I agree with the spirit of what Eliezer has written here: The possibility of p-zombies would suggest epiphenomenalism. If you believe in redness, you should expect it to be causally efficacious.
However, subjective redness manifestly exists; subjective redness manifestly does not exist in any physics known to us; yet the physics we have already have appears to be fantastically predictive in detail, and quite capable of producing intelligent behavior in principle.
The usual way out of this is to deny my second proposition, and say that redness is a type of brain state or property, and therefore as much a part of physics as any other material thing. But the elementary properties one finds in physics are of a very limited nature: quantitative, geometric, causal, probabilistic. How can piling up number and shape, even when glued together by causal relations, create color?
The answer is that it cannot, at least if you restrict yourself to logical, set-theoretic, and other relations truly intrinsic to arithmetic and geometry. This is why people become property dualists, or believers in “strong emergence”.
Another possibility canvassed by Chalmers is a reinterpretation of the mathematical formalism of physics, so that it is about something other than what it appears to be about, namely entities starkly devoid of the “secondary qualities” revealed in conscious perception. This is his panpsychism or panprotopsychism. It’s “pan-” because current physics has monistic tendencies, a homogeneity of kind among its fundamental entities; if any of them are mind-like or qualia-like, one might expect all of them to participate in or at least to approach that condition.
Given the problems of epiphenomenalism, I find this monistic approach more appealing. However, it seems that even those few philosophers who pursue this avenue are hindered by a rather crude notion of mind. People talk as if consciousness is nothing but sensation, and as if sensation is nothing but a pile of pixel-like elementary sensations, and as if these elementary qualia then just need to be identified with something physically elementary. There is a tendency, for example, which denies that there is any phenomenology of thinking, and no such thing as conscious perception of meaning; cognition is unconscious computation, and it’s just the raw feels of sensation which exist and constitute conscious experience.
But in reality, they are just the part of consciousness which is hardest to deny. Also, this mode of thought is least removed from billiard-ball materialism: an object is a heap of particles, a conscious experience is a heap of qualia. Alas, it is not so simple.
If anyone out there really wants to engage in phenomenology, I have a few recommendations. First, you must overcome the conceptual reflex which rejects the reality of “reifications” and “abstractions”. Err in the other direction, see what picture you build up, and then have a go at paring it back. My training curriculum is as follows. First, Chapter IX of Russell’s The Problems of Philosophy, which makes the case for the reality of properties and relations, as entities which exist just as much as “things” do. Then, ideally I would recommend Reinhardt Grossman’s The Categorial Structure of the World, for a mind-stretching example of a systematic ontology in which abstract entities exist just as much as concrete ones do, and in which the relationship between them is also analysed at length. But the book may be hard to obtain, so some other contemporary system of categories might have to do. Then, one should tackle Husserl and Kant, for epistemologically systematic ontology. And the final step—but by now I’m just describing my own research program—is to interpret the world of appearances which one has been describing as interior to a single entity, and to embed that in physics. As I remarked here last month, quantum entanglement suggests an ontology in which there are fundamental entities with arbitrarily many degrees of freedom. It is the one indication we have from physics that old-fashioned atomism (according to which all fundamental entities are simple and perpetually encapsulated) may be radically wrong.
Returning to the theme of zombies—in a monistic ontology like this, you can’t subtract the mental properties and leave the causal relations unchanged, so the critique of epiphenomenalism no longer holds. The best one can do is to imagine a possible world in which the causal laws are isomorphic, but the elementary states they relate are different: are not ‘mind-like’, are not states of consciousness. However—and this is relevant for AI, Friendly or otherwise—there is nothing to guarantee that every causal simulation of consciousness even in this world will itself be conscious, if consciousness is indeed this deeply grounded in “substance” rather than “algorithm”. For example—if conscious states are only ever states of a single elementary entity (e.g. a single tensor factor such as I mention in the previous link), then any distributed simulation of consciousness will not be conscious, even though it will exhibit all the same problem-solving capacities. (The correspondence between these two would begin to break down if they set about investigating their own physical constitution, which by hypothesis is different, so there is a limit to the duplication involved with this sort of “zombie”.)
Note 1: Chalmers does acknowledge the challenge of epiphenomenalism in The Conscious Mind, and discusses it at length. He proposes a number of ways in which the phenomenal properties might still be causally relevant, but does say that if forced to a choice between epiphenomenalism and eliminativism, we should prefer the former. (He develops the monistic alternative to property dualism in later papers.)
Note 2: A few commenters on this blog have tried the old dodge according to which “redness” is just a word. I would agree that it is a categorization whose scope is vague and varies with the individual. But is it not clear that the individual patches of color and shades of red to which it is applied do exist, and that they pose a challenge to the colorless ontology of physics, regardless of how we group them by name?
