Agency

 A key premise of this article might seem absurd on its face:  biological evolution will only produce automata.

The apparent absurdity rests especially on one’s personal experience of feeling, believing, desiring and choosing freely.  So I ask that you suspend your disbelief that you are an automaton, briefly.

I will try to show that our intentionality makes a lot more sense (and many of the problems of philosophy disappear) if we start with our automaton natures, and back into our subjective experience.

The modern age has an advantage in this kind of thinking, with artificial intelligence (AI) as a model of automatic processing.  Descartes, of course, developed his hydraulic model of brain functioning based on the hydraulic automata in the French Royal Gardens.  But we have robots with cameras and audio sensors that respond appropriately to sensory input but do not see or hear intentionally; robots that process data and solve problems but have no beliefs.

Robots mimic intentionality, but are mere rule-followers, utterly lacking in personal awareness. 

Agents are sometimes future-driven (have intentions), while automata always respond to stimuli according to built-in rules.  The classic example is a chess computer program.  It mimics the desire to win, and it mimics peering into the future (at all the possible sequences of move-and-counter-move).  But in reality its behavior is entirely determined by rules the results of which could be calculated with paper-and-pencil.  A chess computer mimics intentionality, but entirely lacks awareness of past, present or future, the use of mental entities that are about something, aims, free will or anything mattering in the least.  I lump all such characteristics together as intentionality.

My premise, then, is that biological evolution produces only automata, so that you and I are, in a key sense, automata.  Our overt and covert behaviors are determined by neural processes that can in principle be explained by microphysical processes.  Our obvious intentionality is actually (in a key sense) simulated intentionality, brought to wonderful sophistication by the forces of evolution, because organisms that work together in producing civilizations and science are more likely to survive.

I’ll try to show that it’s valuable to accept all of this as true, and only then try to deal with the personal experience of intentionality.  Ultimately, however, I hope to turn that order of things on its head.

Let’s be clear what automata lack.  They

do not feel pain or have perceptions

         have no idea that they exist

  have no sense that there is a world, no semantic relations

  have no interest in survival or in the avoidance of suffering

  have no goals of any sort; no desires or aims

  are not originating causes; that is, their overt and covert behaviors are like that of a cog in a clockwork, simply producing such outputs as are determined by their inputs.

In saying that automata are not agents, we imply that the world, in the absence of agency, is simply a causal chain, an interconnected whole with everything having causal relation to everything else, the entirety simply playing out the inevitable given the original causes.  (I ignore randomness, in that I don’t think it impacts questions of free will or intentionality).

How Evolution Might Produce Humanlike Automata

There is all sorts of mimicry in nature.  Some fish are camouflaged as rocks.  Some orchids have petals that mimic the appearance of female flies, thus attracting male flies that pick up their pollen.  An anglerfish dangles a lure.  Evolution, of course, has no intention of mimicking.  It’s just that whatever works tends to stick.

Automata have no awareness of their surroundings.  Instead, they have fixed outputs that are triggered by various inputs.  But evolution rewards mutation in the direction of mimicry of belief.  A housefly seems to believe that your moving hand represents danger, and we can easily imagine this as no more than rule-following mechanisms of a system that lacks sentience or awareness.

Biological evolution is likely to reward any simulation of intentionality, including the ability to learn.  The ganglion of a cockroach’s leg joint is capable of learning, but is surely not conscious or intentional. 

In any event, we can imagine evolutionary processes creating ever more sophisticated response tendencies, until there are automata that write operas and produce word strings that mimic philosophy, without intentionality.

There are few areas where I find value in Freud’s models, but I like his distinction between id and ego, and it can be useful in making a modern distinction between one’s discovered urges and the highly intelligent processing that we do to achieve them.

Roughly, Freud might say that our urges are more likely to be discovered than chosen, and are often rather primitive.  He likens the id to a blind beast, desiring, but unable to model the world and figure out the best ways to get what it wants.

The ego, then, is a sophisticated processing capability.  Activated by the id, it processes the various alternatives, and comes up with plans to get the id what it wants.  Perhaps Freud would treat id and ego as agentic.  But we might equally treat them as automata.  The lower brain simply produces automatic behavioral tendencies that we interpret as urges.  The cortex simply processes alternative execution strategies.  It doesn’t originate motivations, but is enslaved to the motivations produced by the id.

