Repairing Yudkowsky's anti-zombie argument

by [anonymous] 10 min read5th Oct 201179 comments


Eliezer Yudkowsky argues with David Chalmers here on the subject of “philosophical zombies”. I submit that, although Yudkowsky’s position on this question is correct, his argument fails to establish what he claims it to.

To summarise Yudkowsky and Chalmers’s argument:

1. Both Yudkowsky and Chalmers agree that humans possess “qualia”.
2. Chalmers argues that a superintelligent being which somewhow knew the positions of all particles in a large region of the Universe would need to be told as an additional fact that any humans (or other minds possessing qualia) in this region of space possess qualia – it could not deduce this from mere perfect physical knowledge of their constituent particles. Therefore, qualia are in some sense extra-physical.
3. Yudkowsky argues that such a being would notice that humans discuss at length the fact that they possess qualia, and their internal narratives also represent this fact. It is extraordinarily improbable that beings would behave in this manner if they did not actually possess qualia. Therefore an omniscient being would conclude that it is extremely likely that humans possess qualia. Therefore, qualia are not extra-physical.

My objection to Yudkowsky’s argument is that it is not enough merely to demonstrate that the omniscient being would find it extremely likely that humans possess qualia. Probability is a state of partial information; therefore unless the being is certain that humans possess qualia, it is not in fact omniscient regarding this region of the Universe despite the fact that it is postulated to possess perfect physical knowledge about it.

I expect that some Lesswrongians may object to this on account of the fact that 1 and 0 are not probabilities. However, the thought experiment postulates an omniscient being that possesses perfect knowledge about the physical state of a region of the Universe, therefore in the thought experiment absolute certainty is defined to be possible. If this is objectionable*, then the entire argument is badly posed including Yudkowsky’s contribution.

Additionally (although superfluously), it seems that the thought experiment should generalise to any possible configuration of particles in a region of the Universe, since we are trying to prove that qualia are not extra-physical under any circumstances, and a proof of this should not rely on contingent features of the qualia-experiencing beings under consideration. Therefore let us suppose that due to a miracle of quantum tunnelling, the only qualia-experiencing being in the region of space in question is a newborn human infant (I presume wide agreement upon the fact that such an infant does in fact possess qualia.) Is it still the case that the omniscient being can deduce, from the infant’s mental behaviours, the extreme likelihood of its possessing qualia? After all, it doesn’t write philosophy papers and may not even have an internal narrative.

My own solution to the zombie problem is that it is a restatement of the Mary’s room problem. In the zombie thought experiment we are dealing with a mind that has perfect knowledge of a physical human brain (or any brain that produces qualia). Since the Universe’s computational accuracy appears to be infinite, in order for the mind to be omniscient about a human brain it must be running the human brain’s quark-level computations within its own mind; any approximate computation would yield imperfect predictions. In the act of running this computation, the brain’s qualia are generated, if (as we have assumed) the brain in question experiences qualia. Therefore the omniscient mind is fully aware of all of the qualia that are experienced within the volume of the Universe about which it has perfect knowledge.

It is legitimate for us to believe with extremely high probability that the computations occurring in a brain are causally related to the qualia that it produces, for the reasons that Yudkowsky has given; therefore, it is fair for us to state (as I did in the preceding paragraph) with extremely high probability that when it runs a brain’s computations the omniscient mind will experience the same qualia that the brain’s original owner does. The distinction is that we can be extremely confident (by Yudkowsky’s reasoning) that the omniscient mind will itself be certain (by my reasoning) about the existence of qualia within the volume of the Universe about which it has perfect knowledge – whereas if one is trying to prove that qualia are not extra-physical, it insufficient to argue (as Yudkowsky did) that the omniscient mind will itself only be extremely confident about the existence of qualia within the volume of the Universe about which it has perfect knowledge.

There is an objection to the above argument that I would expect readers to suggest. The objection is that I have misinterpreted Yudkowsky’s argument, and in fact my summary of Yudkowsky and Chalmers’s argument should read as follows:

3. Yudkowsky argues that such a being would notice that humans discuss at length the fact that they possess qualia, and their internal narratives also represent this fact. It would use its perfect knowledge of their mental processes to investigate the chain of reasoning that leads humans to refer to themselves as being “aware” and possessing “qualia”. It would thereby discover the cause of their discussing these things, which is extremely likely to provide it with a reduction of the qualia concept. Therefore the omniscient mind will with extremely high probability obtain for itself perfect certainty that the qualia-experiencing beings within the region of the Universe about which it has perfect knowledge do in fact experience qualia.

If this argument was valid, it would have exactly the same outcome as my own conclusion (and a different outcome to Yudkowsky’s argument as I summarised it earlier): we can be extremely confident that the omniscient mind will itself be certain about the existence of qualia within the volume of the Universe about which it has perfect knowledge (rather than: the omniscient mind will be extremely confident about the existence of qualia within the volume of the Universe about which it has perfect knowledge).

I don’t interpret Yudkowsky’s argument in this way, but since it is a closely related meaning I cannot be very sure that he does not intend the above. In any case, I believe that this argument begs the question.

