulative sense data are true on the whole. Coherence theories, reliabilism and functionalism all provide reasons to believe our sense data in the preponderance of situations. But justifying a particular belief on the basis of specific sense data is difficult without granting some sort of special justificatory status to direct observations. The special status comes in various forms: “basic beliefs”, “immediate apprehensions”, and evolutionary, functional or causal explanations. A fair amount of hand waving often seems to go on at this point.

My view is that justification enters the picture a different way. A certain necessary component has been missing from most analysis, a critical ingredient in taking the step from sense data to knowledge about the world: axioms that allow sense data to be organized.

Why is it necessary for sense data to be organized before they can become knowledge? To lead to knowledge, sense data have to be about the world. Photons are about the matter they interact with on the way to the eye; sound waves are about the vibrating matter that gave birth to them. But neither photons nor sound waves come to us labeled. Our minds, and the neurology that supports it, must determine not just the content of sense data, but their provenance. To understand what we are seeing (the sense data) we must know what we are looking at. The second “what” question — what is the object we are perceiving — gets answered instantaneously; by the time we are conscious of a perception, it is already a perception of something. The sense data have been sorted out.

For example, if you hear the sound of a dog barking behind you, you instantly assume the existence of an object (the dog) that the sound is emanating from. As more data come in, you may alter your conception of the object and perhaps even change your mind entirely — maybe you search in vain for any evidence of a dog and decide that your imagination was playing tricks on you, for example. But the natural, default starting point for thinking about perceptions is to assume they are perceptions of objects, and to infer that the objects exist. And for this, it is necessary to accept a priori that objects exist, and that the world is made of them.

So, we begin with an axiom:

• The Axiom of Objects: The world is made of objects.

I suppose this would be a good point to attempt definitions of “world” and “object”. These are not shallow concepts. But for the purposes of this discussion, I’m happy with the common meanings of “world” as everything that materially exists and “object” as a particular thing that materially exists. (For a materialist like me, “materially” is redundant, but I include it to make it clear that here the world means the real world and an object means a real object.) Under these definitions, “the world is made of objects” may at first glance seem to be a tautology: “all that exists is made of things that exist”. But it is actually a meaningful statement. It says that existence is decomposable. This decomposability allows us to slice and dice existence into units that are useful for us, and to do so a priori — before we reason.

But is this statement really an axiom rather than an inference? Can’t we infer the existence of objects from our observations? Perhaps we don’t notice this inference because it happens unconsciously and very quickly. Or perhaps the inference happened at some point in the past, and now we simply recognize perceptions as further instances of the sort which support the inference.

There is a difficulty with this claim, however. To prove the existence of objects, one must prove that at least one object actually exists. To prove this from sense data means that some aspect of some (or all) sense data logically implies the existence of at least one object. This is a stronger claim than saying that the sense data suggest the possibility of objects, or are compatible with the existence of objects. It’s saying that the existence of an object is logically mandated by the sense data.

Yet it seems possible to challenge the necessity of any given object. Consider an apple sitting on a table. You may believe it is possible to prove the existence of the apple based on its color and shape: something which is a particular shade of red and a particular size and shape is very likely to be an apple. There it is, an apple. It’s an object and it exists.

But no, I might say. It’s not an apple. It’s actually two objects, an apple-top and an apple-bottom, fixed together firmly and seamlessly. There would seem to be no fact that you can know about your apple that I can’t also know, when suitably translated to my two-object model, about my apple-top and apple-bottom taken together. You may give me a number of reasons why your model is more useful than mine, or more natural, or more elegant. But unless you can identify something that you can know that I cannot, then you have no grounds to say that my model is incorrect or incomplete, or that your model is necessary.

A potential counterargument would be to point out that each of us sees objects, just not the same ones. Perhaps no single object is necessary, but objects of some sort are necessary.

But even this weaker claim does not hold up. Consider an alternative approach to understanding the world, one without objects. Imagine that you see the world not as objects, but as frequency distributions of various properties in space. For example, instead of perceiving red objects, you perceive patterns of redness. Frequency distributions of various properties considered in combination provide the basis for defining more complex properties and their frequency distributions. So combining the distribution of redness with the distribution of just the right other properties would give you the distribution of appleness, which would happen to coincide exactly with a map of every apple in the world, including the one in front of you.

Are there any facts you can know about the world if you think in terms of objects, that you cannot in principle know if you think in terms of frequency distributions, assuming you translate the facts appropriately from one model to the other? To me the answer appears to be no.

