New paper in Synthese: 'Panpsychism and AI Consciousness'

(Update: a read-only preprint can be viewed here)

 

This is just a quick note that I have a new paper with Corey J. Maley (University of Kansas), 'Panpsychism and A.I. Consciousness', out in Synthese. We use the distinction between digital and analog computing to argue that if panpsychism about consciousness is true, then digital A.I. may be incapable of realizing any kind of coherent first-personal experience--and that this also challenges Giuilo Tononi's Integrated Information Theory of consciousness.

 

Our argument also informs a partner paper that my former student Micah Summers (Florida State University) and I have forthcoming in the Australasian Journal of Philosophy, 'Two New Doubts About Simulation Hypotheses', where we argue that if panpsychism or panqualityism about consciousness is true, then (contra Chalmers) virtual realities aren't fully real, and (contra Bostrom) the probability that we live in a simulation is immeasurably small.

Comments

[postdoc]

Seems to me that whatever it is about our brains that allows for the magic of consciousness may not apply to digital A.I., regardless of panpsychism. The analog and chemical natures of our brains may be necessary elements to creating consciousness even if most stuff isn't conscious.

Maybe you touch on this? I can't read the article, as you didn't upload a pre-print version.

[Marcus Arvan]

Hey postdoc - A read-only preprint is available here: https://rdcu.be/cOIMD

Indeed, we do discuss that question (see pp. 9-11). In brief, we allow that analogous issues may arise for some other theories of consciousness, but explain why panpsychism (specifically, micropsychism) is the only theory of consciousness that clearly and unambiguously has these implications.

[postdoc]

It seems to me that consciousness is somehow created by stuff our brains do, and it seems to me that for many things the kind of stuff used matters. For example, you cannot create electricity using just any materials. The physical characteristics of the materials are important for the higher level function, e.g. generator. Merely simulating a generator using plastic (or on a computer), even if it looks the same, will not produce any electricity.

Now this example may make one think of micropsychism--are not electrons precursors of electricity? What about colors though? Not just any substance is silver or blue or... One requires the right chemicals to get the right colors, but it is a lot less obvious that we can talk of microphysical color precursors. In other words, I can hold that only certain materials or chemicals can be certain colors even though colors are primitive and have no precursors or mechanical-like explanation (particles doing this or that).

So, likewise, I can hold that probably consciousness depends on the materials, as so much seems to, but that there are no microphysical precursors. So, I guess I'm just saying that there is another theory that may rule out A.I. consciousness--a primitive realism, whereby consciousness supervenes on micro-electro-chemical processes or something like this. I guess to me this theory is the most plausible, but I did not see it mentioned. Totally possible I missed it, as I was just skimming.

A question. I can't copy from the paper, but in it you say that micropsychism unambiguously entails that it if it is true, then digital A.I. may not be capable of realizing consciousness. I have a hard time with this sentence. Unambiguously entails that something may happen?? Could you explain what you mean here?

[Marcus Arvan]

Hi postdoc: Thanks for the follow-up. I'm a bit unclear on a couple of things, however.

You say, "So, I guess I'm just saying that there is another theory that may rule out A.I. consciousness--a primitive realism, whereby consciousness supervenes on micro-electro-chemical processes or something like this."

What do you mean here by 'another theory'? The theory you seem to be alluding to here just is the one that we focus on in the paper: micropsychism (as this just is the view that consciousness is primitive and superveves on micro-physics). Do you perhaps mean panqualityism, the view that qualities themselves (redness, greenness, etc.) pervade the microphysical world? If so, Micah Summers and I discuss that in our related paper linked to above--and, as we point out, that view has basically the same implications for consciousness as micropsychism. So, Corey and I would simply say that our argument extends straightforwardly to panqualityism (I think we didn't discuss it because it doesn't appear to be a very common view, and because we really didn't have room to do so at any length).

