Searle's argument against AI and emergent properties--part 3

How this Relates to Searle's Argument

Searle would maintain that there is no reason to think that a computer is conscious, even if it is behaving in the correct way, because the physical processes involved in human brains and computers are different. He also suggests that we can reasonably assume that other people's brains have consciousness because the physical properties are essentially the same. The argument I have just made here should show that this distinction is artificial. Human brains may seem to have the same physical processes, and the same emergent properties, but to represent this with a process detection program in a physical process detector would still require the program to be general enough to detect consciousness in different human brains. We could define some emergent property in a general enough way that it is possessed both by AI systems and human brains and there would be no clearly profound reason why it was less valid. Just as Searle says that human brains involve the same physical processes, we could equally validly say that a computer running an AI program and a human brain involve the same physical processes if we define the physical processes in a general enough way -- and you cannot validly argue that I am allowed to be general in my definitions of properties: you are doing that if you think other human brains are conscious by being similar, but not identical, to yours.

The implication of this is that if Searle can say that other human beings are probably conscious because their brains involve the same physical processes I could equally well say that I can expect certain computers to be conscious because, using more general descriptions of physical processes, they involve the same physical processes as my brain. Any profound nature of this argument is now lost. Searle may seem to be left with one advantage: the sorts of physical processes and properties needed to assume other brains to be conscious are less general than those needed to assume that brains and AI systems are conscious, but it is just a matter of degree. There is no sense in which this establishes that it is likely that other people's brains are conscious and unlikely that computers are not conscious. A third party could walk into such a debate and announce that only Searle's brain is conscious because it has the right physical processes and that neither my brain nor his (the third party's) qualify for consciousness because of the more general physical process detection program needed to include them.

Putting this back into the contexts of wetness and b-wetness, I could argue that consciousness is a very generally defined emergent property like b-wetness. Searle could argue that it is a more narrowly defined emergent property like wetness. A third party could say that we are both wrong and the processes needed for consciousness should be defined much more narrowly than either of us are doing. If Searle has any advantage by using more narrowly defined processes then someone who uses an even narrow definition would have an even greater advantage -- leading to a solipsistic view of consciousness!

None of what I have said here implies that we need to presume that consciousness is not present when the relevant externally observable behaviour of a system is not observed. Searle correctly points out that there are medical conditions which can allow a human to remain conscious but not to show externally observable behaviour that we normally associate with consciousness. We could still presume consciousness to be associated with such systems from an inspection of their interior processes.

What about multiple realizability?

One of Searle's arguments involves what he calls "multiple realizability," and what I call "arbitrariness of interpretation" in some of my articles -- the fact that any physical system can be said to be running any program by making the appropriate interpretation. That sort of issue can be raised for b-wetness. If we define a property so generally that a computer simulating water and a glass of water can have the same property then does this not mean that I can invent any physical property and say that it is shared by any set of systems, given some interpretation? For example, I could define some b-wetness property so that it is found in a glass of water, a block of concrete, a slice of dried out pizza and a cubic metre of vacuum. Does this not mean that trying to define properties like b-wetness is just making things up?

The problem with this is that you can make the same objection against wetness. Wetness may be defined in less general terms than b-wetness, but there is no obvious level of generality of the definition of a property at which properties abruptly become artificial. Multiple realizability is a problem that needs resolving. I suggest an approach in my series of articles: Minds, Substrate, Measure and Value [7,8,9,10,11]. Considering this briefly, as it is not the main subject of this article, we might ask why, if presuming that an appropriately programmed computer is conscious because it works in the same general way as the human brain is valid, it is not similarly valid to say that a slice of pizza is conscious because it works in the same general way as the human brain: we could easily define a physical process detection program that finds the same property in a slice of pizza and a human brain, call it "consciousness" and demand human rights for pizza. I think this would be flawed though. I do not think it would make sense even to start considering an emergent property as a candidate for consciousness unless an abstract description of that property matched, in some way, our own mental experience. The emergent property would need to be a description, on some level, of what it seems like to be us. This would rule out lots of trivial emergent properties that we could easily find in brains and pizza. This does not solve everything, however: we could define an emergent property so that it is a candidate for consciousness and find it in both brains and pizza merely by basing it on a contrived physical process detection program. The important word here is "contrived." The program needed to find this sort of property would be long and an argument could be made that this sort of program would have a "low measure" -- a lower status, in some statistical sense -- in the set of all physical process detection programs.

When we start to consider things in terms of measure, and how contrived our ways of finding properties need to be, this issue of multiple realizability starts to become tractable. It gives us an answer to Searle's assertion that behaviour is of no interest at all [6,7,8,9,10]. Behaviour should be interesting to us because if two systems exhibit the same behaviour this suggests that it will be possible to write a relatively short and uncontrived physical process detection program that "finds" a suitable emergent property in each system. This does not mean that behaviour is everything: it may be possible to find a suitable emergent property, without a very contrived physical process detection program, in a system with no obvious, external evidence of intelligent behaviour. Behaviour, however, should be viewed as a strong indicator of consciousness because it suggests that extracting suitable properties will be easy.

It should be noted that this kind of justification for viewing computers as conscious is not the same as the standard strong AI justification and it is complicated by the issue of measure and it is possible to view this kind of situation with different levels of formality.

Next page >