The Axiom of Objects gives us license to assume the existence of objects. However, we don’t perceive objects directly; we perceive properties. It is conceptually possible, as we saw above, to think of properties as free floating, existing as properties but not as properties of anything. Indeed, if we accept that objects are not logically necessary, and we accept the existence of properties (because we perceive them), then we cannot logically conclude that properties are necessarily properties of objects.

But that is in fact how humans think of properties: as belonging to objects. We take it as given. This suggests that another axiom is at work:

• The Axiom of Properties: Objects have properties.

Together, the Axiom of Objects and the Axiom of Properties allow us to take the leap from perception to knowledge. It’s not the source of justification, it’s just the glue that connects the source of justification to our system of beliefs.

Here is the cybernetic model, in a nutshell:

    Φ(r) & α(q) & ε(m) & ε(p) ⇒ q

where

• Φ(r) means that r is a desirable state;
• α(q) means that q is a choosable action;
• ε(m) means that q is known to cause r when p; and
• ε(p) means that the state of the world is known to be p.

I propose that this model, with a little elaboration, allows conscious behavior to be explained in physical terms in a way that doesn’t make consciousness redundant.

Consider the physics of a traffic light. The light turns green, and cars move. The light turns red, and cars stop. Clearly there is a cause and effect relationship between the color of the light and the motion of the cars. But most things in the universe don’t move when you shine a green light on them and stop when you shine a red light on them. Furthermore, there is no known simple law of physics, or complex law for that matter, that relates colored light and motion of bodies. Yet unless we admit to dualism the causal chain of events in a traffic intersection is a chain of physical events following strict physical laws. What’s the deal here?

Well, a lot of it at either end is straightforward. The physics of the light itself are pretty well understood. Light from either an LED or incandescent light source passes through a colored lens, across space and into a driver’s eye, exciting color-sensitive cone cells in the retina, which send signals via the optic nerve to the driver’s brain. Working from the other end, the change in motion is caused by changes in the flow of fuel into the engine and the application of brakes, which are caused by the pressing of the accelerator and brake pedals, which is caused by the movement of the driver’s legs, which is caused by nerve impulses originating in the driver’s brain.

In between, though, the causal chain leads into and out of the driver’s brain. Once in the brain, the simplest explanation seems to mental, not physical: the driver sees the light; if the driver perceives the light to be green, the driver decides to press the gas peddle; if the driver perceives the light to be red, the driver decides to press the brakes. Then the brain emits a signal and the chain goes physical again.

Interestingly, the brain is not actually a necessary part of the traffic light system. It would be quite possible to replace the mental parts of the causal chain with purely physical components. For example, you could attach a video camera to the hood of a car, connect it to a computer that continuously analyzes the images from the camera to see if it contains a red or green patch of the size, shape, position and context consistent with a traffic light, and connect the computer to servomechanisms which depress the accelerator or brake pedal as appropriate. Not a trivial exercise by any means, and I’d be leery of riding in such a car, but it’s certainly within the realm of possibility.

And if you find the software in such a system to be a little too brain like, you can do away with it: set up a bunch of red and green light sensors, focused in the expected direction of where a traffic light would be, and run wires through carefully adjusted amplifiers and high-pass filters to switches operating the servomechanisms. I definitely wouldn’t ride in that car, but it’s imaginable that with enough cameras and enough fine tuning it might do the right thing more often than not, which would be enough to prove the point — the point being that anomalous causal chains do not have to be mental.

But is this chain really physical? It’s certainly not natural. One could argue that it is in fact mental, because such a setup of cameras and lights would never arise naturally, and could only possibly exist by virtue of an agent desiring to set up such a system, thinking of how it should work and doing all things needed to construct it. In other words, if the causal chain itself is not mental, the cause of the causal chain is.

However, this is not true of all such causal chains. It might be true of traffic lights, but it is certainly not true of the countless biochemical signaling systems which abound in living creatures of all kinds, with and (presumably) without minds, and which behave just as anomalously as traffic lights.

It does seem to be true that persistent anomalous causal chains seem to extend only as far as living things, and things made by living things. Anomalies in the non-living world – weather anomalies, for example, such as a warm day in January where I live, or a cold day in August – do occur but are invariably ephemeral; the causal chains appear out of nowhere and disappear quickly. Perhaps being able to initiate anomalous causal chains that persist is the defining characteristic of being alive.