You write: "you say that micropsychism unambiguously entails that it if it is true, then digital A.I. may not be capable of realizing consciousness. I have a hard time with this sentence. Unambiguously entails that something may happen?? Could you explain what you mean here?"

I'm not exactly sure what you think is unclear here. We argue that micropsychism entails that for a mechanism (e.g. the brain or computer) to generate (or realize) a coherent macrophenomenal manifold (like the coherent first-personal experiences that you and I have), that mechanism must (somehow) manipulate fundamental microphysics in an analog manner. The claim then is that this is precisely what digital computers do not (and cannot) do--so, micropsychism entails that digital computers cannot as a matter of fact realize coherent macroconsciousness.

[postdoc]

In your paper you define micropsychism differently, as being the view that phenomenal consciousness or its precursors exist at the microphysical level of reality.

I don't know what panqualityism is as defined. Most people accept that red and green are pervasive in the world.

I looked it up in the SEP and it says the view is this:

"According to panqualityism the protophenomenal properties are unexperienced qualities. Our conscious experience is filled with experienced qualities, e.g., those phenomenal qualities involved in seeing colour or feeling pain. Panqualityists believe that such qualities are only contingently experienced, and that in basic matter they exist unexperienced."

This view strikes me as making some serious category mistakes and isn't at all coherent.

The view I am talking about is simply the view that consciousness is primitive and supervenes on electro-chemical processes. Whether it can supervene on merely electrical process, as in a modern computer, is an open question. But there are obvious reasons to be skeptical due to the fact that not just anything can supervene on anything.

What's weird about the unclear passage is saying that something unambiguously entails that something MAY be the case. I'm sure it's clear if you read the entire paper. I was just skimming out of interest and in order to answer my immediate questions.

It's an interesting topic. Thanks for contributing to it.

[Marcus Arvan]

Hey postdoc: the view that you suggest here—that consciousness supervenes on electrochemical processes (but not micro physics more broadly)—seems to me to amount to a kind of type identity-theory, which we do discuss in the paper.

The “weird” feature of the passage you refer to (our claiming that only micropsychism unambiguously entails that digital AI *may* be incapable of coherent consciousness) is explained a couple of times in the paper. We claim that if our interpretation of the analog-digital issues is correct, then micropsychism unambiguously entails that digital AI can’t be conscious. The “may” enters into the picture insofar as it is a matter of ongoing debate whether our conception of the analog-digital distinction is correct—though we think our conclusion probably follows on other ways of understanding the distinction too. We just can’t defend all that in the paper—hence the “may” claim.

Hope this helps, and glad you found the paper and broader topic of interest!

[postdoc]

The view is most certainly not type-identity theory. The view is that phenomenal properties are not reducible to physical properties but supervene on them. It's a view in metaphysics that would be called primitivism or common sense realism or something like this.

Anyway, it's interesting to note that this view may also preclude digital A.I., depending on what the supervenience relations in fact are.

I'm not making any point that is a problem for your paper. But if we combine primitivism with a plausible thesis we can reject A.I consciousness as implausible.

Plausible thesis. The material matters. Often powers come with certain materials. For example, we know that with colors that merely simulating a paint's microphysical structure like on a computer will not produce the relevant color. We can see this for other things as well. You can simulate an atom all you want on a computer but will never generate the relevant atomic forces. For whatever reason subatomic particles have certain forces associated with them. Likewise for whatever reason (maybe none at all) certain colors seem to be associated with certain micro-physical processes/structures. So, there is every reason to be suspicious that a computer simulating a brain would be conscious.

Corollary: There are elements of our brains that seem kind of similar to a computer, neurons firing or not and sending electricity around. However, our brains maybe should be more understood as like a generator. The electricity moving around is important but the stuff it's made out of is too, because a brain isn't simply processing information for people; it's giving rise to consciousness.

[Marcus Arvan]

Hey postdoc, I guess I'm still missing something, but that sounds a lot like micropsychism or panqualitysm to me and more or less the argument we advance in the paper!