I agree totally. As someone who has read Chalmers entire book, it’s frustrating to read so many people misinterpreting his views. Chalmers is in no way committed to a strict epiphenomal dualism; with the Zombie argument Chalmers is merely demonstrating the intellectual bankruptcy of traditional reductionist dualism. He hesitantly endorses epiphenomenalism only because he is granting the materalist as many possible premises as he can to make his point. The materialist would seemingly want to hold to these facts:
1)The physical world is causally closed. All physical activity can be explained fully by the laws of physics.
2)The fundamental constituents are in no way conscious and exist as solely extrinsic entities in space and time.
3) The mind is fully reducible to physical activity.
Chalmers grants premise 1 and 2 but then asks, if this is true, why not Zombies? Why couldn’t the universe, made as it is of unconscious entities, simply allow for arrangements of matter exactly like us but without any internal subjective reality? This basic argument seems to me irrefutable; an argument in the face of which materialism must wither and die. Chalmers is fully aware that the most problematic aspect of this argument is the Zombie’s ability to “think” about consciousness. And it is here that Yudkowsky finds Chalmers theories descend into absurdity. But why, if the materialists premises are correct? Indeed it seems, according to a classical view of physics, that if a super-mathematician had perfect knowledge of all arrangements and properties of matter from a very early time in the universe he could predict perfectly the future existence of a book called “‘The Conscious Mind’ written by David Chalmers” and he could make this prediction without any reference to any of Chalmers’ mental states. If this is possible, (and a materialist would seemingly HAVE to agree) then it seems wholly possible that a Zombie could indeed write books on consciousness without consciousness being at all present. Throughout his rebuttal to Chalmers, Yudkowsky tacitly endorses a interactionist view of consciousness where the behavior of a philosopher musing on the vagaries of mind cannot be understood without reference to subjective experience as part of the causal chain. But of course, this gives up the ghost! Chalmers, in fact, does not rule out interactionist dualism, a view which, if true, would make Zombies impossible. Unfortunately, this rules out not only materialism but the causal closure of the physcial world! Chalmers doesn’t want to go out on such a limb. The “defense” of epiphenomenalism Chalmers mounts in his Zombie argument is an attempt to grant as much to the materialist as logically possible. As counter-intuitive epiphenomenalist dualism is it’s still preferable to materialism, a view which is simply impossible. But, again, there are still OPTIONS available to us other than such a strange counter-intuitive position as epiphenomenalism; options which Chalmers ENDORSES! A point Yudkowsky fails to acknowledge.
To speculate further (and digress somewhat), it does seem some sort of panpsychist monism is the most satisfying of all possible psychophysical theories; not only does it have the virtue of maintaining the causal closure of the physical world, it, like the most far-reaching theories of theoretical physics, unites different phenomenon into one structure. It still is problematic however that our thoughts about consciousness, while impossible to be “divorced” from conscious experience (zombies are impossible in a monist view of reality), aren’t caused by consciousness in the same way that the existence of zebras “causes” our beliefs about Zebras. It seems the role of consciousness in the causal chain is deeper and more non-linear in any view of consciousness that isn’t full-blown interactionist dualism. Perhaps this is actually a virtue of monism. Maybe we should expect that subjectivity itself, the thing allows us to have “beliefs” at all (the Zombie has no beliefs as a belief is wholly defined as a subjective intentional state), should play a more “grounded” role in our belief formation mechanism, including beliefs about consciousness. Consciousness doesn’t “cause” our beliefs about consciousness, it merely makes them true. Weird, but not impossible.
If I had to bet, an ultimate theory of reality would be an information theory in which the fundamental constituents of the universe are subjective “monads” undergoing computational processes; I favor this view metaphysically because subjectivity has irreducible “intrinisic” qualities, physical entities are wholly defined by external structure and function. It tells us what the universe is made of “in and of itself,” perhaps penetrating the world of unknowable reality that Kant believed we could never see. And if, as Tononi and Koch’s recent scientific work suggests and consciousness IS integrated information (a view I first heard fuzzily articulated in, that’s right, Chalmer’s The Conscious Mind), then such a theory could explain ALL the phenomenon of the universe, from the cosmological to the phenomenological (because it turns out the former is merely a version of the latter). But this is highly speculative.
Anyway, I don’t think Zombies can exist. Neither probably does Chalmers. But it’s the materialist who has the impossible task here; deny the zombie but maintain the physical world as closed and fundamentally unconscious. He can’t do it.
Why couldn’t the universe, made as it is of unconscious entities, simply allow for arrangements of matter exactly like us but without any internal subjective reality?