By “knowledge of the future” I don’t mean by powers of prognostication or parapsychology; I mean knowledge that is earned (via evolution) or learned. Nor do I mean knowledge of the exact future that will happen; I mean knowledge of various futures that may happen, depending on one’s choices. Nor, finally, need this knowledge be perfect; just somewhat better than random. The point is that if one can make choices, and can make better-than-random predictions about the possible outcomes of the choices, then one is in a position to extract value from something that in conventional terms doesn’t physically exist — the difference between what is and what might have been.

Teleosemantics is a theory of representation that says that mental representations reference the external world via their purpose, not their cause — that is, a mental representation is the way it is not because the thing it represents caused it to be that way, but because that’s the best way for it to achieve the function it was evolved to do. Teleosemantics was formulated as an alternative to causal theories of representation, which hold that a causal chain of some sort connects things in the real world to their mental representations. Imagine you see a snake and you think “snake”: a causal theory asserts that “snake” (i.e., the representation of a snake in your mind) means snake because it’s caused by an actual snake. A teleosemantic theory, however, asserts that “snake” means snake because you have evolved to react to snakes, and a mental representation of a snake such as the one in your head when you think “snake” is useful as a part of the reaction mechanism.

The notable advantage of teleosemantics over a causal theory is that it explains how you can see a rubber snake and think “snake”, without “snake” having to mean rubber snakes as well as real snakes. After all, if the meaning of any mental representation is the thing that caused it, then pretty much any mental representation means just about anything, because humans are very good at mistaking things for other things. Teleosemantics doesn’t have this problem, because if meaning derives from purpose, and the purpose of a mental image of a snake is to avoid real snakes, then it’s still about real snakes even when prompted by the observation of a rubber snake.

But teleosemantics has its own problems. For one thing, it seems too narrow: how could such a theory account for all the things we can think about that don’t have an apparent purpose? For another thing, it seems too tied to biology and evolution: teleosemantics says that mental representations are determined by biological needs and evolutionary history, and are only indirectly connected to the things represented, whereas intuitively it seems that the connection must be much more direct.

I agree with the critics that there must be a deeper connection between mental representations and their referents, something mediated perhaps by biological mechanisms but not fundamentally defined by them. But I agree with the teleosemanticists that the connection is by way of function. My solution is essentially to recast the teleosemantic explanation in cybernetic terms and transport it to the level of basic physics and logic, thereby becoming fundamental and universal (to a physicalist anyway).

To see how this works, let’s revisit the snake example. What is in your head when you see a snake, and what connection does that thing in your head have with the actual snake?

The teleosemanticists say, for starters, that what happens in your head is an evolved capability. I fully agree. Without doubt we are what we are because of evolution, including our mental faculties. If you see a snake and move carefully away, it makes perfect sense to say you are exhibiting behavior that was favored by evolution, and that what happens in your head has something to do with that evolved capability. But the fact that capabilities are evolved doesn’t tell us what they are and how they work. Evolution can tell you why birds have wings, but to know how those wings work you need to know aerodynamics. Likewise, it’s certainly true that you are able to evade the snake because of the abilities afforded you by evolution. That’s why you can evade the snake. What I would like to consider is how you can evade the snake. How is this behavior implemented? This is not a question evolution can answer.

Avoiding snakes is a directed cybernetic process. A cybernetic process is one that operates on feedback: the state of the system at time t + 1 is a function of the state of the system at time t and the state of the world, as sensed by the system. A directed cybernetic process is one that functions towards a goal, defined as a particular state of the world at time t + 1. In our example, the goal is the state of not being imminently threatened by snakes.

But we are not talking about hard-wired snake avoidance, we are talking about seeing something, thinking it might be a snake, and deciding to avoid it — in other words, not just avoiding snakes, but consciously avoiding snakes. Consciously avoiding snakes is a self-directed cybernetic process.

This is not to say that self-directed cybernetic processes are always conscious — but they always involve something similar to mental states: information about the state of the world, for one thing, and goals, for another. They are functionally similar, to be precise, which does not require being phenomenologically similar.

