Does the computational theory of mind explain anything?

Question

In science, when you theorize that X reduces to Y, you propose a theory that links X and Y in some causal way. Physicists don't just say, "What you experiences as a gas is really a swarm of fast-moving molecules"; instead, they use statistical mechanics to show that with this theory, the kinetic energy of the molecules, taken in aggregate, can explain gas pressure, temperature, and the ability of expanding gas to do work. Similarly, biologists don't just say, "The apparent design of the various capabilities of the various species can be reduced to a long sequence of random mutations"; instead they offer a theory of natural selection to explain why the adaptions seem to have a purpose, to have function.

The computational theory of mind doesn't seem to have this characteristic. Those who make this claim say, "Our mental states can be reduced to a set of physical states/functional states/some other feature of some sort of computation. Now, clearly this sort of theory has the potential to explain at least some aspects of behavior, but I don't see how it explains anything at all regarding the mind.

When I'm thinking, I generally don't experience anything that I would call algorithmic unless I'm consciously executing an algorithm in my mind. What is the connection from mental states comparable to how statistical mechanics connects molecules to the gas laws or how natural selection connects random mutation to apparent design? Where is the theory about the relationship between computational state X causes desire Y or emotion Z or intention W?

Note, it's not enough to show cognitive studies about how certain chemicals are associated with certain emotions, or activity in certain parts of the brain are associated with certain intentions, or evolution can justify certain desires; those observations/theories still leave an explanatory gap between the physical and the mental. Statistical mechanics shows mathematically how the kinetic energy of many molecules is mathematically equivalent to pressure. Where is that equivalence in computational theories of mind? And without that connection, exactly what is the value of those theories?

ADDENDUM:

I've gotten lots of knowledgeable answers, but none of them really address my question (except the one that says that there is no such explanation). Several of the answers point to non-mental things that can be explained by a computational theory of the brain, but I'm not asking about that. Others have pointed out that there is no other explanation for mental states, but that is no reason to think the computational model is correct if it can't explain anything either. Several people have suggested a two-step process where the brain acts like a computing device and causes mental states, but that's just another way to say that one can model the brain as a computing machine; it doesn't explain how mental states just are some aspect of the computation. In fact, it seems to assume a sort of Cartesian dualism, where the brain does something and the mind is an independent sort of substance that just follows along.

So, once again, I'm not asking about computational models of the brain; I'm asking about computational models of the mind. How is anything that happens in the mind reducible to something that happens by computation.

Answer 1

How is anything that happens in the mind reducible to something that happens by computation.

As of 2023, nobody knows how the interesting parts of the mind (i.e., the feeling of "self" or of "being aware") works. You won't get an answer here or anywhere else that does not require either some amount of belief unfounded in logic or objective facts; or the trust that one of our existing theories will eventually be fleshed out fully to explain your questions.

Not a single one of the philosophies of the mind out there are strictly scientifically falsifiable or able to pass skeptical dismemberment like you are doing here. Or put the other way round, as NotThatGuy commented, "evidence supporting [non-dualism] is overwhelming and we have no contrary evidence".

The closest one can (in my opinion and experience) get is to experience the working of the mind carefully, i.e. by meditation, and arrive at the personal belief that non-dualism might be closer to reality than dualism. That still does not give you any clue into how anything works, but it lets one expect (if one makes that experience and is open to the conclusion) that the mind is a part of the regular, physical world, not something "on top" or "aside of" physicality. And once you are convinced that the mind (even though we do not know how) is purely physical, then the conclusion that it is simply evoked by the visible structure of the brain (i.e., neurons, signal propagation, computation) is one which a philosopher can adapt, even without knowing how this gives raise to consciousness or other high-level features of the mind.

Finally, note that the Computational Theory of Mind does not state that the brain is equivalent to what we think of as a computer, but that it is a computational system, i.e. a system (with no specific implementation) which processes information. At first glance, and formalized through the Chinese Room Experiment and other thought experiments, it seems like at least on a high level, this seems like a very easily explained intuition - you can interpret all sensations as input, and all actions as output, and there you then have a perfectly functional computational system (implemented as a black box).

As to what it explains - it does not need to explain anything. As of right now, the mind is firmly in the purview of philosophy, not science. Specifically since you have made abundantly clear that you are not talking about the brain, but the mind. Philosophers are perfectly allowed to wonder about how things work, without any "use" or benefit except thinking about an interesting way to explain the world.

Answer 2

Does the computational theory of mind explain anything?

Obviously. It explains human thought and behavior in terms of physical computation (SEP). Whether or not you like the explanation shouldn't be confused with the fact it is an explanation.