[postdoc]

I was not abundantly clear before. This view I am calling primitivism is not a kind panpsychism. The view is that consciousness is something created by the brain that is not identical to anything physical in the brain or any physical process or anything brainy (for a lack of a better word). The analogy is with color. The color primitivist holds that colors are in a sense created by the "micro-stuff" doing this or that or being in such a way or another, but he holds that colors aren't any of those things.

Perhaps the talk of color is confusing. The color primitivist might hold that atoms are colored or at least some atoms. Certainly it's plausible that some molecules are colored. But the primitivist need not say that the colors of macro objects are made up out of the colors of micro-objects. This is a cool idea but I've never heard it mentioned before. I very much doubt it's true. Not even simple paints combine in a way that can be said to be constitutive, as if blackness consists of Cyan, Yellow, and Magenta. And with lights, colors just don't add up this way at all. What matters with objects is how they reflect light but that's also very messy due to the phenomena of metamers, whereby different spectral power distributions can produce the same color experiences. So, the common understanding of color primitivists is that colors supervene in a messy way that is non-constitutive.

So, again, the idea is that consciousness is created by brains kind of like how electricity is created by a generator. If you insist of thinking of electrons as precursors to electricity (would it be a priori?), you can think in terms of colors instead. Gold is golden colored but there is no gold precursor in a gold atom or in the protons and neutrons. Certain things look gold, not only gold itself. Why is this so? Probably no reason. So, again, the idea is that consciousness is created by brains kind of like how colors are created by micro-stuff being certain ways or doing certain things. Like colors, there is no reason to hold that the micro-stuff is necessarily colored--gold is gold colored even though none of its atomic parts are plausibly colored gold or colored at all. So, looking at consciousness again, we need not hold that the micro-stuff doing this or that is conscious or proto-conscious, while holding that the micro-stuff doing this or that creates consciousness. However, we may think it's more than just processes or functions that matter, because that's just how it is with so many things.

With a computer all that matters is that you manipulate symbols that you've ascribed meaning to in a logical way. A computer can be realized using all sorts of materials because there are all sorts of things that we can use as symbols and all sorts of ways of manipulating them. Computing is not a basic process in this sense. But whatever the brain is doing it can't exactly be said to be computing because we have not ascribed any meaning to any symbols and our brains certainly can't do this. The brain is much more correctly thought of a generating thoughts and ideas etc. And once you start thinking of it as a generator instead of a computer, it becomes plausible to think that the substances/materials matter.

Anyway, that's the best I've got right now! Good luck with your research!

[postdoc]

I read some more about panqualityism.

Color primitivists might hold that colors are as we see them, so the colors are the phenomenal colors. The primitivist might say that these phenomenal colors define the phenomenal characters of our experiences in an ostensive fashion. One might say this about other secondary properties too like sounds and smells.

However, the secondary property primitivist is not making a panqualityist claim. That would require a lot of additional baggage. For one, he does not hold that ALL mental phenomenon have external world counterparts--like how the colors of objects define our experiences of colors. For example, feelings and emotions would be different. For feelings there is not an appearance/reality distinction. If I feel pain in my foot, then there is a pain in a my foot; it doesn't matter whether there is any damage etc. With colors it's different. My foot can appear red but not be red; the appearance could be an illusion. So, colors are objective but pains are purely subjective.

Additional baggage is this combination issue. It seems in some way Coleman wants to explain consciousness in terms of phenomenal properties that exist in fundamental matter or something like this. The primitivist has no desire to ground consciousness in fundamental matter or anything. The phenomenal characters of color experiences are defined by external world phenomenal colors (a red experience is an experience of THAT). However, there is no attempt at a reduction here, no attempt to reduce the subjective to the objective. Clearly experiences of colors are not colors or anything like that, and their characters don't depend on their being colors.

