Searle's argument against AI and emergent properties--part 4

Minds and Reference Class

It may seem strange to talk about considering properties that could be candidates for our own mental states, but I would argue that that is what is done informally by science and humans in their everyday lives anyway. All that you can ever experience, by definition, is your own mental state. When you consider possibilities about how the world is it can only mean that, formally or informally, you are considering possible ways in which the emergent property of your mental state can have come about. This means that any scientific description of reality, as far as you are concerned, if properly expressed, should be a formal description of your abstract mental state and how it relates to the rest of reality. As we do not know everything about our own mental states or reality we have some uncertainty about our mental situations. We have a reference class of possible mental states we could be in and for each of these a reference class of possible descriptions of how those mental states relate to reality. I would argue that the best way of approaching consciousness is to look at it in this way and ask what kinds of situations we should be prepared to admit in principle into our reference classes of possible situations in which our mental states could be. I explore this idea in another article [10]. When considered like this, questions such as "Can a computer be conscious?" become "Is it conceivable that a description of my mental state which involves it being an emergent property of the processes in a computer could ever in principle be part of my reference class of possible situations?" I suggest that such a reference class should include every possibility which involves some formally describable way of relating a valid description of a mental state to physical reality and that the possibility, in principle, of including computers in such a reference class suggests admitting them as reasonable candidates for consciousness, though possibly with some consideration of measure.

We can get an idea of what this reference class view would mean by considering various thought experiments. For example, suppose you wake one morning remembering that a mad scientist you met at a party had abducted you and was going to make you unconscious and scan your brain during the night, making a computer simulated copy which was to be woken the next morning in a virtual reality. Assume that you live in a world where the technology to do this is routinely available -- and nothing that Searle says opposes such an idea; in fact it could be argued that his materialistic view of consciousness pretty much mandates it being possible. Situations similar (in some ways) to this are described in science fiction novels [12,13]. Are you the original person or are you the copy? If you approached this by having an automatic bias against being the copy on account of consciousness being unlikely in computers I would say you are taking a flawed position. You cannot even use an argument such as "I know human brains tend to be conscious because they have the sorts of processes causing my consciousness." In fact, if Searle's methodology is valid you cannot even be sure that human brains are conscious: you could be in the computer experiencing consciousness with memories imported from a brain that had never experienced consciousness: you might only be experiencing consciousness at all as a result of copying. You would need to assume you are not in a computer program to get Searle's reasoning off the ground in the first place. The only way to approach this, without making artificial special cases and assumptions, is to consider the reference class of possible abstract descriptions of your mental states and, for each, the reference class of possible ways in which the relationship between the mental state and the rest of the world could be described -- how the mental state could follow-on from everything else as an emergent property. In practice, the reference class of possible ways of relating a mental state to the rest of reality is probably not going to depend too much on the subtleties of the mental state and you can probably do almost as well by selecting one reasonably likely mental state and considering the reference class of possible, formally describable situations in which it could be. Considerations of measure should play a part in this kind of consideration, as discussed in my other series of articles [6,7,8,9,10]. The point is that such a reference class approach seems to be almost mandatory to make sense of this kind of situation, particularly if we start to consider other issues that are discussed in these other articles, and admitting its use is very suggestive that a very generalized approach to consciousness, with consciousness described as an emergent property in the most general terms possible, is warranted.

Are the physical processes even remotely similar?

Some people may reject my assertion that physical, emergent properties can be considered at such a general level, or they may accept it, but say that even the matter of degree involved is hugely important because the physical processes in human brains are almost exactly the same, allowing a very narrow classification of physical process to be considered, while any physical processes that appropriately programmed computers and brains have in common, even if these are admitted, are much more different from each other, requiring a very wide classification of physical process to be considered.

Against this, I would first say that it is basically an appeal to incredulity. Another point I would make is that we cannot even be sure that the physical processes in different people's brains are even remotely similar. In fact, we cannot even be sure that the physical processes in two brain cells or even two carbon atoms are remotely similar.

How can this be the case? The processes may seem very similar from our human point of view, considering what is known about physics, but we do not know that that physics is fundamental. It could be that underpinning a "fundamental" particle there is a lot of physics, and a lot of processes, that we do not know about. If this happened to be the case we could not be sure that two "fundamental" particles (though of course they would only seem "fundamental" to us if this were the case) were there because of remotely similar processes. Suppose we choose some apparently fundamental particle of type "X." It may be that any particle of type X just happens to be a particularly stable and common type of emergent property that arises in many different ways, with many different types of processes underneath it and it may be that if we had to consider the (currently unknown) lower level physics as well, then the description of the general type of processes underpinning X particles needed to find them, using the physical process detector discussed previously, lacks any simple one-to-one relationships and is vastly more complex, more abstract and, to us, apparently more contrived than the sorts of programs needed to "find" similar physical properties in brains and appropriately programmed computers.

If this were the case it would cause problems for Searle's position. How could we say that two brains use the same general kind of physical processes when it was not even possible to say that two "fundamental" particles in the two brains, or even in the same brain, existed according to the same physical processes? We could say that the processes underpinning the particles do not matter and that the appearance of "fundamental" particles as being the same means that we can ignore what is underneath, but that seems to be specifically what Searle wants to prohibit us from doing on a larger scale with computers and brains. If process is everything, and behaviour is nothing, then how could we say that one particle of type X was the same sort of emergent property as another particle of X if only the behaviour were similar? Another way out of this problem may seem to be by saying that whatever causes two particles of type X must be the same sort of phenomena, even if how it occurs is different. For example, suppose the physics underpinning particles of type X is the "zod soup". (I am just inventing some nonsense physics terms here to avoid having to attack or defend any specific views in physics.) It might seem that we can declare all X particles to rely on the same processes by virtue of all being based on "zod soup" physics, but just being based on the same physical laws would hardly justify this. It may be that the description of what is going on in the "zod soup" to get one X particle is vastly different from what is going in the "zod soup" to get another X particle and even if the two kinds of processes can be made to sound the same with generalizations, we could do the same with computers and brains: I doubt, for example, if Searle would be impressed if I said that computers and brains have the same causation for their behaviour because they both use condensed matter. It is hard to escape this: if physics at a low level happened to mean that two X particles, if we applied Searle's standard, did not share the same causation then it would be invalid to ignore this and declare the same causation to exist in things made of X particles.

I am not saying that physics is like this, but I am saying that Searle's argument rests on the assumption that it is not. It does not end with particle physics. Even if the underpinnings of particle physics "cooperate" with Searle, we could ask the same questions about what, if anything underpins the physics underpinning particle physics. Searle's argument depends on physics not turning out like this all the way down.

I would be interested in knowing if Searle accepts the possibility, even in principle, that physics could turn out like this. If he does, I would also be interested in knowing what he thinks the implications would be for his argument if physics did turn out to be like this.

Next page >