Freaky Functionalist Fights Paper Computer

Do You Believe in Magic?

Non-AI art by non-AI Pete Mandik

Thanks for taking my bait, Eric. I was hoping to trick you into engaging in a conversation in which you attempted to answer the sincere questions I have about your view (which i legitimately find interesting), and you fell for it! MUWAHAHAHA!

One thing your response has helped me to see is that the stuff about “appropriateness” and the stuff about “magic” are actually two distinct and (unbeknownst to you) unrelated arguments. Regarding appropriateness, the point of that line of thought seems to be something that a functionalist not only is able to agree with, it’s pretty much just the whole point of functionalism. One standard way of defining functionalism—and this applies to all flavors of functionalism e.g. computational functionalism, neurofunctionalism, teleofunctionalism—is that what makes some state of a system of states a MENTAL state (e.g. a belief, a desire, a sensation of pain) is not any intrinsic property of the state (e.g. having 22 billion sodium ions in it), but a set of causal relations that state bears to other states in the system. So, for example, a functionalist will hold that what makes a state a desire to drink beer is one that will cause, in concert with a belief that there’s beer in the fridge, an intention to go open the fridge, and so on. An analogy to such relational specifications is the way machine parts have their functions. What makes some hunk of metal a brake pad isn’t the specific material that it is made of, but the causal relations that it bears to the brake rotor, wheel, etc. That same hunk of metal might be put into a different set of causal relations, and now it’s not a brake pad any more, but instead a paper weight, a door stop, the pendulum of an up-cycled steam punk grandfather clock, what have you. For mental states that are supposed to be information bearing, for example, pain sensations that presumably carry information about tissue damage and disturbance, the functionalist says that a state of activation in a nociceptor cell doesn’t carry that information because of its intrinsic properties, but because of its causal relations to the states of the rest of the system, the states that count as short term memory that there’s been damage in the ankle, the desire to prevent further ankle damage, the intention to get an ice pack from the freezer, etc. Any worries that arise about such functionalist theories being circular, since they seem to be defining mental states by relations to other mental states, has been put to rest by the logical technique of Ramsification, so-named after its inventor Frank Ramsey, and employed by functionalist philosopher David Lewis to give rise to what we now call Lewis-Ramsey functionalism. To make a long story short, each of the defined mental state types can receive a unique definition in terms of the other state types, and so all of the resulting definitions are provably non-circular. (Suzi Travis has blogged about Ramsification in her recent post about LLMs and Predictive Processing theories of consciousness.)

So, anyway the story so far: the best sense to be made of your “appropriateness” just is the core point that all functionalists would endorse: no part of a system of states is going to carry information for the rest of the system all by itself; it is only in virtue of its relations to the rest of the system that it counts as carrying information that your ankle is sprained. None of this is magic, unless it seems magical to say that a hunk of metal cannot all by itself be a brake pad.

So, what’s going on with your accusations that computational functionalists are committed to magic or the supernatural? As best as I can tell, your paper-computer argument is just another version of the “weird realization” objections to functionalism like Ned Block’s Chinese Nation argument, which is an heir of Leibniz’s “mill” argument from section 17 of his Monadology, and inspiration of Searle’s Chinese Room argument (which has some brilliant additional elements that make it more than just a “weird realization” objection). Such arguments are also close cousins of explanatory gap arguments against physicalism.

Weird realization arguments exploit the apparent multiple-realizability entailment of functionalist theories: Since it’s the relations between physical states that matter, it’s left open that some other physical states could play the same causal roles. This, of course, is a big IF. IF microchip states could play the same causal roles as neuron states, THEN you could replace my neurons with microchips and I would feel pain. IF Chinese people with walkie-talkies could play the same causal roles as my neurons, then a Pete consciousness could be realized by a suitably networked nation of China. IF a bunch of slips of paper being passed around could play the same causal roles…you can see where this is going. The functionalist doesn’t assume and doesn’t assert that, e.g. a bunch of papers shuffled COULD play such causal roles, they are just saying that IF they did, then they would, by the lights of their theory, give rise to pains, hopes, fears, the thought “I am Pete” etc.