What Leads Us To Assert Agency?

Why should we insist that we are agents?  I submit that it is not our intelligence – akin to Freud’s ego.  Contrary to the world according to Star Trek, neither is it our human emotions – the id. The emotions and intelligence of a system can be accounted for by automatic processes.

But we might add another outdated term:  the will.  In self-modeling, I note external pressures toward certain behaviors, along with my personal power to resist.  I note barriers to my goals, but personal power to persevere.  I note the capacity to imagine possible futures, and my power to strive for what I want.  Whatever the reality, it is both adaptive and consistent with such personal experience to model oneself as an agent.  It’s also the simplest, most powerful self-model.

In addition to noting acts of will, we cannot help noting that we have conscious streams, including mental entities that have aboutness.  There is the tendency among some philosophers to deal with consciousness as though it were a free-floating entity, able to be analyzed separately from the subject who experiences it.  But I think that conscious experiences are inextricable from the one who produces or observes them.  The key is to have somebody home who is conscious, who is intentional, who genuinely desires, who is sometimes able to manipulate the environment to achieve his or her ends.  Insofar as I claim to experience a conscious stream, I seem to imply being an agent.

An agent is one who actually means anything, who is aware that there is a world out there, who personally experiences, believes and desires, and who is sometimes an originating cause.  An agent is whatever is personally conscious in these various ways.

I doubt that there is meaning to pure consciousness without agency.  There is no conscious stream of mental entities such as perceptions without there also being personal experience, belief and desire.  Consciousness implies a personal perspective.

The Declaration of Agency

We are self-declared agents.  Is our claim true?  Imagine Robbie, an early 21st Century robot, designed to learn to operate successfully in a simplified environment.  Robbie has processing software for each of the following:

Receiving and processing data from its (controlled) surroundings and fitting them into a relatively sophisticated model of its world

 

Having goals and rules for prioritizing them

 

Having modestly effective models of what will influence the accomplishment of those goals

 

The tendency to execute behaviors that move it closer to its goals

 

Having a primitive model of itself, including all the above, and the tendency to modify that self-model with new data.

Let’s stipulate that Robbie is not remotely an agent.  It is built within the technological limits of AI at the beginning of the 21st Century.  But asking whether we would grant it agency is, I think, the wrong question.  We will learn something by asking, should we design Robbie to “conclude” that it is an agent?  I think that we would be twisting things strangely if we did otherwise.  Look at it from Robbie’s “viewpoint”.  In its self-model, Robbie would in the normal course of things model that it interferes in the course of things (its own processing is a determinant of its behaviors).  It properly models that it directly perceives the world, and has beliefs about it (it cannot effectively model itself to be unaware).  It has a modest model of the future with itself included, and it would be natural for it to model its choices as reflecting internal preferences (in that it discovers “preferences” (priorities) that are predictors of its behaviors).

Robbie is designed to model future outcomes and their consequences as to Robbie’s goals, and to behave appropriately.  Modeling itself as an automaton might actually stand in the way of this sort of processing, leading Robbie to await its next behavior, rather than actively “choose” it.  Robbie’s design powerfully mimics intentionality, and modeling itself as an agent is by far the simplest and most powerful model.

Let us say that we build the processor for Robbie with the syntactic equivalent of semantic space.  For example, it has a “semantic space” for colors, such that its receipt of a given wavelength by its camera activates a location in a simulated multi-dimensional space, relative to red, yellow, blue, black, etc.  If asked to describe the relation of this wavelength to any one of those colors, its processor produces an answer based on this “semantic space”.  Similarly, its “semantic space” for sounds locates a tone relative to frequency, amplitude and timbre.

We would appropriately design it to assert that it perceives color and sound directly, and that redness, “semantically”, is more closely related to blueness than to the sound of a trumpet.  When it recalls redness, it will activate the identical “semantic space” that it uses for seeing the color of an apple, and will truly “assert” that it recalls the same red.