In fact, there simply are no other decent alternatives to the scientifically minded. To explain thought, the Ancient Greeks had the classical elements which are meaningless. Descartes tried to insert an imagined supernatural being into the pineal glad which is silly. The homonculus is at best a clumsy metaphor for a biological control system. At least seeing the brain as an organ that conducts physical computation isn't laughable to the educated mind. Plus it has the upside of being fully friendly to evolutionary thinking as neurons and the endocrine system are explainable in terms of biochemistry.

Is computationalism fully explanatory? Obviously not. It gets somethings right, like providing a scientific basis for psycholinguistics, for instance. Plus, it has the added benefit of being consistent with computational neuroscience. What other alternative is there for trying to explain phenomenalism? Vitalism? Voodoo? Golem magic? Mythical gods?

observations/theories still leave an explanatory gap between the physical and the mental.

Yes. Duly noted decades ago by Chalmers. Your entire post might be called an introduction the hard problem. And even though the issues you raise are well subsumed by Chalmers, he himself doesn't retreat from the notion that the mind and body can be reconciled and that computation plays a role in explanation. He simply accepts property dualism. But what he doesn't do is declare the notion of thought as mental computation as meaningless. That a Turing Machine is an abstraction of a human mathematician means that we shouldn't view human thought as a synonymous with the modern von Neumann architecture; it clearly isn't. But that connectionism supports both analog and digital computation, and that symbolic systems can be seen as special cases of connectionism means that neurons (and biochemical processes of the endocrine system) clearly have an explanatory role in thought. Whether one wants to add more exotic notions to computation like quantum entanglement or that connectionist models should be treated as instances of analog computing instead of digital computation doesn't get away from the core premise that human thought is essentially tied to physical computation.

Where is the theory about the relationship between computational state X causes desire Y or emotion Z or intention W?

It's right here. It's called the somatic marker hypothesis:

"Somatic markers" are feelings in the body that are associated with emotions, such as the association of rapid heartbeat with anxiety or of nausea with disgust. According to the hypothesis, somatic markers strongly influence subsequent decision-making. Within the brain, somatic markers are thought to be processed in the ventromedial prefrontal cortex (vmPFC) and the amygdala.

You're right to claim you don't introspect your neurons or your limbic system functioning, because, evolutionary, that would be stupid to provide your conscious mind that information. Neural reentry serves the purpose of providing a control system to keep the systems of an organism moving in the same direction to survive, not providing a catalog of experience of the full body's function for your philosophical curiosity. Just because you can't witness your amygdala functioning consciously by internally visualizing a stream of non-ordinary differential equations doesn't mean that the it isn't a lump of tissue that can be modeled used theoretical neuroscience. Take a lesson from Descartes. He comes across as arrogant to think that his personal introspection is somehow an authoritative source on truth in the universe. It clearly wasn't. Nobody's is. That sort of certainty died with Hume's argument of inductive fallibilism.

Has computationalism moved beyond first generation cognitive science to embrace connectionism? Yes. Might it move into a third, fourth, and fifth generation? Probably. But it ain't going anywhere because metaphysical speculation based on emotional personal introspection is a defective remedy for ignorance, and the slow churn of the scientific methods will chip away at the problems continuing to move philosophical questions into the domain of the sciences.

Answer 3

You are correct that we have no reason to even speculate that "quality" can emerge out of abstract information. This is because not a single mathematical structure has been known to derive a "qualitative" aspect. If you have an axiomatic structure, you can infer other mathematical truths from it. That is the closest thing to an "emergent phenomenon" that you have in mathematics. So, "quality is an emergent phenomenon" is just a bunch of hand-wave with no meaning, as there is no such thing as "quality" in the framework of mathematics.

With that said, I want to point out that the usual approach of computationalism is to say that "qualia don't exist". That is, right now you have immediate access to abstract information about sound and colors and that the information has no qualitative aspect to it. A computationalist can always just say that they themselves only have access to "information" and you would have no way to prove them wrong.

However, to those of us who do not deny qualia, computationalism puts the cart before the horse. This is because we take qualia as things that we know to exist first and foremost. Physics is something that we derive from our subjective experience by filtering out the qualitative aspects and by trying represent our knowledge using symbols. To test physics, we have to map these symbols back to aspects of our subjective experience.

Computationalism then asserts that this filtered out product is the only thing that exists and that the original source was an illusion. This, of course, puts the cart before the horse. And it is probably also factually wrong as it is the standard viewpoint among physicists that there is no final model of physics

Answer 4

A great question! And no, I don't think the computational theory explains much :) Even though technically valid, it's too general a description to be useful. It's like describing an mp3 player (remember those?) as a computer that runs algorithms. Even though it is and it does, it might be more helpful to describe it as a record player (and, as such, no different from a cassette player that is perfectly analog and computes nothing).