Primitivism about consciousness should be understood similarly to primitivism about colors and smells etc. It's the common sense view or something like that. It doesn't want to reduce consciousness or ground it in fundamental matter. It doesn't want to build it up out of fundamental conscious properties or proto whatevers. What's the common sense view? Well, it's something like this: the vast majority of stuff is not conscious. It takes a kind of machine, a.ka. brains, to create consciousness. What the brains due is they give rise to a new kind of thing, a person we might call it, a thinking thing, and how this happens is not explainable in a way that is intelligible. It just happens, as does everything. All contingent truths are mysterious. Why should should anything to anything?

Anyway, something like that probably isn't very popular but views like that exist in the color literature. It surprises me that they seem foreign to you.

[Marcus Arvan]

Hey postdoc: thanks for clarifying - it's very helpful. I'm not terribly well-versed in the color literature, which probably explains some of this. But quite frankly, the view seems problematically mysterious to me, in ways that seem worse than panpsychism or panqualityism (which are already seem notoriously mysterious to some people, though not to me).

For here's the crucial part to me. You write: "What the brains due is they give rise to a new kind of thing, a person we might call it, a thinking thing, and how this happens is not explainable in a way that is intelligible. It just happens, as does everything. All contingent truths are mysterious. Why should should anything to anything?"

The nice thing about panpsychism, panqualityism, and functionalism for that matter, is that they at least provide intelligible explanations for how conscious minds are what they are. Panpsychism holds that just like a painting is built out of smaller constituents, so too must consciousness be--and the only way to do it is for phenomenal consciousness to be primitive in much the same way that mass, charge, electrons, gravity, and the strong and weak nuclear forces are primitive at a microlevel. All of this seems intelligible and *plausible*, to the extent that physicalism and functionalism seem to leave phenomenal qualities out--and panqualityism just extends the same line of thought further to all of nature, holding that qualities are in all things. Conversely, if one rejects this picture, functionalism is the obvious alternative--and it too at least provides an intelligible-looking explanation for the mind.

In contrast, saying "What the brains due is they give rise to a new kind of thing, a person we might call it, a thinking thing, and how this happens is not explainable in a way that is intelligible. It just happens, as does everything", just seems to me to amount to a kind of McGinn-ian Mysterianism. If there are *explanations* (such as panpsychism) that work better, why not prefer them? Yes, there's an obvious sense in which all contingent truths are mysterious (it's mysterious why the gravitational constant is what it is, among other things)--but what good science and philosophy do is provide explanations where they are available. For instance, we don't just throw our hands up and say "Well, cells just *somehow* give rise to life." We have organic chemistry and biology to explain how they do so, and in a way that is illuminating, reducing them to particular mechanisms and fundamental constituents.

Long story short: these seem to me like the kind of things that good theories do, and it seems to me like panpsychism, panqualityism, and functionalism all arguably do so better than the alternative being floated here. Then again, as Corey and I speculate in the paper, maybe some kind of future science of consciousness will be able to pin down the relevant correlations between what goes on in fundamental physics and conscious experience, such that we might have *some* idea which of the above hypotheses is best favored by our evidence--and who knows: maybe it could be your (interesting) proposal! I'm not at all sure how a science of consciousness might accomplish this kind of thing, but I don't like to arrive at skeptical conclusions prematurely, as science has a pretty good history of surprising us.

Anyway, thanks for the fun conversation. Your proposal is (I think) pretty cool and worth thinking about, and I agree, the Universe is suffused with mysteries. The real trick is to figure out which ones admit of explanations and which one's don't! :)

[postdoc]

Hey Marcus,

I'm glad I was finally able to explain the view to you in a way that made sense!

I'm not going to argue for the view here. There could be lots of different arguments for it.

1. The other views all fail (popular in the color lit). For consciousness for example, panpsychism seems to struggle with the recombination problem. As a primitivist, I think this problem is intractable.

