In this article I give a background to and description of Searles Chinese Room argument; demonstrating its effectiveness; and confirming the critique as decisive; in refuting the tenets of Strong AI. I also intend to show that whilst Searles opponents fail to employ suitable methods to counter the argument, the longevity of the discussion lies in a shared oversight. This coincides with my intention to reveal embodied cognition as fundamentally misconstrued.

BACKGROUND

Berkeley, an idealist, argued that everything is a figment of the mind, and that to be is to be perceived. This is considered an extreme point of view, and is not very popular. In contrast, Descartes posited the mind as a unique substance but, whilst Descartes’ position may seem more natural in that it fits with common beliefs of having a body and a mind (perhaps soul), it is flawed – for a separate mental substance cannot interact with or have a causal effect on the physical world.

Descartes suggested that a particular organ may enable physical interaction with mental substances, however this was quickly realised to be fallacious, which led to the suggestion that the mind and body do not interact but are fortuitously in perfect synchronicity – a situation commonly attributed to divine design. However, the development of modern science called for a physical account of the mind, spurring a return to physicalist theories.

The physicalist agenda of behaviorism claimed that mentality is a way for referring to the disposition of causal agents; that all we have, or need access to, are the observable patterns exuded by some internal ‘black box’ of cognition. This black box would account for Descartes ‘ghost in the machine’, and would not require a reliance on non-physical substances. But omitting any explanation of the black box is philosophically weak. Furthermore, it would seem that reasoning and language are more complex and formal than ‘behaviors’.

The concept of formality was developed by the work of Alan Turing, and the Church-Turing thesis provided a suitable theory of formal computation in the Turing machine. However the Turing machine is an abstract concept, and must be instantiated. The idea of these instantiations being to the body as software is to hardware is known as the mind-program analogy, and is the basis of Strong AI, which suggests therefore that computation is cognition. This contrasts with Weak AI which claims no more than the possibility of simulating cognition with computation.

THE CHINESE ROOM

My considerable description of the background is necessary, as the Chinese Room argument is very specific in its premises and aim. Here is the thought experiment :

• Put a man with no understanding of Chinese in  a room.
• The room has a slot at one side through which Chinese symbols are passed in.
• There is also a table explaining, in English, which Chinese symbol is the correct response to the one passed in.
• There are then a pile of symbols which the man can pass out through a slot in the other end of the room.
• Given sufficient completeness in the system and, in particular, in the table of suitable responses, the man should be capable of outputting the correct Chinese response to questions passed in, yet never gain any understanding of Chinese.

It should be clear that such an example is possible, and that the man does not gain an understanding of Chinese. People often execute methodical tasks without understanding, such as in assembly line work. It is, therefore, up to the proponents of Strong AI to explain this. However it is at this point that the crux of the argument is often misunderstood, so I will demonstrate the premises and aim in a simple list form:

• The argument is against the functionalist claims of Strong AI – it is not against Weak AI.
• Strong AI, in the mind-program analogy, claims to explain the ‘black box’ mind of behaviorism, by correlating the mind with instantiated software.
• The Chinese Room instantiates a Turing machine, but places a human being at the centre, performing the operations.
• Thus, the outcome is that a Turing machine, which is the computational paradigm, can be instantiated around a mind, and yet that mind does not gain any understanding of the computations being performed.

It is important to realise that Searle is not attacking the idea of people being thinking machines. In fact, he agrees with it. Furthermore, it is irrelevant whether or not Searle (or anyone else) can tell the difference between real minds and instantiations of Turing machines – he is not attempting to show that people are not thinking machines, but that instantiation of a computational program – in whatever medium – does not in itself impart understanding. For Searle, Strong AI fails to succeed or to be useful specifically because it is a claim about programs rather than machines, and an uninstantiated paradigm of a program is no better than the non-physical mind of the dualist in explaining causation.

It is typical for replies to Searles thought experiment to suggest either that there is some combination of things that does understand (known as the Systems reply), or that some form of automaton instead of a static room would fare better (the Robot reply). However both these and similar objections overlook the simple point that instantiation of a program does not in itself impart understanding – on realising this, it should be obvious that redefining the instantiation cannot affect the outcome.

Searle argues that only a person, or something with equivalent causal powers, can have cognition or understanding, not by virtue of instantiating a program, but by being a particular thing that can have Intentionality about things, which imparts meaning and enables cognition and understanding.

At this point I would state that Searle succeeds in successfully showing proponents of Strong AI to be incorrect in their claim that the mind is to the body as software is to hardware, at least unless they also wish to claim either that theirs is in fact a dualist position, or that they make no claim whatsoever in regard to the mind being causal. However, when Searle continues his exposition of what this causal agent is, he and his protagonists, in my opinion, come to share the same misconception.

MY ARGUMENT

The Turing test, which forms part of the basis of Strong AI, is misrepresented. It provides a criteria for accepting a machine as thinking, but it is an arbitrary criteria. Accepting a machine as thinking, and knowing that a machine thinks, are not the same. Turing was aware of such issues, which is why in his paper he abandons the task of answering the question ‘Can Machines Think’ and, instead, replaces it with an arbitrary test for ‘thinking’. The Chinese Room is an example of a Turing machine, and it passes its Turing test, but yet it still seems to have no understanding.

The mind-program analogy suffers another inherent weakness, in that natural physical phenomena can represent solutions to problems far better than discrete computations, and there is a possibility of the existence of super-Turing computation. Both of these issues are unaddressed by current theories.

When the Systems, Robot or other replies (such as introducing this or that type of program, or this or that level of complexity) are made, they either simply fail or end up in agreement that a certain type of complex object, and not just a program, instantiate cognition and understanding.

These are arguments over WHERE cognition and understanding ARE, having already made assumptions over what either of them IS. And yet, part of the elegance of the Turing test is that it gives us a way round this issue, and it seems that Strong AI overlooked that. Searle (albeit intentionally) overlooks it too, which is what enables him, in the Chinese Room argument, to invalidate Strong AI. However, this does not provide a suitable basis for continuing his argument that particular things have cognition by virtue of their Intentionality. He gives, and I have, no reason to believe this.

MY CHALLENGE

I claim that cognition and understanding can not be found IN anything. Also, I have a further thought experiment:

• Take Searles thought experiment, and generalise.
• Take the basis of any stance on embodied cognition, say Searles Intentionality.
• Theorise a situation where that basis fails to impart cognition.
• Conclude that, therefore, that basis does not impart cognition.

When Searle claims that things of a particular type have cognition by virtue of their Intentionality, he goes a step too far, as do those that try to specify yet more new types of thing that may instantiate cognition.

A machine that passes the Turing test is taking part in cognition – in fact anything can be, as cognition is based on the arbitrary specification of a level of interaction. The Turing test is merely one such specification.

I therefore state that cognition actually IS interaction, and that as such it can be easily shown to exist, by the existence of interactions rather than of individual things. Whilst my claim clearly requires further clarification, particularly to highlight why it is not simply a form of philosophical interactionism or similar point of view, I end this article issuing the challenge to anyone to prove the existence of any thing, let alone a mind of any sort, without relying on its interactions.

REFERENCES

• Berkeley, G. (1710) A treatise concerning the principles of human knowledge. Ed. Dancy, J. (1998) OUP.
• Descartes, R. (1641) Meditations on first philosophy. In: Discourse on Method and Meditations on First philosophy, (1980) Hackett Publishing Company Inc.
• Searle, J. R. (1980) Minds, brains, and programs. In: Behavioral and Brain Sciences 3 (3), 417-457.
• Turing, A. M. (1964) Computing machinery and intelligence. In: Mind 59, 433 – 460.