Because an arrangement of matter exactly like us would (if under the same set of physical laws as us) be conscious.
I agree with the spirit of what Eliezer has written here: The possibility of p-zombies would suggest epiphenomenalism. If you believe in redness, you should expect it to be causally efficacious.
However, subjective redness manifestly exists; subjective redness manifestly does not exist in any physics known to us; yet the physics we have already have appears to be fantastically predictive in detail, and quite capable of producing intelligent behavior in principle.
The usual way out of this is to deny my second proposition, and say that redness is a type of brain state or property, and therefore as much a part of physics as any other material thing. But the elementary properties one finds in physics are of a very limited nature: quantitative, geometric, causal, probabilistic. How can piling up number and shape, even when glued together by causal relations, create color?
The answer is that it cannot, at least if you restrict yourself to logical, set-theoretic, and other relations truly intrinsic to arithmetic and geometry. This is why people become property dualists, or believers in “strong emergence”.
Another possibility canvassed by Chalmers is a reinterpretation of the mathematical formalism of physics, so that it is about something other than what it appears to be about, namely entities starkly devoid of the “secondary qualities” revealed in conscious perception. This is his panpsychism or panprotopsychism. It’s “pan-” because current physics has monistic tendencies, a homogeneity of kind among its fundamental entities; if any of them are mind-like or qualia-like, one might expect all of them to participate in or at least to approach that condition.
Given the problems of epiphenomenalism, I find this monistic approach more appealing. However, it seems that even those few philosophers who pursue this avenue are hindered by a rather crude notion of mind. People talk as if consciousness is nothing but sensation, and as if sensation is nothing but a pile of pixel-like elementary sensations, and as if these elementary qualia then just need to be identified with something physically elementary. There is a tendency, for example, which denies that there is any phenomenology of thinking, and no such thing as conscious perception of meaning; cognition is unconscious computation, and it’s just the raw feels of sensation which exist and constitute conscious experience.
But in reality, they are just the part of consciousness which is hardest to deny. Also, this mode of thought is least removed from billiard-ball materialism: an object is a heap of particles, a conscious experience is a heap of qualia. Alas, it is not so simple.
If anyone out there really wants to engage in phenomenology, I have a few recommendations. First, you must overcome the conceptual reflex which rejects the reality of “reifications” and “abstractions”. Err in the other direction, see what picture you build up, and then have a go at paring it back. My training curriculum is as follows. First, Chapter IX of Russell’s The Problems of Philosophy, which makes the case for the reality of properties and relations, as entities which exist just as much as “things” do. Then, ideally I would recommend Reinhardt Grossman’s The Categorial Structure of the World, for a mind-stretching example of a systematic ontology in which abstract entities exist just as much as concrete ones do, and in which the relationship between them is also analysed at length. But the book may be hard to obtain, so some other contemporary system of categories might have to do. Then, one should tackle Husserl and Kant, for epistemologically systematic ontology. And the final step—but by now I’m just describing my own research program—is to interpret the world of appearances which one has been describing as interior to a single entity, and to embed that in physics. As I remarked here last month, quantum entanglement suggests an ontology in which there are fundamental entities with arbitrarily many degrees of freedom. It is the one indication we have from physics that old-fashioned atomism (according to which all fundamental entities are simple and perpetually encapsulated) may be radically wrong.
Returning to the theme of zombies—in a monistic ontology like this, you can’t subtract the mental properties and leave the causal relations unchanged, so the critique of epiphenomenalism no longer holds. The best one can do is to imagine a possible world in which the causal laws are isomorphic, but the elementary states they relate are different: are not ‘mind-like’, are not states of consciousness. However—and this is relevant for AI, Friendly or otherwise—there is nothing to guarantee that every causal simulation of consciousness even in this world will itself be conscious, if consciousness is indeed this deeply grounded in “substance” rather than “algorithm”. For example—if conscious states are only ever states of a single elementary entity (e.g. a single tensor factor such as I mention in the previous link), then any distributed simulation of consciousness will not be conscious, even though it will exhibit all the same problem-solving capacities. (The correspondence between these two would begin to break down if they set about investigating their own physical constitution, which by hypothesis is different, so there is a limit to the duplication involved with this sort of “zombie”.)
Note 1: Chalmers does acknowledge the challenge of epiphenomenalism in The Conscious Mind, and discusses it at length. He proposes a number of ways in which the phenomenal properties might still be causally relevant, but does say that if forced to a choice between epiphenomenalism and eliminativism, we should prefer the former. (He develops the monistic alternative to property dualism in later papers.)
Note 2: A few commenters on this blog have tried the old dodge according to which “redness” is just a word. I would agree that it is a categorization whose scope is vague and varies with the individual. But is it not clear that the individual patches of color and shades of red to which it is applied do exist, and that they pose a challenge to the colorless ontology of physics, regardless of how we group them by name?
