Dan Ports
2002/11/26
21L.448

A Rebuttal to SearleÕs Chinese Room Argument

	The ultimate goal of artificial intelligence research is to create computer systems that can simulate thought at a human level. The task of creating such a system encompasses a variety of technical challenges. It also prompts a philosophical question: if such a system were created, how much intelligence can be attributed to it? Some have claimed that it must be capable of ÒthinkingÓ or ÒunderstandingÓ in the same manner as a human, while others claim it is simply blindly following rules. John Searle, in his paper ÒMinds, Brains, and Programs,Ó argues against the notion that Òthe appropriately programmed computer really is a mindÓ (353). He uses his famous Chinese Room example to claim that ÒintentionalityÓÊis the difference between a thinking human and a computer that is merely processing information according to rules. According to Searle, though the machine may be behaviorally indistinguishable from a human, it is impossible for anything that does not have Òthe same causal powers as brainsÓ to be intentional, and therefore it cannot be intelligent (369). This argument is flawed because the claim that only human brains are capable of intention is not as trivial as Searle makes it appear. When considered from the perspective that humans are a form of biological machine, SearleÕs notion of intentionalism becomes an arbitrary, unclear distinction.

	One definition for intelligence is based on observable behavior: a system is intelligent if it behaves intelligently. This is the definition proposed by Alan Turing in his paper ÒComputing Machinery and Intelligence.Ó In this paper, he introduces the Turing test, which involves allowing a human judge to communicate via a text-based terminal with a computer program and another human. The judge is not told the identities of the two others, and the computer is programmed to imitate a human; the judge must determine which is really the human and which is the machine. If the two cannot be distinguished, then it follows that the computer must have mental abilities comparable to those of a human. The power of the test comes from its ability to test such a wide variety of subjects. TuringÕs example transcript shows that a conversation can test knowledge of subjects as diverse as poetry, arithmetic, and chess; he claims that Òthe question and answer method seems to be suitable for introducing almost any one of the fields of human endeavourÓ (6). In order to pass the test, the computer must have a strong foundation of general knowledge and human-level ability in different fields; if it does not, the Turing test will expose it. According to Turing, a computer that can successfully pass the test must have Òthe intellectual capacities of a man,Ó and it can reasonably be said to be intelligent and ÒthinkingÓ (5).

	John Searle accepts TuringÕs claim that machines can eventually be made capable of passing the Turing test. However, he rejects the notion that this will make them intelligent. He illustrates his argument using a famous thought experiment: a man who does not speak Chinese is placed in a room and given a book of rules in English which he can understand. He is then passed a set of Chinese symbols, and he follows the instructions in his book in order to assemble a response from Chinese characters. By doing this, he is responding coherently to questions posed in Chinese, and indeed it is possible that, with a large enough rulebook, he would be able to pass the Turing test in Chinese. But Searle claims that ÒI have inputs and outputs that are indistinguishable from those of the native Chinese speaker É but I still understand nothingÓÊ(356). By this reasoning, a computer cannot be intelligent: it is simply performing Òcomputational operations on purely formally specified elements, [which] by themselves have no interesting connection with understandingÓ (357). Searle argues that a system is intelligent if it is capable not just of giving the correct output but also understanding the meaning, and this requires intentionality. This creates a dichotomy between conscious, intelligent systems with intentionality, and machines that are merely manipulating symbols. He rejects the ÒdualismÓ of strong AI, which postulates that mind and brain are different, and Òprograms are independent of their realization in machinesÓ (371). For Searle, the mind and brain cannot be separated, because an intelligent mind must have intentionality and intentionality is purely Òa biological phenomenonÓ exclusive to the brain (372). Thus, the only way a system can possibly be intelligent is if it has a nervous system like that of a human. Searle writes that, though they can be used to construct computing machines, Òstones, toilet paper, wind and water pipes are the wrong kind of stuff to have intentionality in the first place Ñ only something that has the same causal powers as brains can have intentionalityÓ (369). By this reasoning, a computer program, even if it meets TuringÕs behavioral standard for intelligence, cannot be intelligent by SearleÕs standard of intentionality.