We can describe a directed cybernetic process functionally. The physical embodiment of the cybernetic process is a cybernetic system, conventionally referred to as an agent. Let p be a particular state of the world. Let q be a particular action by the agent. Let r be the state of the world that results. So, we can say: if p is the case, and an agent does q, the result will be r. We can also express this as a logical proposition:

p & q ⇒ r

Let m stand for the above proposition:

m = (p & q ⇒ r)

A directed cybernetic system works toward a state of the world. Let the function Φ(r) mean “works toward r”, or, in other words, the agent has chosen r to be a goal. Let the funtion α(q) mean that doing q is an available choice to the agent, that is, q is actionable. Let ε(p) mean that the agent knows p to be the state of the world., or one might say, the agent knows p. Let ε(m) mean that the agent knows m, that is, knows that if p is the state of the world, then doing q will result in r. Then the behavior of a directed cybernetic system may be described as follows:

Φ(r) & ε(m) & ε(p) & α(q) ⇒ q

Or, restated in plain English: if r is a desirable goal, and an agent knows that doing q when the state of the world is p results in r, and the agent believes the state of the world to be p, and q is an option available to the agent, then the agent will do q.

To be sure, this is a simplification. For one thing, it’s just a snapshot; in reality, the state of the world, the available actions and even the goals change over time. For another thing, the state of the world is too complex to consider in its entirety; in practical terms, only a relatively small number of properties of the world can be considered at once. Finally, there are many goals and actions in play at any given moment, and they are not guaranteed to be consistent or compatible.

But such complications don’t negate the fundamental principles expressed in this simplified formulation. One such principle is that what the system needs to know is not the same as what the system’s goal is. Or, in terms of the formula, to achieve r, the system needs to know p and m. This explains the apparent disconnection between a functional explanation and the intuition that a representation is related to what it represents: in a cybernetic system, while the function of a representation does determine content, the content is not of the function, but of the state of the world that causes or enables the function. I call this a reverse-causal theory of content: the mental chain of causation, from function to representation, is the reverse of the physical chain of causation, from content of representation to function.

So the cybernetic model explains why the cause of our mental representations can be so different from their content. As stated so far, however, it doesn’t explain why there are mental representations in the first place. After all, it holds for thermostats as well as human minds, and themostats do not need mental representations.

Still, the cybernetic model is the starting point for such an explanation, once we examine the complexities and opportunities the human cybernetic system has evolved to handle.

The first complication is the fact that our knowledge of the world comes to us in bits and pieces over time. This means that ε(p) — our knowledge that the state of the world is p — will in general require memory and integration of sense data over time. A mental representation of some sort seems a natural model for storing the result of such a process.

A stronger rationale for mental representation comes from a consideration of the practical limitations of the cybernetic model as described so far. When we talk about the state of the world, of course we don’t mean every single detail, or even every single detail it is possible for us to know; there are simply too many. Luckily in practice most of the details we might consider will be irrelevant for most purposes. In the snake example, what counts is a fairly small subset of details: those that determine that a snake is within striking distance of us. Even so, the possible combinations of details this encompasses is astronomically large: think of all the ways a snake might look, all the sizes, shapes, colors and patterns in all the possible orientations and contexts. And this is just for a single snake: of course we want the same behavior to hold if we see two or more snakes, or, for that matter, if what we see is not a snake but another animal that poses a similar risk.

The only practical way that I can think of to handle such complexity is through some kind of generalization. Let’s define P as a set of world states that includes p, Q as a set of specific actions that includes q, and R as a set of world states that includes r. Let’s stipulate that the relation p & q ⇒ r holds for any p ∈ P, q ∈ Q and r ∈ R; we will call this proposition M.

We can now write a generalized version of the cybernetic formula:

Φ(R) & ε(M) & ε(p) & ε(p ∈ P) & ε(q ∈ Q) & α(q) ⇒ q

Or: if R is a desirable type of goal, and an agent knows that doing an action of type Q when the state of the world is of type P results in a R-type state, and the agent believes the state of the world to be p, and knows that p is a P-type state, and knows that q is a Q-type action, and knows that q is an option available to the agent, then the agent will do q.

This may be more complicated to describe, but is much easier to implement on a practical basis. Without generalization of this kind, an agent would have to essentially relearn a lesson for every minor variation of the relevant factors. But it does impose an additional requirement on agents: they must have an understanding of types. A type is by definition abstract: it encompasses many instances, including the possibility of instances that have not yet been encountered.

Whereas a system that deals only with concrete instances could conceivably work entirely with sense data rather than representations, this seems less feasible in the case of a system that also deals with abstractions. On the other hand, if a system uses mental representations to deal with concrete instances, handling abstract types may be a small step: all the system needs to do is to reuse representations. If “snake” is a representation, and this representation is applied whenever the sense data fit certain criteria, and a number of different concrete instances can evoke the suitable sense data, then “snake” is a type, and the system is capable of generalization.