The revised argument assumes that the qualia concept is almost certainly reducible. However, doubt regarding this appears to be the entire motivation for Yudkowsky and Chalmers’s debate. If Yudkowsky were to regard it is a given that qualia are reducible, then why not replace his 6,600 word post with the following: the thesis of reductionism is proven beyond reasonable doubt to be true. Therefore qualia, like other phenomena that we observe, must be reducible to the level of quarks. Therefore if a mind possesses perfect knowledge about a region of the Universe at the level of quarks, it necessarily understands qualia and recognises their existence because qualia are merely higher-order phenomena composed of quarks. QED.

Since Yudkowsky did not do that, I presume that he does not believe that the reducibility of qualia can be taken for granted. Regardless of what he believes, I would nonetheless criticise the revised argument on the basis that qualia should not be assumed to be reducible.

An observation about qualia: they do not appear to be susceptible to definition. All of the definitions of qualia that I have encountered have been either nonsensical, or liable to be interpreted (à la Dennett) as mere computational properties of the brain.  And definition, done properly, is in fact the same act as philosophical reduction.

For example, qualia might be defined as “subjective qualities of experience”. Subjective in this context means the same thing as incommunicable. A definition of something as essentially incommunicable is tantamout to defining it as indefinable, which is nonsense. “The inner listener” and similar attempted definitions are interpreted by Dennett in the sense of the brain’s having a parallel computational structure – not the intended referent at all. And the “mysterious redness of red” could either be interpreted as a reference to the nature of redness as a derived or computational property of objects rather than a fundamental property, or as gibberish (since there are no inherently mysterious phenomena).

Nonetheless, since (I believe with extremely high probability) we all possess qualia, we are able to figure out the intended referent of words like “qualia” and “consciousness” (at least before the word consciousness was philosophically co-opted by Dennett). We are aware of the existence of one indefinable concept, and we can shore up our mutual recognition of terms that refer to it by discussing the apparent relationship between the properties of this concept and the state of our physical brains, which seem perfectly reducible to quarks and susceptible to definition.

In this post, Yudkowsky writes:

Suppose that a 747 had a fundamental physical existence apart from the quarks making up the 747.

What experimental observations would you expect to make, if you found yourself in such a universe?

If you can't come up with a good answer to that, it's not observation that's ruling out "non-reductionist" beliefs, but a priori logical incoherence. If you can't say what predictions the "non-reductionist" model makes, how can you say that experimental evidence rules it out?

My thesis is that non-reductionism is a confusion; and once you realize that an idea is a confusion, it becomes a tad difficult to envision what the universe would look like if the confusion were true.

May I offer, as a suggestion of what an authentic irreducible concept looks like, qualia?

There are two reasons why it is rational to hold a completely reductionist view of the Universe. Firstly, reductionism is a historically successful means of explaining things and solving problems, and always defeats non-reductionism. Secondly, non-reductionism is a priori logically incoherent in the sense that Yudkowsky describes.

However, what if the qualia concept is actually a counter-example to both of these? Of course it’s far too early to conclude that reductionist means have failed to explain qualia – we need to learn more about the brain first. But is it not also reasonable to suggest that our experience of qualia is exactly what we would expect the universe to look like, if irreducible phenomena were to exist? We almost all agree that qualia exist, and in fact my belief in the existence of qualia is the last belief of which I can imagine anyone dissuading me, yet we have thus far (despite much philosophical inquiry and a fair amount of neuroscience) been incapable of reducing the concept to the slightest extent. By comparison concepts such as shouldness and couldness, which are complex computational properties of the brain, have already been reduced a level or two within our map of reality.

To summarise, I argue that neither version of Yudkowsky’s anti-zombie argument, as I interpret it, is sound. The first version is unsound because it fails to demonstrate that an omniscient mind possesses the same level of confidence in its judgements about qualia as it does about the physical Universe, and the second version is unsound because it presumes that qualia are reducible, which is unwarranted. I also propose my own anti-zombie argument, which I believe demonstrates that qualia are not in fact extra-physical (although they may yet be irreducible).

*I see the premise of an agent with perfect knowledge as a helpful simplification in the thought experiment. But I believe that it could be exchanged for a fallible superintelligence without the conclusions changing. The contrast between perfect certainty and infinitesimal uncertainty would be replaced by minor uncertainty and infinitesimally increased uncertainty. The problem still exists with Yudkowsky’s argument that if qualia are not extra-physical, then uncertainty about qualia should be no greater than total uncertainty regarding the physical configuration of reality.



It seems to me that I should have paid more attention to (i.e. re-read) Eliezer’s post “the generalised anti-zombie principle” before writing this article. Because in it he states:

“Consciousness, whatever it may be - a substance, a process, a name for a confusion - is not epiphenomenal; your mind can catch the inner listener in the act of listening, and say so out loud. The fact that I have typed this paragraph would at least seem to refute the idea that consciousness has no experimentally detectable consequences.”


Could we define the word "consciousness" to mean "whatever actually makes humans talk about 'consciousness'"? This would have the powerful advantage of guaranteeing that there is at least one real fact named by the word "consciousness". Even if our belief in consciousness is a confusion, "consciousness" would name the cognitive architecture that generated the confusion.

