There are some weaknesses in my presentation: I have summarily
dismissed interactionism and externalist intentionalism in my
discussion. It is conceivable that an externalist intentionalist
argument might be able to derail the functionalist argument that I
have presented here.
If externalist
intentionalism is established, however, an epiphenomenalist such as
Chalmers or myself will be no happier than if internalist
intentionalism is established. It is conceivable that an argument from
externalist intentionalism might be strong enough to derail
internalist intentionalism, yet not strong enough to establish
externalist intentionalism, and it is remotely conceivable that such
an argument might be appropriated by phenomenalists to establish their
case against internalist intentionalism. Such forays will have to
wait for later.
Another weakness of Harman's theory and our fixed-up version is that it relies on empirical speculation about the construction of human beings. This empirical speculation may turn out to be wrong. I don't take up this objection in this paper because I believe that Harman's theory, or something quite like it, is more or less correct (except for its claim to explain phenomenal properties). I think Chalmers would also agree with this. Searle and Dreyfus argue that functionalist theories cannot possibly explain intelligence. I'll have to leave most of this debate for other occasions.
Ned Block has an argument that he claims to be a serious, perhaps even fatal, challenge to functionalist explanations of phenomenal properties. I would likely to briefly address this argument, since I believe that, unfortunately, it does not really pose much challenge at all.
Block asks us to imagine that Eliza is raised in a room in which everything will change its color randomly once every few minutes. Eliza is taught the names of colors, so she can accurately tell you the color of any object she is looking at, but she has no associations, for example, of red with fire or blue with the sky. For Eliza, particular colors are not associated with particular objects, because all of the objects she has known her whole life have constantly changed color. Thus, Block claims, she would not likely develop any preferences or inclinations for one color over another. There would be no functional difference between the colors. Yet, we can imagine that her color experience is just as vivid as ours, even though there is no functional difference between Eliza's representations of different colors. To ward against arguments that humans might be genetically programmed so that different color experiences innately function differently, Block asks us also to imagine that Eliza is genetically engineered, if need be, to remove any such innate asymmetries.
Despite Block's claim to the contrary, I do not see this thought experiment as posing any problem for functionalism. A functionalist will just point out that even if it turns out that Eliza will say her different color experiences are vastly different (though we have no real reason to believe that it will turn out this way) and it also turns out that Eliza's different color experiences function nearly identically despite her claims to experience them very differently (another dubious assumption), there is no contradiction for functionalism here. All that is required is that Eliza be so constructed that she judges the experiences to be vastly different. A robot could certainly be constructed to incorrectly judge things that are similar to be vastly different, and we have no reason not to allow that Eliza, or even us, might be constructed in such a manner.
Block might respond to this by saying that Eliza might not know her color experiences are very different from each other. Therefore, she would not judge her color experiences to be different from each other, and this leaves the functionalist without any explanation for the fact that her color experiences are actually quite different from each other. To this the functionalist will just bite a very easily digested bullet and say that if Eliza's color experiences function identically, and they are judged identically, then there is in fact no important difference between them.