I propose, then, that we can imagine designing a robot that takes the intentional stance as to itself.  It might not at all pass muster for our criteria for sentience, intentionality or agency.  In fact, we could predict its every string of words, because it is only outputting by means of rules that we created.  Nonetheless its design has it declare itself to be an agent, and to declare that it has the direct, subjective experience of redness when it processes its camera’s input from an apple.  It might have no way of finding out whether its assertion of subjective experience means anything like what you and I mean.

In summary, a robot that is not in the least intentional can be designed to self-analyze and declare its own intentionality for appropriate reasons.  Such a design is highly adaptive, and it would not be surprising to find that nature favors such a design as it mimics intentionality.

Because we designed Robbie to mimic intentionality, it is natural for its self-model to take the intentional stance.  In the same way, because evolution tends to produce systems that mimic intentionality, it is natural for us to use the intentional stance in self-modeling.

You have the option of using the intentional stance with me, or not.  But you don’t have that option with yourself.  Given your design, you will model yourself to personally experience, believe, desire and interfere in the course of things, because no other model does remotely so well in fitting your design or needs. 

I feel certain that I am an agent, because of my own subjective experience.  But we now add the caveat that Robbie might assert having precisely the sort of subjective experience that you and I have.  One must conclude that I can no more tell whether I am an automaton than Robbie can.  (This conclusion leads one to ask, what is it about a belief that makes me so sure that I am personally believing, rather than doing something akin to what Robbie does.)

It Is True That We Are Agents

One is tempted to conclude that materialism holds, and that we are merely automata, with our claims of agency being false.

But such a conclusion depends on the false notion that there is a single truth that alone can make sense of everything in the world.  To the contrary, we require two explanations of reality:  the agent explanation and the automaton explanation, because there are valid purposes to each (see the “Knowledge” article), and neither can ever be adequate to describe all that is to be described.

The automaton explanation is radical materialism.  If we ignore randomness, I suspect that the ideal version of this explanation is complete, able to predict and explain the physical world, including human behavior, ideally. 

But it has a fatal weakness, in that it asserts that there is no one to make assertions.  Once we stop conflating it with the agent explanation, it fails to allow for meanings, belief, the semantic relation, pain, desire, things mattering and more.

The agent explanation properly accounts for all of those things.  It asserts that we are genuine agents with free will who use genuine semantics and have genuine qualia.  Assuming that the agent explanation otherwise mirrors the materialist explanation, it can successfully explain both the physical and the mental.  The weakness of this model is that it is difficult to picture oneself as both an originating cause and as a mere link in the causal chain.

Dennett uses the notion of an intentional stance as a mere strategy.  I will use the term ‘subjective stance’ as a genuine belief as to oneself – that I am an agent.  Taking this stance creates a personal perspective that makes knowledge possible.  We can produce a true explanation of the world from that perspective, but a different true explanation if we simulate taking no perspective at all, in the automaton explanation.

The subjective stance is a declaration of separateness.  Whereas the automaton is a cog, this stance is a claim to be a surroundings, observing and influencing them from a particular perspective.

The very declaration creates separation in a manner that did not previously exist.  The automaton model implies that the world is a unity, with everything causally inter-related and no present-time originating causes.  But the subjective stance is a personal perspective that creates an observer, outside of the world and modeling it.  It creates an agent, an originator of causes, interfering in the course of things.

That is, by casting yourself as personally perceiving the world and pursuing aims, you become one who breaks the causal chain, within the agent explanation.

This becomes well defined only as you develop a sophisticated model of the world.  Such a model makes assertions about the world, and must include you as a believer of them.  It must take your personal perception of your surroundings to be I/it claims about the world.  It must include your aims as powerful personal relations to the world, as in, I desire that.  And it must include you as an agent with free will, personally being pulled by your aims, personally considering alternatives, and sometimes in charge of choosing your path.

Knowledge is a three-part relation that entails agent, meanings and world (see “Knowledge”.)  The agent explanation, in order to be complete, must adequately account for agent and meanings, as well as the world.

Of course, you are free to model yourself, simultaneously, with the automaton explanation.  You can acknowledge the power of that model, but you cannot take it on for your personal experience.  That is, the automaton explanation will inevitably seem fatally flawed from the subjective stance, because it fails to deal with agent, meanings and the various intentional relations.