Mitchell -
I agree totally. As someone who has read Chalmers entire book, it’s frustrating to read so many people misinterpreting his views. Chalmers is in no way committed to a strict epiphenomal dualism; with the Zombie argument Chalmers is merely demonstrating the intellectual bankruptcy of traditional reductionist dualism. He hesitantly endorses epiphenomenalism only because he is granting the materalist as many possible premises as he can to make his point. The materialist would seemingly want to hold to these facts:
1)The physical world is causally closed. All physical activity can be explained fully by the laws of physics.
2)The fundamental constituents are in no way conscious and exist as solely extrinsic entities in space and time.
3) The mind is fully reducible to physical activity.
Chalmers grants premise 1 and 2 but then asks, if this is true, why not Zombies? Why couldn’t the universe, made as it is of unconscious entities, simply allow for arrangements of matter exactly like us but without any internal subjective reality? This basic argument seems to me irrefutable; an argument in the face of which materialism must wither and die. Chalmers is fully aware that the most problematic aspect of this argument is the Zombie’s ability to “think” about consciousness. And it is here that Yudkowsky finds Chalmers theories descend into absurdity. But why, if the materialists premises are correct? Indeed it seems, according to a classical view of physics, that if a super-mathematician had perfect knowledge of all arrangements and properties of matter from a very early time in the universe he could predict perfectly the future existence of a book called “‘The Conscious Mind’ written by David Chalmers” and he could make this prediction without any reference to any of Chalmers’ mental states. If this is possible, (and a materialist would seemingly HAVE to agree) then it seems wholly possible that a Zombie could indeed write books on consciousness without consciousness being at all present. Throughout his rebuttal to Chalmers, Yudkowsky tacitly endorses a interactionist view of consciousness where the behavior of a philosopher musing on the vagaries of mind cannot be understood without reference to subjective experience as part of the causal chain. But of course, this gives up the ghost! Chalmers, in fact, does not rule out interactionist dualism, a view which, if true, would make Zombies impossible. Unfortunately, this rules out not only materialism but the causal closure of the physcial world! Chalmers doesn’t want to go out on such a limb. The “defense” of epiphenomenalism Chalmers mounts in his Zombie argument is an attempt to grant as much to the materialist as logically possible. As counter-intuitive epiphenomenalist dualism is it’s still preferable to materialism, a view which is simply impossible. But, again, there are still OPTIONS available to us other than such a strange counter-intuitive position as epiphenomenalism; options which Chalmers ENDORSES! A point Yudkowsky fails to acknowledge.
To speculate further (and digress somewhat), it does seem some sort of panpsychist monism is the most satisfying of all possible psychophysical theories; not only does it have the virtue of maintaining the causal closure of the physical world, it, like the most far-reaching theories of theoretical physics, unites different phenomenon into one structure. It still is problematic however that our thoughts about consciousness, while impossible to be “divorced” from conscious experience (zombies are impossible in a monist view of reality), aren’t caused by consciousness in the same way that the existence of zebras “causes” our beliefs about Zebras. It seems the role of consciousness in the causal chain is deeper and more non-linear in any view of consciousness that isn’t full-blown interactionist dualism. Perhaps this is actually a virtue of monism. Maybe we should expect that subjectivity itself, the thing allows us to have “beliefs” at all (the Zombie has no beliefs as a belief is wholly defined as a subjective intentional state), should play a more “grounded” role in our belief formation mechanism, including beliefs about consciousness. Consciousness doesn’t “cause” our beliefs about consciousness, it merely makes them true. Weird, but not impossible.
If I had to bet, an ultimate theory of reality would be an information theory in which the fundamental constituents of the universe are subjective “monads” undergoing computational processes; I favor this view metaphysically because subjectivity has irreducible “intrinisic” qualities, physical entities are wholly defined by external structure and function. It tells us what the universe is made of “in and of itself,” perhaps penetrating the world of unknowable reality that Kant believed we could never see. And if, as Tononi and Koch’s recent scientific work suggests and consciousness IS integrated information (a view I first heard fuzzily articulated in, that’s right, Chalmer’s The Conscious Mind), then such a theory could explain ALL the phenomenon of the universe, from the cosmological to the phenomenological (because it turns out the former is merely a version of the latter). But this is highly speculative.
Anyway, I don’t think Zombies can exist. Neither probably does Chalmers. But it’s the materialist who has the impossible task here; deny the zombie but maintain the physical world as closed and fundamentally unconscious. He can’t do it.
Because an arrangement of matter exactly like us would (if under the same set of physical laws as us) be conscious.