	A major problem with SearleÕs argument is that it draws an unreasonable distinction between the human brain and machines. This can be best understood by considering the perspective that the human body is a form of machine. This may seem to be a shocking idea, since so many philosophies and religions hold that humans, animals, and machines are in some respect fundamentally different. However, in a physical sense, they are more similar than they might seem. A machine is a collection of parts assembled into a functional system. This can also be said of the human body: it is composed of a set of organs that operate together, and these organs are in turn made up of cells, increasingly smaller parts. The major physical difference is that a human is a machine composed of biological parts, whereas a man-made machine is generally made from wood, metal, or similar materials. Yet when they are reduced even further to the atomic level, both humans and machines even share the same constituent elements; they are simply arranged differently. This applies as well to the brain. Searle claims that the human brain is intelligent because it is capable of understanding the input it manipulates. He acknowledges that a machine with an artificial nervous system would be equally intelligent, and that Òit might be possible to produce consciousness, intentionality, and all the rest of it using some other sorts of chemical principles than those that human beings useÓ (368). Searle claims that the only way a computer can be intelligent is if it simulates a biological brain. This claim seems unreasonable because it implies that there is some intrinsic property unique to the brain that makes it capable of understanding. The brain is simply a machine made up of neurons in some particular complex arrangement, which could conceivably be replicated on a sufficiently powerful computer. Searle admits that such a system would be intelligent. But this arrangement is simply a system that takes inputs and generates an output by processing the data through neurons. The neurons are biological computational elements, and there is no reason that they could not be replaced with digital computational elements. If this is done, then the brain has been transformed into a program for a digital computer of the very sort that Searle claims cannot be intelligent. But the two are equivalent, and it cannot reasonably be said that one is intelligent and the other is not. SearleÕs assertion that only a brain can be intelligent is flawed because the brain is simply a computing machine that happens to be implemented using biological components. The distinction he draws between biological and mechanical or electronic implementations is not nearly as significant as he claims it to be.

	SearleÕs argument centers on the notion of intentionality: that the human brain can have intention, but computer programs cannot. Unfortunately, this notion of intentionality is not a well-defined, meaningful concept. Intentionality is not a natural, intrinsic property of an entity, but an arbitrarily assigned label that depends on perspective. Norbert Wiener suggests the impossibility of assigning intention when he writes in God and Golem, Inc that Òa hen is merely an eggÕs way of making another eggÓ (36). This situation described in this statement is not an unusual one; the cycle of Òbiological alternation of generationsÓ in which an egg produces a hen and a hen produces an egg is well understood. However, it is surprising to see it phrased this way because we expect it to be the hen rather than the egg that has consciousness and intention; we expect that the hen produces a egg in order to fulfill its goal of creating another hen, not vice versa. Yet either viewpoint can be valid. When viewed from an evolutionary perspective, both the hen and egg are equally necessary for the species to continue to exist, and each one creates the other; the two exist as a ÒdualityÓ (35). This implies, as Wiener suggests, that though we normally attribute intention only to the hen, the egg can also be seen as having intention. The choice is purely arbitrary; neither is naturally more intentional.
	Even Searle agrees that intentionality is a quality arbitrarily assigned by humans when he says that Òwe find it natural to make metaphorical attributions of intentionality to [our tools] É but no philosophical ice is cut by such examplesÓ (358). Searle claims that the purpose and intentionality of tools is merely a metaphor we have ascribed to them for our convenience in describing them. We may view them as having intention, but this is meaningless; it is not a real form of intention, as it certainly does not make the tools conscious or intelligent. If we accept this, then we must question whether all other intentionality is equally meaningless, including the intentionality we assume the human mind to have. If we again treat the human brain as a machine made up of biological components, we observe that the individual neurons do nothing more than transmit and combine signals; they cannot have individually have intentionality or consciousness. The human mind comprises an arrangement of these neurons. Applying the rules of physics, we could conceivably predict the thought or action the brain would produce in response to any particular stimulus: it is an output determined entirely by the reaction of the brainÕs neurons to a certain input. A similar argument that human behavior is deterministic is found in the Book of the Machines in Samuel ButlerÕs Erewhon. The Erewhonian author claims that free will is merely an illusion, using the example of a train driver who Òcan stop the engine at any moment that he pleases, but he can only please to do so at certain points which have been fixed by ÉÊan unseen choir of influences, which makes it impossible for him to act in any other way than oneÓ (218). The Erewhonian uses this to argue that humans are similar to machines, because both behave deterministically not Òspontaneously.Ó In this regard the brain is essentially the same as a computer: it is a machine composed of elements that lack intentionality, and it produces a deterministic outcome according to a set of rules. Thus if we accept SearleÕs premise that a computer program lacks intentionality, we must say that the same is true for the human brain. This means that intentionality is not a useful distinction for evaluating intelligence.

	SearleÕs premise is that a system may behave intelligently Ñ as measured by the Turing test, for example Ñ without actually being intelligent. He claims that in order to be intelligent, it must be consciously capable of understanding what it is doing, which requires intentionality. This argument is flawed because the notion of intentionality is really an arbitrary distinction rather than a natural property, as WienerÕs example of the egg and the hen helps demonstrate. His notion that intelligence can only exist in brains because only brains are capable of intentionality is refuted by considering the brain as a form of biological machine. When it is treated simply as an organized collection of neurons, then it becomes an information processing system very much like a computer program. This implies that a system can still be intelligent if it is implemented electronically rather than biologically.