The effect is that each subjective stance (yours or mine) is the very basis for our known world.  It creates semantics – the I-meaning-object relation.  In doing so, it also creates free will, pain, the future, knowledge and anything mattering.

We don’t need to endow it with metaphysical mystery to see that it is the sine qua non.  We can acknowledge ourselves as automata with one explanation.  But we must acknowledge our agency with an even more fundamental model of reality.

I doubt that there is independent truth as to whether a housefly is an agent, or as to whether it feels pain.  A housefly is certainly an automaton in one sense, and one that nicely simulates intentionality.  It has neural processes that act as though they are entities that are about its surroundings, asserting that this is food, that is danger and over there is something to mate with.  When injured, the housefly writhes.  But we can guess with confidence that it does not have a self-concept, and doesn’t distinguish between the injured body part and the pain.

If pain is a real, immaterial entity, then there is truth about whether the fly feels pain.  If pain is, instead, a relation between an injured body part and an agent created by a self-concept, then there is no independent truth about whether the fly (or anything else that lacks a self-concept as agent) feels pain.  Rather, it is a matter of convention in how we apply the term ‘pain’.

Dennett seems to imply that the intentional stance is a mere cognitive strategy.  The implication is that there might come a day in which the single correct way to understand ourselves is from the automaton explanation.  I don’t think that works.

I personally feel pain, and it would be silly to take the stance that I am an automaton that is incapable of pain (see “Dualism” for a discussion of pain and other qualia).  However, like Dennett, I acknowledge that I can without contradiction treat people other than myself either with or without the notion of intentionality.  Only by inference from my own case can I conclude whether there are other agents.

I conclude that there is a true automaton explanation.  However, that explanation is inadequate to explain key aspects of experienced reality, and must be rejected as the sole explanation.  We require two non-combinable explanations, both of which are genuinely true (see “Knowledge”).

By the way, the conventional scientific explanation, in my view, fails to carry materialism nearly far enough.  It ought to deny that there are knowers, and therefore that there is knowledge.  It ought to deny that there are intentional creatures, and therefore any meanings.  That is, it ought to deny that there are genuine semantic engines.

The conventional scientific explanation does not at all reach those conclusions.  Perhaps the typical thoughtful proponent would assert that evolution somehow brought about genuine meaning users, in a way to be determined some day.  This is a position I can’t easily refute, but I suggest that it leads to all sorts of problems that philosophers today struggle with.

Context

Note that my model of the world is private to me, even if innumerable others use the identical model, each from their own perspective.  My model occurs only in my experience, even though it is heavily influenced by society, and is known to be shared to a large degree.  We know by inference that there are other agents, but all that happens appears within each person’s private experience.

This is important because it means that my awareness is the very context for experienced reality.  When I’m unconscious, that experienced reality disappears (although I can later infer that a great deal happened while I was unconscious).

Agency is the very context for perception, belief, desire and choice.  There can be a model of reality in the absence of agency, but one cannot adopt that model consistently for oneself.

So what do I mean, here, by context?  The automaton explanation is provided from no given context.  But in my personal experience, I as subject am a given, even in extreme circumstances.  For example, when I introspect on my own conscious stream, it occurs in an I/it relation, with the “I” as the very context.  If I were to have an out-of-body experience, suffer complete amnesia, or have a psychotic break where I find myself to be Napoleon, the “I” would remain as the very context.

In any given first person perspective, the context is one of I/it relation to one’s surroundings, including, I perceive, assert, intend and initiate causes.  There can’t be semantics outside of this context. 

 Roughly, context means perspective.  The world occurs for me from my first person perspective.  I am the subject who observes, experiences, etc.  In addition to the self-as-object, there is the self-as-subject.

Note that grammar leaves us unable to say anything about the subject (I) without making it into an object (me).  The subject is the very context for experienced reality, rather than something that can be analyzed as an object.  When I say, “I am a man”, the predicate, “man”, is something I have.  If I awakened in a female body, I would say, “I am now a woman, but I am still I”.