Applying the same logic to the human mind, one could say, for example, that a neural network (whether artificial or inside a human head) runs statistical analysis and statistical inference, not algorithms. One could also propose that neural network's purpose is to distill experience (its training) into John Locke's "simple ideas" (or Kant, I believe, called them "intuitions", or data scientists call them "models").

And that's what Daniel Kahneman's System 1 is, essentially -- a very deep neural network.

As for Kahneman's System 2 (our rational mind), it's more akin to a laptop running a role-playing game: Its purpose is to visualize the Reality, including its internal operation, in 3-D. That's how we understand things, by discovering (through continuous attempts to visualize) interactive models¹ of them. Unlike our laptops, however, our System 2 doesn't come preinstalled with its copy of the Simulation. So, instead, we've evolved to spend extra 5-7 years in childhood -- so that each human child could assemble² their own copy, piece by piece and under the guidance of knowledgeable and supportive adults, as well as older children.

And that's what we could do better, our educational system: Teach every child the art of understanding (the art of "driving" their System 2) more intentionally and systematically. As it stands, it's touch and go.

 

¹ Or Locke's complex ideas.

² The Ancient Greek word for the Simulation was Logos -- and it's a derivative from an even older verb lego, which means "to assemble".

Answer 5

At this point in time, i am not aware of a single complete unifying computational model of mind. Instead "Computational model of mind" seems more of a paradigm expecting that an actual model of mind will be invented one day.

As a paradigm, computational models seek to offer an explanation to limitations and disturbances of mind. Like why we sleep, why we dream, why our mind works different under the influence of drugs, how we can learn faster, we we suffer optical illusions, why it's difficult for most to learn and apply maths (but easy for some), why some people are significantly less able to solve problems or live independently, and so on.

As a paradigm for explanation of the mind, it seems like the only materialistic alternative, all other explanations of mind would be dualist and use "magic", and thus not explain anything in a better way unless magic was real and properly understood.

Most generally, such a paradigm explains why the mind can do all things a computer can do (while not yet explaining the things the mind seem to be able to beyond that). As in "why can minds do number calculations? Because minds are partly like computers."

However this does not imply the mind is a computer with all properties, only that such abilities are explainable by considering the mind as partly a "weak" computer for such tasks.

Some other things that could be considered "explained" are: why it takes time to get a joke, or do calculations, why sometimes we fail to remember a thing we know, the effects of split brain experiments, why mind is developed by schooling. If mind were just magic, those things would lack for a reason. A magic mind has no obstacle to flawlessly always remember everything, do perfect math, come to conclusions instantly, become developed without school, function as a unity with a split brain.

A paradigm without any complete unifying model is unsatisfying, but not unusual as a way to cope with limited observability.

As an example mankind already had paradigms of biology without microscopes, of geology without long distance travel, of cosmology without telescopes, of atoms, of the weather, of evolution, ... Any such early paradigm would define expectations of how a future model might explain observations.

Answer 6

What it's attempting to explain is "why some thoughts take longer than others" (insert appropriate epistemic caveats here; it's been a long time since I've approached philosophy in anything resembling an academic context).

Clearly this is empirically true. Even for the same thought, "4 x 6 = 24" takes longer for a child to grasp initially than after they've memorized it. And it similarly takes slightly longer to figure out "4 x 6 x 25" than "1 x 2 x 3".

The attempt to generalize this to everything in the brain being some sort of computation may come from the observation that there are only so many different kinds of neuron in the brain, and neuron activation looks awfully similar to the things you can do with logic gates.

Where the computational theories are weakest is in my view is qualia, in similar fashion to behaviorism (as per Ryder's answer). Qualia are treated as a side effect, though I suppose there is the possibility of them being treated as "subtle information changes".

This may be testable in the near-ish future by observing whether, or to what degree, "smart" prosthetic or otherwise inorganic programmable augmentations (for instance, cars) can transmit qualia of the kind we'd expect from introspection.

Cars pretty obviously transmit qualia in the sense of "your visual field changes when you drive them somewhere" and "there's vibration you feel through the seat while they're on", but there's definitely something incomplete in that explanation. You can walk away from a car, after all. On the other hand, lobotomies and (color)blindness illustrate that the brain is also not inviolable, and that qualia can be changed or even removed without also removing their recipient entirely.

(If an afterlife exists, qualia don't even need a brain, but a computationalist might point out that the types of signal, and thus qualia, received without a brain are going to be very, very limited.)

Answer 7

You may have some questionable ideas about how science and philosophy works, and what the computational theory of mind is.

The computational theory of mind states that the human mind is an information processing system and that cognition and consciousness together are a form of computation. Given the evidence we have, this seems to be the most likely explanation for consciousness, of the explanations available (of which there aren't all that many).

Evolution seems like the most likely explanation given the evidence of the genetic variation between different generations of organisms.