I will look at one more factor that argues for the existence of mental representations. The value of a cybernetic principle expands enormously if it can be created proactively rather than in hindsight: i.e., if the agent can come up with new formulas based on existing formulas, existing knowledge and mental processes (logic, intuition, etc.). In fact, we do this all the time, by using our imagination. We imagine possible future states of the world; we imagine possible actions we might take; and we imagine the result that might occur from an action. What could imagined states and actions and results possibly be, other than mental representations?

Note that none of the foregoing requires biology or evolution; rather, biology and evolution take advantage of possibilities inherent in the way the world operates.

In short: the laws of physics predict the power of imagination.

Preobjective ontology may be explained as follows:

1. We perceive what exists. That is, what we perceive to really exist, generally really exists.
2. However, we don’t perceive the “what” in “what exists”. We presume the what, and we perceive that it exists.
3. This presumption, of the “what” in “what exists”, is contingent and continuously changing.

The first point posits realism. The second point defines object mapping: to think about anything, or talk about anything, we have to objectify it. Perceptions are assigned to objects; perceptions determine properties; therefore, properties belong to objects, and not directly to existence. The third point states that the object mapping can and does change. In practice, it changes from moment to moment as our needs and attention change.

I think of preobjective ontology as a form of physicalism (the view that what’s real is just what’s physically real, the atoms and photons and other such things that physics says are real), but one that is more limited in its assertions than conventional physicalism. Classical logic cannot be directly applied to ontological questions, because it depends on reference and abstraction, which are not possible in a preobjective arena of discussion. But preobjective reasoning is possible, for example using a tool I dubbed material logic.

Existence precedes objectification. It sounds simple and straightforward, but it raises a subtle paradox for a physicalist. Objects are conceptual, by this way of thinking. However, if all existence is physical, then objects, too, must physically exist in some form, even though they are concepts. But how can an object exist preobjectively?

My answer to this is that objects are not as simple as they seem. An object is a feedback mechanism: a linkage between an effect and a cause, the way a thermostat is a linkage between temperature (effect) and a heat source (cause). In this case the effect is the state of the brain when it is thinking about, for example, tables, and the cause is sense data corresponding to tables. The table-brain-state and the table-sense-data must preexist the table object, because the table object is composed of the table-brain-state, the table-sense-data, and the control mechanism that connects them. Strange as it may seem, the table-brain-state must exist first, and only then the table object.

But it’s actually not so strange, when you think about how the table-brain-state might have come about. Surely it evolved from something simpler. My theory is that it evolved from the big-rock-brain-state, when our ancestors first began to figure out that rocks that were flat on top and about waist high were more useful than other big rocks (this cries out for illustration…)

And then there is entropy and the arrow of time.

The second law of thermodynamics predicts a steady increase in disorder (entropy) in a closed system over time. Like other laws that operate at large scales, the second law can be derived from the statistical behavior of large numbers of molecules. It has also been abundantly confirmed in empirical observations.

But there is a fundamental truth introduced by the second law which is oddly inexplicable at the micro level: the arrow of time. The second law of thermodynamics says that balloons don’t unpop, rust doesn’t turn to iron, and eggs don’t unscramble. The funny thing is that no other physical laws prohibit these things from happening. Is it possible that something we experience so immanently and intimately — the one-way flow of time — is just a statistical artifact? Many physicists and philosophers are unconvinced.

I won’t weigh in on that particular question. But I will say that there is a loophole in the second law: it only holds for a closed system as a whole, and doesn’t rule out local increases in order. In fact, it happens all the time: it’s called life. Every living creature is an island of reverse entropy, every living cell a little machine that absorbs disordered matter, burns some of it as fuel, and uses the rest to build things — such as new little machines.

Is life a special case? No. Every air conditioned home or car is also an island of reverse entropy. Perhaps you may consider air conditioners to be extensions of life, “things-engineered-by-life”, along with spiders’ webs, termite hives, hermit crab shells, water wheels and semiconductors. Even in the realm of life itself, discernible order occurs at many levels: proteins, organelles, cells, organisms, societies. If life is a necessary cause, it’s one with a multiple personality disorder.

From a reductionist physicalist point of view, it’s hard to get much of an explanation for this order — it seems to be random, a statistical artifact like entropy itself. In essence, from the reductionist point of view, there is nothing to explain: life follows the laws of physics like everything else, and in the closed system of which it is a part, the second law is honored.