Many functionalists have, of course, taken the “bait” and said, yeah, sure, ok. What’s the problem? Let’s say, yeah, the whole Chinese room system, with papers, book, reader, understands Chinese. Yeah, the whole Chinese nation communicating via walkietalkie instantiates a group mind that thinks it’s the red-blooded all-American hero, Pete Mandik. Yeah, the paper computer feels pain. Why not? What’s the problem?

Here, the anti-functionalist seems to not have anything to say beyond “it’s weird!”. It’s unintuitive! “I don’t see how a bunch of shuffled papers could give rise to a feeling of pain, or an experiencer of said pain, or an understander of the Mandarin word for pain, or whatever”.

Besides being an exceedingly weak argument, the problem with “it’s weird”-style objections to functionalism are their adjacency to explanatory gap aka hard problem arguments (and their cousins the zombie arguments, the Mary arguments) against physicalism. If the failure to see why or how a bunch of physical anythings can give rise to an experienced of pain is a good reason for thinking pain wouldn’t arise under those conditions, then it doesn’t matter which physical thing you plug into the theory. Explanatory gap arguments can’t be selectively wielded against just one particular physicalist theory (like a paper computer theory), they can be wielded against all physicalist theories. Suppose someone were to propose some allegedly nonfunctionalist physicalist theory, like, that a state of experiencing pain just was a fluctuation in a certain kind of electromagnetic field. An explanatory-gap anti physicalist could just say “that’s weird! I don’t see how a fluctuation in an electromagnetic field could be an experience of pain”. A zombie-conceived could say, for pretty much the same “reasons” “I can conceive to the field giving rise to no conscious experiences whatsoever. “You physicalists ALL believe in magic, as far as we can tell,” say the anti-physicalists.

One common retreat among physicalists who reject functionalism is to lean on the a posteriori/empirical nature of their theory. Such strategies are what Chalmers calls Type-B Materialism. They concede that there will be no transparent explanation of WHY their favorite physical state gives rise to an experience of pain, but they console themselves with the possibility of empirically proving that the particular physical state type goes along with the particular mental state type. It can’t be magic if it’s empirically supported!

This is cool as far as it goes. Something that happens as a matter of empirical regularity just is what it means for something to be non-magical or non-miraculous, as Hume point out in his “On Miracles”. But it cannot go so far as to cut any ice against the OG functionalists (who Chalmers lumps in with Type-A Materialism). As Chalmers himself will gladly allow, you can have hypothetical empirical evidence for any functionalist theory you want, via a “gradual replacement” scenario. Take someone who is clearly a conscious experiencer, and gradually replace each of their neurons with microchips, or Chinese people with walkie-talkies, or automatic paper shufflers. Whatever empirical evidence you have at the beginning of the process that the person has conscious experience is evidence that you will continue to have after all of their neurons have been replaced with, e.g. microchips, even if that person is you yourself!. Regarded strictly as empirical theories, as opposed to philosophical ones, functionalism has no disadvantage against its nonfunctional physicalist rivals. The differences only show up when we presume that what we are trying to do is purely philosophical—armchair analysis of what “conscious” means. In the armchair, the rules of the game include “if you can conceive it without conceiving it giving rise to consciousness, then you win and it loses”.

Bottom line: “it’s weird” objections only cut ice if you’re playing a non-empirical game. If you want to play an empirical game, you need to lay off the thought experiments. No one has built a paper computer and called it conscious. Probably no one ever will. And if they did, you’d have to apply strictly empirical reasons for rejecting their claims. “It’s weird!” won’t cut any ice.

Bonus Reading Recommendations:

Mandik, Pete. (2017). Robot Pain. In: Corns, J. (ed.). The Routledge Handbook of Philosophy of Pain. New York: Routledge. (pp. 200-209). https://philpapers.org/archive/MANRP-4.pdf

Mandik, Pete and Weisberg, Josh. (2008). Type-Q Materialism. In Chase Wrenn, (ed.), Naturalism, Reference, and Ontology: Essays in Honor of Roger F. Gibson, New York: Peter Lang Publishing. pp. 223-246. https://philpapers.org/archive/MANTM.pdf