The context about which we’re talking is not a mind-independent part of reality.  Rather, I’m talking about the practicalities of how we self-model, whether or not it is legitimate.  My model includes me in two ways, as a subject and as an object.  The object has mind, personality, etc.  But the subject is the one experiencing from my perspective, and is fundamental to there being knowledge and agency.

When Robbie self-modeled as an agent, it created a model of the world in which Robbie as subject is the very context for all that occurs, and so is outside of that modeled world.  Robbie can develop a true model of the world from that context, in which Robbie genuinely believes, experiences, desires and chooses.  There is no model that is inherently more adequate than that one.

Cogito ergo sum might simply mean, I think therefore my mind must be real.  But it might also mean, I think therefore there is a subject.

In the automaton explanation, there is no separation from the causal web; there are no originators of cause.  In the agent explanation, context implies separation, with the observer being outside of what is observed (and in that sense free of the causal web).

I suggested that the automaton and agent explanations are non-combinable.  I argue elsewhere (the “Knowledge” article) that the two explanations are precisely mappable onto each other – in no way conflicting as to the world.  They can be equally accurate in prediction.  But they differ fundamentally in the context from which the world is modeled.  Agency as a context is fundamentally separate from the physical world, permitting personal belief and experience, and being the foundation of free will.

Free Will

It is true in the automaton explanation that I lack free will, and it is true in the agent explanation that I have free will, despite the fact that the one explanation can be accurately mapped onto the other.  The difference that matters is that there is no inherent context for the automaton explanation, whereas the agent explanation takes the context as primary. 

The agent explanation grows out of my having personally modeled the world.  I am the creator of the experienced world:  it actually occurs within the context that I am.  The objects in that world might be caused, but I am not an object.

Now, I do discover myself as an object in that world, and I might well discover that object to be a mere cog.  But my use of the automaton explanation cannot legislate me out of my inherent position as subject, agent, context for the experienced world.

Without me, there is only an undiscriminated, meaningless whole.  Only my context allows for the I-meaning-object relation that creates semantics, and so creates a discriminated, meaningful world.

As agent, I have intentions – aboutness that enables the semantic world, and aims that drive my behavior.  When I encounter genuine alternatives, I choose teleologically, according to my purposes and forecasts, rather than as a mere relay of prior causes. 

For example, suppose that I gain access to a computer that perfectly predicts my every behavior, based on the automaton explanation.  This would simply be added data, which I would use for my purposes, sometimes doing as predicted, sometimes seeing my predicted folly and avoiding it, and sometimes doing otherwise just to be contrary.  It is my very semantic relation to those predictions that illustrate my agency.

I can acknowledge that my mental processing is supervenient on neural processing, and is thus fully determined.  My free will, then, is not a breaking free from the causal chain.  I, as agent, do freely choose what is to my benefit, using neural processes to do so.  From my personal perspective, I must analyze my situation and direct my course, unable to surrender my agency to the causal chain.  Knowing that others have already determined which choice I will make is interesting, but it doesn’t free me from the responsibility of determining my own future.

Do we, then, have free will in each other’s agent explanations?  In principle, I can model myself as an agent and everyone else as automata.  In practice, we almost universally model each other as agents, each creating his or her own experienced world, and each having purposes.  We have thus created a true, shared agent explanation.

 The Intentional Stance

In The Intentional Stance, Dennett describes the marvelous power of treating entities as though they were intentional.  He seems to imply that there are no truly intentional creatures, and no genuine semantics.  He is correct that I am merely a rule-following system in the automaton explanation.  What he misses is that I am also truly a user of semantics in the agent explanation.  I personally mean something by my mental contents.  I have a rich mental life of genuine and sometimes visual meanings.  My words have full, rich semantic implication, both denotative and connotative, whereas my syntactic events have none at all, being merely complex aspects of an automatic reaction that will generally lead toward survival.

You and I acknowledge each other’s subjective manifesto, granting agency to one another, and building on the assumption that each has a full mental life.  Out of that granting comes the fullness of civilization, the building of meaning across generations. 

But the foundation is a single agent’s subjective manifesto, and all that it implies.  I can grant agency or not to anyone else.  But my own agency is a certainty, in the only possible known world:  the one experienced by me personally.