But this is willful blindness. There is clearly something to be explained, and if science cannot explain it then magical and supernatural explanations will fill the void.

Not to fear, though. Indeed there is a logical explanation. It goes like this:

1. Systematic local reversal of entropy is caused by systematic feedback.
2. Systematic feedback is caused by directed behavior.
3. Directed behavior is caused by a Turing machine or the equivalent.
4. There exist Turing machines.
5. Therefore there may exist systematic local reversals of entropy.

Shorter version: life is a computation.

But the explanation is actually broader than life. It begins with a fundamental relationship between feedback and entropy. And this makes sense, because feedback can be seen as a way to reverse the arrow of time in a limited but real sense.

To understand this, consider the iconic case of the rudder of a boat and the steersman who controls it. Moving the rudder causes the boat to turn. The arrow of time leads from cause to effect, from changing rudder position to changing boat direction. But in the mind of the steersman, changing the direction of the boat comes first, changing the position of the rudder second. Boat direction is the independent variable; rudder position the dependent variable. This is easy to see by considering the possibilities: the steersman could just as easily decide to turn right as turn left, but once the boat direction is decided the rudder position is determined and cannot be otherwise.

So, in this sense, in a cybernetic system (a system based on feedback) an effect may lead to a cause. This, I would argue, is the root principle underlying the reversal of entropy. It’s the first and most important proposition in the five-point explanation above.

The second point states that the whole arrangement of rudder and steersman is not random; the rudder exists in the first place explicitly to enable the cause, in order to enable the achievement of the effect. So not only does effect precede the cause, desire for the effect precedes the possibility of the cause. The feedback system exists to fulfill a purpose.

The third point states that for the purposeful behavior to work, it must entail knowledge, in this case of the motion of boats and the effect of rudders. This could be conscious knowledge achieved through observation and reason. Or it could be genetic knowledge acheived through evolution. Either way, it is the result of computation, and computation requires a Turing machine.

The fourth point posits the existence of one or more Turing machines. So where do Turing machines come from in the first place? This is, of course, the perfect spot to throw in a supernatural explanation, an outside intelligence to act as the first cause. But a naturalistic account is surely more plausible: given that local systems of reverse entropy (life) have arisen naturally in the material universe, Turing machines must be endemic to nature.

As a materialist, I find the natural explanation better than the supernatural one, even if it requires going further than the conventional physicalist argument generally runs. In essence, the cybernetic view requires accepting a new duality: the duality of information and matter. Duality is not dualism; it is dual identity, two equally valid ways of seeing the same thing. The materialists says (correctly in my view) that there can be no information without matter. The cybernetic materialist says that, in addition, there can be no matter without information.

This is not as mystical as it might sound. It takes nothing away from physicalism, and offers nothing to the dualist. On the contrary, it seeks to expand the physicalist explanation to concepts, abstractions and logic, in other words, the building blocks of thought: they are fundamental attributes of matter, like charge and mass, that are inseparable from matter yet may be described on their own terms. Do charge and mass exist the same way matter exists? I don’t know the answer, but whatever it is I believe it also holds for logic and arithmetic. Perhaps they don’t extend quite all the way to fundamental particles; perhaps, like entropy, they emerge at larger scales. Either way, they inhere in physical existence.

Finally, we get to point five, the conclusion: if you are a cybernetic materialist, naturally there is life, and life reflecting on life. That’s the cybernetic explanation.

So if instead of stopping at

'P' → P

you go on to

P → ?

what do you end up with?

My intuition about disquotation is that disquoting is a step towards reality. Maybe disquoting again would bring you the rest of the way. So if you disqoute a reference to a thing you get that thing.

Unless of course that thing doesn’t exist. Disquoting is not magic. Disquoting a reference to a thing that doesn’t exist gets you nothing.

So let’s try it with a proposition: “There exists an elephant”

'There exists an elephant' → There exists an elephant
There exists an elephant → [an actual elephant]

One thing that jumps out here is that you can’t play this game without an elephant. But, there’s no rule that says the elephant must actually be in your possession. The elephant must exist somewhere. If we both believe that to be a fact, that’s good enough for us to take it as a valid proposition.

So what happens when something doesn’t exist?

'There exists a unicorn' → There exists a unicorn
There exists a unicorn →

What happens is that you can’t make a statement, because you can’t find a unicorn.

Unless you’re Marco Polo, in which case:

'There exists a unicorn' → There exists a unicorn
There exists a unicorn → [a rhinoceros]

Note that even if you can make a statement, there’s no guarantee that what exists is what you say exists — only that it exists.

