Searle's argument against AI and emergent properties--part 1

By Paul Almond

While it is reasonable to regard consciousness as an emergent property of a physical system there is no profound sense in which it can be said that different people's brains work according to the same kinds of processes and an appropriately programmed computer and a human brain would work according to different processes. Any difference between these situations is just a matter of degree and any argument that we should presume other people conscious because their brains work in basically the same sort of way could also be used to justify presuming an appropriately programmed computer conscious.

John Searle is a philosopher well-known for taking a position against strong artificial intelligence (strong AI) [1]. Strong AI, at least according to Searle's classical definition, is the idea that a system is conscious if it performs the correct computations. Searle makes some arguments that strong AI is wrong and even incoherent [1,2,3]. I will not be trying to refute all those arguments here. This article is about Searle's assertion that consciousness is an emergent property of a physical system and that this makes it unlikely that an appropriately programmed computer, just by performing the correct computations, would be conscious. "Emergent property" simply means some higher level property that a system has due to what is physically happening in it at a lower level.

An example of this, which I will use throughout this article, is wetness. Wetness arises due to the low level behaviour of molecules in a system. There is nothing supernatural about wetness: it is simply a high level property associated with low level processes and it disappears when we look closely enough at the molecules. Using a computer to simulate the behaviour of molecules in a fluid would not make real wetness. Our simulation would be just that - a simulation. We would not get real wetness because our simulation does not involve the actual processes needed to cause real wetness. To cause that you need the particular physical process of molecules moving around in a fluid. The physical processes underpinning the simulation are completely different: they are just movement of electrons through semiconductor junctions. Electrons are not wet and movement of electrons does not cause wetness. Searle makes a similar argument about consciousness, saying that a simulation of a mind should not be expected to duplicate the consciousness of that mind any more than a simulation of the movement of molecules in a fluid should be expected to duplicate wetness, because the physical process underlying brains and computers are completely different.

I will argue that this idea is flawed.

Consciousness as an Emergent Property

Searle maintains that behaviour in itself is insufficient for consciousness. He says that consciousness is an emergent property of a system in which the right sorts of physical process are occurring. As an example of how this sort of reasoning works, imagine a large number of hydrogen and oxygen atoms, joined by covalent bonds to form water molecules, and a computer simulation of the interaction of a large number of hydrogen and oxygen atoms joined by covalent bonds. The simulation captures some of the behaviour of water, but the simulated water is not real water. It is not really wet. Searle would maintain that something like wetness is an emergent property of matter undergoing particular physical processes and that it is the same for consciousness: matter, undergoing the appropriate physical processes, has the emergent property of consciousness. Searle says that there is nothing mystical about this: it is just a fact about how physical systems behave. It is no more magical for consciousness to emerge from a system undergoing the relevant physical processes any more than it is for wetness to emerge from atoms undergoing the appropriate processes. Of course, you can get wetness in other ways, by using different atoms and making different molecules, but all of these ways will involve basically the same sort of physical processes. These physical processes are out of reach of the computer simulation and it will never cause real wetness. Likewise, a computer simulation of intelligent behaviour, in itself, is insufficient for consciousness. There may be ways of generating consciousness other than in human brains, using the same general kinds of physical processes. As we do not really know what these processes are yet, we cannot totally rule out a computer undergoing them and generating real consciousness, but Searle maintains that there is no reason to think consciousness is there, just because you have the right behaviour:

"…that is the old mistake enshrined in the Turing test. If this principle were correct, we would all have to conclude that radios are conscious because they exhibit intelligent verbal behaviour. But we do not draw any such conclusion." [4]

To Searle, the underlying physics and processes are everything. You can think it probable that other people are conscious, not because they act as you do, but because you know that the physical workings of their brains are essentially the same: the same general kinds of physical effects occur in their heads as in yours, so the same emergent properties can be expected:

"Where knowledge of other minds is concerned, behaviour by itself is of no interest to us; it is rather the combination of behaviour with the knowledge of the causal underpinnings of that behaviour that form the basis of that knowledge." [4]

The Problem

I disagree with Searle's view that consciousness being an emergent property of physical systems makes it absurd to associate consciousness with the appropriate behaviour. My problem is that this idea seems to assume that there are two kinds of things we can observe in the world. One type of thing is a particular physical process, or the existence of particular physical objects; for example the interaction of water molecules and the existence of the real water molecules. I could produce two glasses of water and nobody would have problems with the idea that the same general physical processes are involved in each. The second type of thing is some kind of abstraction of a physical system in which the physical processes are not necessarily the same. For example, most people would admit that the same kind of abstraction could be made of a glass of water and a computer simulation of a glass of water, but would say that the underlying physical processes were different in each case. Underpinning the real wetness is the interaction of water molecules. Underpinning the simulated wetness is the interaction of electrons, which are not going to produce wetness and in any case are behaving in a way which appears, on casual inspection, totally different from the molecules in the glass of water. It is this kind of obvious analogy that makes Searle's position seem so attractive: we know it is stupid to think that simulations of water molecules involve real wetness. My position is that this distinction between different kinds of properties is flawed. I am not saying that the simulated water is wet. What I am saying is that there is no profound difference between physical properties and processes and the sorts of abstractions of these things that are captured in computer models: everything is physical processes.

Suppose we want to know if some system is undergoing a particular kind of physical process. Whether I am right, or whether Searle is right, we should be able to formalize our approach, our rules that tell us when a particular physical process is happening. Suppose we do this by making a machine that I will call a "physical process detector." The physical process detector works a bit like my "algorithm detector" in another article [10]. It has a set of probes with measurement tips that it can move around. A computer program controls the probes and the machine captures the measurements that it needs. The computer program analyzes the data and tells us whether or not the physical process(es) in which we are interested in is (or are) present. As emergent properties are caused by processes in a system then a physical process detector is also a physical property detector. We can use it to search for any emergent property that we wish, provided that we can define the processes associated with that property well enough to write the program. For example, we could program the machine to detect wetness and it would take readings to see if atoms were present and doing the sorts of things which cause wetness as an emergent property -- using rules for "wetness detection" that we programmed into it. Maintaining that such a machine could not be programmed, in principle, to detect something like wetness would be maintaining that the concept of wetness could not be formalized, even in principle, and elevating it to a "supernatural" status. I have argued in another article about the "supernatural" that this sort of idea would be incoherent [11].

If we properly programmed our machine to detect wetness and put it next to a glass of water it should detect wetness. If we put it next to a computer simulating the behaviour of water it should not detect wetness. (We will ignore any tiny amounts of water on the computer, for example from perspiration on the fingertips of people typing on it.): electrons are not wet and I will never persuade any of you that they are. However, suppose I defined a new type of property of a system and called it "b-wetness." We could program the physical process detector to detect b-wetness when placed near a glass of water and when placed in front of a computer that is simulating the behaviour of water. It does this because the definition of b-wetness fits with processes in both the glass of water and the matter in the computer.

Next Page >