2. None of the other views make sense. This is perhaps proto-wittgensteinian. For example, one might say there are conceptual problems with talking of proto consciousness or attributing conscious properties to inanimate things. I mentioned before how I think panqualityism is conceptually flawed.

3. Primitivism is the common sense view and none of the arguments against it work, so it is the view we should hold.

Those are just some really brief argument sketches.

A quick reply to your argument against primitivism to the effect that it's mysterious. Yes, primitivism says that there is no philosophical explanation for consciousness--no explanation whereby consciousness is intelligible in any strong sense of the word. But the primitivist says these kinds of explanations are not available to us.

You cite biology. Now I'm not a biologist and know little about the field. But there is nothing in primitivism that precludes biology as a science. It's better if we look at physics where I know slightly more. Primitivism about physics doesn't say that we can't propose atoms and parts of atoms and powers for those parts and try to explain things using these theories. What primitivism says is that these theories don't ultimately provide the kinds of deep intelligible explanations people want. Eventually you end up with brute facts basically, whereby you just have to say things like this particle just does this or that where there is no reason for why this is the case. So our attempt to make the world intelligible fundamentally rests on a 'no reason' claim and so fundamentally always fails. This is not to say that the theories are not useful or even true. It's just to say that they don't really explain in a deep way.

The panpsychist wants to push the brute facts of consciousness into external, atomic particles or something. I guess this is so we can unify physics with consciousness and have a fundamental theory that explains everything. But it's entirely possible the brute facts are at a much higher level. Sure there are practical reasons to prefer a fundamental theory that explains everything from the movements of photons to people's feelings. But I think presently that it doesn't look like it's going to happen. That's why I think primitivism is closer to the common sense or intuitive view. Either way, panpsychism cannot make consciousness intelligible; it just pushes the problem somewhere else. So, I don't find this argument against primitivism that it makes the world mysterious particular compelling.

[Grant Castillou]

It's becoming clear that with all the brain and consciousness theories out there, the proof will be in the pudding. By this I mean, can any particular theory be used to create a human adult level conscious machine. My bet is on the late Gerald Edelman's Extended Theory of Neuronal Group Selection. The lead group in robotics based on this theory is the Neurorobotics Lab at UC at Irvine. Dr. Edelman distinguished between primary consciousness, which came first in evolution, and that humans share with other conscious animals, and higher order consciousness, which came to only humans with the acquisition of language. A machine with primary consciousness will probably have to come first.

The thing I find special about the TNGS is the Darwin series of automata created at the Neurosciences Institute by Dr. Edelman and his colleagues in the 1990's and 2000's. These machines perform in the real world, not in a restricted simulated world, and display convincing physical behavior indicative of higher psychological functions necessary for consciousness, such as perceptual categorization, memory, and learning. They are based on realistic models of the parts of the biological brain that the theory claims subserve these functions. The extended TNGS allows for the emergence of consciousness based only on further evolutionary development of the brain areas responsible for these functions, in a parsimonious way. No other research I've encountered is anywhere near as convincing.

I post because on almost every video and article about the brain and consciousness that I encounter, the attitude seems to be that we still know next to nothing about how the brain and consciousness work; that there's lots of data but no unifying theory. I believe the extended TNGS is that theory. My motivation is to keep that theory in front of the public. And obviously, I consider it the route to a truly conscious machine, primary and higher-order.

My advice to people who want to create a conscious machine is to seriously ground themselves in the extended TNGS and the Darwin automata first, and proceed from there, by applying to Jeff Krichmar's lab at UC Irvine, possibly. Dr. Edelman's roadmap to a conscious machine is at https://arxiv.org/abs/2105.10461

[Consciousness]

Postdoc: You claim: “So, again, the idea is that consciousness is created by brains kind of like how electricity is created by a generator.”

The truth of that claim may be dependent on the truth (of the assumption) that consciousness a physical entity. But is consciousness a physical entity? If not, then can a physical entity like the brain create consciousness if it’s non-physical?