What I find highly intriguing about this is that in the final disquoting step you move from a bivalent logic to a monovalent logic — instead of being true or false, a statement merely, but literally, exists. There is no opposite, because in this unusual logic if something doesn’t exist it can’t be a statement.

Yes, logic. I think of this final step as a kind of logic, a funny kind of logic but still a formalizable, verifiable, arguable logic. I call it material logic.

Material logic makes claims of existence. That’s all it does. There are no explicit properties in material logic, other than the property (if you consider it that) of existence. But this is precisely where propositional logic leaves off (I have classical logic in mind, but I think it applies to intuitionistic and other logics as well).

In classical logic you can propose that an elephant exists, and you might even be able to prove the proposition, given the existence of certain evidence, and perhaps even prove things about the evidence, based on other evidence — a causal chain, you might say — but at some point you have the original physical evidence. What makes it “physical” is that it physically exists. But this claim of existence must simply be accepted and can’t be proven. This is an unaviodable consequence of the representation problem: ultimately, we can only logically analyze our mental representations of things, but the connection between the mental representation and the physical thing lies outside of logic. Naming a thing doesn’t prove the thing exists.

The value of material logic is that it deals exclusively with the last step, the one which classical logic takes as a given — claims of physical existence. And, as the exercise above shows, it’s possible to trace a path from propositional logic to material logic, thereby extending the reach of logical analysis into the existential realm. This is not by itself a solution to the representation problem, or a way to prove physical existence, but it is, I believe, a tool for achieving progress towards these goals, by providing a better way to talk about physical existence and its relationship to mental states. Better in two ways actually: closer to what’s actually going on, and easier for us to reason about.

Material logic is, essentially a formalization of object mapping — the mechanism by which mental representations are established, according to the object-oriented ontology I have proposed. To make it useful, we need to take it a little farther than what I’ve described up to this point (but not much). First, however, let’s review what we have so far.

• A statement in material logic is a statement that something exists. There are no names, categories or properties in material logic, so the statement doesn’t say what exists, just that something exists.
• A statement in material logic consists of something that exists.
• There is no way in material logic to state that something doesn’t exist. So material logic is monovalent (there is no negation).
• Since every statement says that something exists, and every statement consists of something that exists, every statement in material logic is correct — you simply can’t utter an incorrect statement.

Material logic’s infallibility and monovalence mean that you can’t use material logic to claim truth or falsity. But certain truth claims are also existence claims, and they can provide a bridge between propositional logic and material logic.

However, if we want to think about material logic we first face a bit of a problem: material logic is unparseable. Take the case of operators. Operators in material logic must necessarily look different than those in propositional logic — in fact they won’t look like anything at all, because an operator is an abstraction and material logic cannot express abstractions. All we can do materially is express the result of applying an operator, if there is one. And since there may be many ways a particular result came to be, you can’t see what the statement was just by looking at the result.

Luckily, we aren’t required to talk about material logic in material logic. Clever creatures that we are, we can come up with readable, analyzable notation that represents material logic without being itself material.

Continuing with the convention I used above, let’s represent the mapping of object x onto the real world like this:

[x]

We can say that [x] is the materialization of x, or, conversely, that x is the representation of [x].

We can describe the basic relationship between propositional logic and material logic as follows:

∃x → [x]
[x] → ∃x

or simply

∃x ↔ [x]

Now we’re in a position to define operations. I’ll define three: intersection, composition and subtraction.

The intersection of two materializations consists of the substance they share. So, for example, the intersection of my head and my skeleton would be my skull. We can express intersection with familiar notation:

[x] ∩ [y]

Intersection implies some degree of identity: if two objects intersect, then they are in part or in whole the same in substance.

The composition of two materializations consists of the substance belonging to either or both. The union operator seems appropriate:

[x] ∪ [y]

The third operation, subtraction, consists of the substance that belongs to one materialization but not another, and is represented like this:

[x] - [y]

One shortcoming with the above notation is that not every instance of it corresponds to a statement in material logic. In particular, because two of the operations, intersection and subtraction, are not closed, it is possible to describe an operation that does not yield a materialization and is therefore not a materiological statement.

In this regard it’s useful to remember that an expression like [x] – [y] is merely a stand-in for the material result of the operation, providing the result exists; otherwise the expression is nonsense. Some expressions, such as [x] – [x], are inherently nonsense. Here classical logic can come to the rescue. Precisely because [x] – [x] is not a statement in material logic, we can state categorically, in classical logic, that ∃([x] – [x]) is false.