General relativity seems like the most likely explanation given the evidence of the movement of physical objects.

And so on.

That's pretty much what scientific theories are: the most likely explanation for some evidence.

"When I'm thinking, I generally don't experience anything that I would call algorithmic unless I'm consciously executing an algorithm in my mind"

The computational theory of mind states that even your emotions would be computational, regardless of whether or not it feels like they are.

"It's not enough to show cognitive studies about how certain chemicals are associated with certain emotions, or activity in certain parts of the brain are associated with certain intentions, or evolution can justify certain desires"

Those all form part of the evidence supporting the computational theory of mind, and do seem like sufficient evidence to accept the theory.

Which other theory better explains those things? If there is no such theory, we should provisionally accept the computational theory of mind.

"Those observations/theories still leave an explanatory gap between the physical and the mental"

Much of science leaves explanatory gaps.

General relativity, for example, proposes that every physical object warps spacetime around it, which results in objects moving towards one another. But we don't know what exactly spacetime is (if it is anything more than just space, and time), or why it gets warped by objects.

This is just how we discover knowledge: we get more and more knowledge, one bit at a time. There will always be some gaps between the bits we know.

"I'm not asking about computational models of the brain; I'm asking about computational models of the mind"

I'm not aware of any computational models of the mind that's distinct from the brain.

We have no clear or definitive evidence to suggest that there is some element of human consciousness that's distinct from the brain. People making claims of such things also typically claim that such things are outside the realm of scientific enquiry, and that the lack of evidence is a feature rather than a bug.

"The mind is an independent sort of substance that just follows along" - this seems to add an unnecessary hypothesis. You might as well just... not have a hypothesis that the mind exists, and you wouldn't lose any explanatory power.

Mind-body dualism is a philosophical idea that's getting less and less plausible as we discover more and more about reality and the brain.

"Exactly what is the value of those theories?"

A scientific theory is a statement about what is (to the best of our knowledge) true. It's not a statement about what information is useful or has value. Whether or not a scientific theory is useful is a different question entirely.

Although the implications of our consciousness being nothing more than firing neurons would (or should) be quite significant to religion, as much of religion relies on some sort of "soul" that exists outside the body.

It could also be applicable to the ethics of artificial intelligence at some point in the (likely distant, or only hypothetical) future, because one could argue that humans deserve human rights due to our consciousness, and a machine with a functionally identical consciousness should therefore be given similar rights.

This would also suggest that the philosophical concept of free will doesn't exist, which may have implications for how we think about justice. If someone's actions is nothing more than a product of their biology and environment, it would make sense to focus on rehabilitation and what would minimise future harm, rather than punishment.

It may also have consequences for the diagnosis and treatment of mental illnesses and other mental conditions. If we accept that some harmful impulses are part of someone else, and they have no choice in the matter, medical treatment that would suppress those impulses (or just therapy) may be more appealing than if we were to consider those impulses to be them freely choosing to act in that way (even if suppression-based medical treatment doesn't have the best history).

Answer 8

What is the connection from mental states comparable to how statistical mechanics connects molecules to the gas laws or how natural selection connects random mutation to apparent design?

The brain has far more parameters than an ideal gas or a model in population genetics, so instead of something analytic, analogous to the equipartition theorem or Price equation, you might only ever have something numerical or black-box, as in machine learning.

Where is the theory about the relationship between computational state X causes desire Y or emotion Z or intention W?

You want computationalism to explain behavioural states, as well as mental states. Gilbert Ryle's behaviourism aside, you'll say there's more to having a mind than acting like a p-zombie. To a computationalist, what more there is concerns which computations are which mental states, which is a separate task from connecting computations to physics, which is what I'll focus on.

As a warm-up, we claim the audiovisual output of my phone, and its response to external stimuli such as someone else ringing it, are explicable in terms of computation. This explains how apps written in code let it do new things, how video playback gets every pixel right, and so on. It also implies a property dualism hardware/software distinction analogous to the brain/mind distinction. Physical laws don't explain the logic of programming languages, but they explain how particles ensure the phone obeys the code.

An expert can say code converts as thus to binary, which manipulates the semiconductor as thus. To do as much with the brain, we'd have to e.g. know the physics of its running in binary or another digital base or analog. Here are analogies to programming we'd need:

• What are the software's standard data types and how are they stored and edited?
• What chunking of concepts makes sense to describe the programming in terms of high-level languages? (The implications for bit manipulation of Python -> C -> Assembly -> Machine code -> Binary might not be close to what the brain does, but if it is we could describe the behaviour in something closer to English than binary.)

It would be nice if we could "read the source code" (like this joke?), but we can only look at how the brain forms and behaves. Our phones ultimately run on physics to avoid an infinite regress of the form, "to run software A it needs to first run software B". Maybe neurology can find similar facts.