So Vladimir Nesov appears to be correct in that I was wrong to assume that Yudkowsky was necessarily referring to the same kind of "qualia" or "consciousness" that Chalmers was.

Rather than further delving into guesses about Yudkowsky’s intentions, I’ll just attempt to clarify my conclusions in general:

1. I notice that most humans regard themselves as possessing “qualia”. Formerly this concept might have been known as “consciousness”, but at this point that term is ambiguous. No-one seems able to define qualia, despite their insistence that it is a real concept. Qualia are considered to be related to brain processes in some way.

2. Writers such as Dennett believe that they can (or have already) described approximately the physical process by which the brain computes an internal narrative containing statements such as “I am aware that I am aware”, and similar references that many would understand as referring to qualia. Specifically, according to wikipedia: “Dennett's view of consciousness is that it is the apparently serial account for the brain's underlying parallelism.”

3. As I believe I have demonstrated, if we assume that qualia exist then we are bound to believe with extremely high probability that a being that has perfect knowledge of a physical brain is fully aware of the qualia that are being produced in this brain. It has no more uncertainty about qualia than it has about the physical brain states. I also believe that Yudkowsky’s argument did not prove this, for reasons already stated.

4. “3” does not allow us to conclude that “qualia” are the cause of our making statements such as “I am aware that I am aware”. These statements can be explained according to Dennett’s eliminative materialist view, in which qualia are held not to exist on the basis that the concept is resolutely indefinable and therefore unreal. However although “qualia” are apparently indefinable, since a given human’s degree of belief in the existence of qualia is typically extremely high, many humans are unwilling to apply normal standards to the concept. One possibility, which I favour, is to expect that we will ultimately discover that the brain produces statements about consciousness for a reason along the lines of what Dennett describes, but that this physically instantiated computation serves a double role* both as a reducible causal explanation for why we talk about “qualia” and in producing the irreducible phenomenon of qualia by means of a psycho-physical bridging law. This would mean that qualia supervene upon brain states, but the causality operates only in one direction. In this scenario qualia are an irreducible phenomenon, and since our means of investigating the world necessarily involve the tool of reduction we cannot expect to understand qualia or any putative “psycho-physical bridging law”. Although this appears to be similar to irrational scientific confusions of the past, there is reason (as I have argued) to view qualia as a legitimate exception.

5. Alternatively, it may turn out to be the case that when we learn more about the brain, we will discover a reduction of qualia that satisfies everyone who claims to experience qualia. In this case “qualia” would indeed be the cause of our referring to ourselves as possessing qualia, and qualia would also be seen to be real. However, as I have discussed the concept of qualia possesses unusual features that make this seem somewhat unlikely.

*Think of it as being the algorithm, and “how the algorithm feels from inside”. In this case, since we are the algorithm, both are real!

So on reflection, it appears that I have quite a large disagreement with Yudkowsky:

It seems to me that Yudkowsky’s argument here does not prove what it is supposed to prove – that complete physical knowledge entails complete knowledge of qualia – and this is necessary in order for qualia not to be some entirely airy-fairy mystical concept that neither of us agrees with. The only way in which it could be interpreted to prove that, is if we simply assume that the concept of irreducible qualia is disallowed and he only means to refer to "reducible qualia" or "qualia-eliminative reducible consciousness" – this is both unwarranted, and sheds a rather confusing light on why Yudkowsky took 6,600 words to make his point.

If Yudkowsky is an eliminativist about “qualia”, i.e. he is actually attempting to prove that complete physical knowledge entails complete knowledge of reducible, qualia-eliminative consciousness (per Dennett), he is both in conflict with a majority of people’s extremely strong beliefs and he could have countered Chalmers with a very brief argument stating simply that talk about consciousness necessarily has a cause, and this cause must be a computation in the brain reducible to quarks.

If Yudkowsky is a reductionist regarding “qualia”, i.e. he is actually attempting to prove that complete physical knowledge entails complete knowledge of reducible qualia only, the question remains why qualia still appear to be entirely elusive to definition and again he could have countered Chalmers with a very brief argument stating simply that talk about qualia necessarily has a cause, and this cause must be a computation in the brain reducible to quarks.

If on the other hand Yudkowsky is ambivalent about these options, I don’t believe that it is sensible for him to refer under a single banner “consciousness” to the concepts “irreducible qualia” and “reductive explanation of qualia” and “reductive qualia-eliminative explanation of consciousness” because statements that are true of one are not true of the others. So he should stick to arguing about the consequences of one at a time.

For example, despite the fact that I agree that complete physical knowledge entails complete knowledge of irreducible qualia, I don’t see how this proves that irreducible qualia are the cause of our making statements such as “I am aware that I am aware”. So if Yudkowsky argues that complete physical knowledge entails complete knowledge of consciousness as though this proved that consciousness is the cause of our making statements such as “I am aware that I am aware” then he must only mean consciousness in the sense of reducible qualia or qualia-eliminative consciousness. And I have already discussed the problem with assuming these concepts at the expense of irreducible qualia, and asked the question why it takes 6,600 words to refute Chalmers under those assumptions.