There is one more very important step we need to take with material logic for it to be of most use: we must incorporate motion and change. This will also bring us face-to-face with limits to our knowledge, gaps that cannot be eradicated. It’s a big question, and I’ve only begun to put together an answer, but here are some of the main points:

• Motion and change correspond to transformations in material logic. We can catalog and study such transformations.
• We experience persistence by detecting continuity in sense data, and we experience motion and change by detecting changes in sense data. The computation behind these mental processes corresponds to finding the transformation that best fits a series of materializations.
• However, what we experience in this way is neither persistence of substance nor persistence of objects, it’s the persistence of the mapping between them.

Substance may change in form but lasts forever, as far as we know. Abstract objects are themselves timeless and changeless. An object exists physically if and only if, and only as long as, it’s mapped to substance. My hope is that material logic will enable new insights into this mapping, and thereby into what it really means to say that a thing exists.

A name clearly is a property; the question has been, a property of what exactly. This becomes a problem when we consider the names of fictional characters, cases of mistaken identity and other situations where the exact thing being named is ambiguous, incorrect or even nonexistent. To address this problem, many philosophers have turned to exotic ideas such as many worlds theories and modal logic.

From an object-oriented point of view, however, the source of the problem is clear: the objective gap. Names are rigid designators of objects, not substance. An object is a mental construction which is imprecisely, incompletely and tentatively mapped to reality. The object itself always exists, as an object; but mapping it to something real doesn’t always mean that reality is what the object says it is. The object called Santa Claus clearly exists, and might be mapped to all sorts of things — a picture, a guy in a costume at the mall, the words “Ho ho ho!”. All of these things are real, and they are all Santa Claus in a sense, but they are more something else than they are Santa — other objects can be mapped to them with more plausible claims of existence.

But mapping Santa Claus to things that aren’t literally Santa Claus is not only possible, it can be useful. It lets us judge the quality of the mall guy’s Santa performance — how convincing is his protrayal? How much like the “real” Santa?

Let’s say, though, that we recognize the guy in the Santa suit, and he’s someone we know named Al Jones. In that case we will apply another object, “Al Jones”, which is an identity rather than a role. So this mapping will be more exclusive and permanent: we expect the man to remain Al Jones even after changing out of the suit, and we don’t expect the man to also be Bill Smith or anyone else. In other words, we accept this mapping’s claim to existence.

But a funny thing happens at that point. Instead of judging flesh-and-blood Al Jones against the “Al Jones” object, the way we judged him against the “Santa Claus” object, we judge the “Al Jones” object against Al Jones. That is, we compare the object to our perceptions, and if there’s a problem we modify the object to match the perceptions, or throw it out.

So if we walk up to Santa and find out we were wrong, and see that it’s actually Bill Smith in the Santa suit, we must unmap “Al Jones” and map “Bill Smith” instead. This has no impact on the existence of the real Al Jones or the real Bill Smith, or for that matter the “Al Jones” or and “Bill Smith” objects. Ontologically, nothing has changed. All that changed was the mapping.

In many ways, this view is resembles contextualism, only as an ontology rather than an epistemology. Contextualism says that what we can claim to know depends on the context. An object-oriented ontology says that we what we can claim exists depends on the context, context in this case meaning the selected object mapping. In particular, we cannot say that an object exists based purely on its own properties. Those properties were selected on the basis of a mapping, and other properties were not selected. To justify our belief in the object we have to look at the whole picture.

In this way, objects are like shadows. A shadow exists only by virtue of non-shadow; if there weren’t light to be obstructed, there wouldn’t be a shadow. A shadow is a subtraction.

The process of deciding what exists is also largely a process of subtraction, of removing noise so we can detect the signal. There are an infinite number of things that might exist, and many, many that might plausibly exist at any moment, so many that we must eliminate huge numbers of them at every turn.

But we also have the power to use mappings that don’t completely fit reality, but are nevertheless useful — metaphors, analogies, figures of speech. In such cases, the object’s properties are more valuable than its claim to existence, so the later is disclaimed.

Logical propositions are in this category. To reason about an object, knowing its properties is more important than whether or not it exists. In fact, you could say that much of the power of classical logic is precisely the ability to reason about things that you’re not sure really exist, or even know for a fact don’t exist — things that lay in the future, for example.