Does Searle’s Chinese Room argument establish that the mind is not a computer program?

I recently looked at the case for artificial intelligence and I revisited the arguments laid out by John R Searle in his argument that so called strong artificial intelligence cannot evolve from a computer program.  Using the Chinese Room (CR) argument advanced by John R Searle, I examine whether his contention that the mind is not a computer program is true in the context of what constitutes artificial intelligence (AI) and thus the computational theory of the mind is false.  In order to do this I will establish the agreed basis of artificial intelligence, outline the Turing Test for evaluating machine intelligence, describe the CR thought experiment and then evaluate the Seale’s principle claim that the CR experiment is merely a syntactic process requiring no semantic understanding and demonstrating strong AI to be false.

In order to evaluate Searle’s CR argument we first need to have a clear understanding of artificial intelligence and in particular the form of AI, so-called ‘Strong AI’, at which Searle is directing his thought experiment.  Strong AI is the philosophical thesis that appropriately programmed computers have minds in exactly the same sense that we do.

During the 1950s, Alan Turing proposed a simple test to evaluate whether a machine is making an adequate simulation of the human mind.  The ‘Turing Test’ states that ‘if a computer can pass for a human in online chat, we should grant that it is intelligent’. 

This leads us to divide AI into four specific categories:-

                AI1         Computers are capable of thought;

                AI2         Only computers are capable of thought;

                AI3         A machine can think simply as a result of instantiating a computer program;

                AI4         Computer models are useful in the study of the mind.

Clearly AI4 is the weakest form of AI and is not considered to be particularly controversial. However, the AI argument builds from AI3 through to AI1 as the claims become stronger. Indeed if AI2 is true then we are all computers!  The combination of AI1 and AI3 suggest that a thinking machine is possible and all we need to do is write the appropriate program and run it to demonstrate a thinking machine (Wilkinson, 2005, pg 100). Strong AI of this form suggests that a suitably programmed computer can understand natural language and actually have other mental attributes similar to humans whose abilities they mimic. For this reason if the AI3 form of AI can be undermined the entire strong AI argument is invalid.  Consequently AI3 is the form of AI that Searle considers ‘strong AI’ and which his CR argument is intended to show to be false and hence the mind is not a computer.

Searle’s CR experiment is a thought experiment which imagines an English speaker who knows no Chinese, locked in a room full of boxes of Chinese symbols (the database) together with a book of instructions for manipulating the symbols (the program).  Imagine the people outside the room send in other Chinese symbols (data input) which, unknown to the person in the room are questions in Chinese.  By following the instruction book, the man is able to pass out Chinese symbols and answer the questions correctly (the output).  Searle contends that the man in the CR has actually passed the Turing Test for understanding Chinese without having to understand a word of Chinese.  Put simply the Searle argument may be sum up as follows:-

Premise 1.           If ‘strong AI’ is true, there exists a program for Chinese such that if any computing system runs that program, that system may be considered to understand Chinese;

Premise 2.           The program may be operated by anyone, without understanding Chinese;

Conclusion.         Therefore, ‘Strong AI’ is false (from premise 2).

Premise 2 conforms to the CR and as such the inevitable conclusion is that running a program does not constitute understanding.  Searle’s central claim is that the CR experiment shows that it is not possible for syntax to result in semantics.  Syntax in this context refers to the way in which the Chinese symbols are manipulated as opposed to semantics which relates to the meaning of the symbols.  Essentially the program is purely a syntactical symbol system which merely manipulates the symbols and entirely lacks semantic properties without any understanding of their meaning (Wilkinson, 2005, pg 105).

Before I consider the case against Searle’s CR experiment, it would be useful to describe the programming process.  For a process to be programmable, the process must be able to be constructed as an algorithm.  For a process to be algorithmic the following must apply:-

1.                   Every step is specifiable entirely without ambiguity;

2.                   At every step, there is no ambiguity about what the next step must be: no insight, inspiration or creativity is needed;

3.                   Provided each step is correctly executed, the procedure will produce the desired result in a finite number of steps.

All computers, whether based on traditional step-by-step von Neumann architecture or parallel processing, are programmed in an algorithmic manner (Wilkinson, 2005, pgs 100 - 101).  The key issue is the ability to ‘frame’ the question in an algorithmic format to enable the computer to work.  Searle considers the computer program based on the algorithmic formulation to be purely syntactic, therefore, unable to be semantic and hence computer programs cannot produce minds.  Searle’s contention may be shown as follows:-

Premise 1.           Programs are purely formal algorithmic processes (syntactic);

Premise 2.           Human minds have mental contents (semantics);

Premise 3.           Syntax by itself is neither constitutive of, nor sufficient for, semantic content;

Conclusion.         Therefore, programs by themselves are not constitutive or sufficient for minds.

Premise 3 supports the CR thought experiment.

Now there are a number of criticisms to Searle’s CR argument against Strong AI, however, they are essentially variants on the so-called ‘System Reply’.  Essentially the counter argument is that no single element should be considered as the ‘mind’ of the CR, and certainly the man at the centre running the process does not understand the Chinese language, but rather it is the whole system operating together that develops an understanding of Chinese and hence a semantic grasp of the Chinese symbols.  Searle replies that if the man in the room were to memorise the instruction book and the database of Chinese characters and become the entire system, he still would not be able to attach any meaning to the formal symbols even though he is now the entire system.

Others criticise the fact that the man in the room is deprived of any sensorimotor connection to the world and that these are vital missing factors.  Searle counters that the Chinese characters could be the outputs of a television camera and the outgoing symbols be the commands to a robotic arm. He now has a connection, but still no understanding.

Others suggest that the CR does not model the way the brain works as complex neural network.  Searle counters by applying Ned Block’s Chinese Gym thought experiment, whereby millions of people in a huge gym are connected to one another by walkie-talkie radios passing instructions to one another acting as individual neurons participating in a neural network.  On the other hand, the same issue regarding semantics applies in this case as well.  The Chinese Gym does not understand Chinese any more than the CR.

Daniel Dennett’s challenge to the CR is based on the issue of complexity.  He contends that Searle’s experiment is far too simple and that this is the reason for no apparent understanding being present within the system.  Dennett argues that with increasing complexity you will get unexpected results resulting in radical changes in behaviour known as emergent properties which only occur when a system is sufficiently complex.  Such radical changes in behaviour or properties of complex systems are common in the natural (Wilkinson, 2007, pg 113-114).  Consequently, Dennett argues that any system capable of conversing in Chinese would most likely be far more complex than the CR experiment and it would be difficult to say confidently that the computer system did not understand Chinese.  Searle contends that running increasing complex programs are no more than more complex algorithms and are just as syntactic as previously described and not capable of semantics however complex.  Whilst I agree that a complex computer program is just an assembly of multiple simple routines, increasing complexity often leads to increased computing capability and can result in unexpected capabilities and even outcomes.

In the same vein, the challenge posed by Patricia and Paul Churchland suggests that the issue is one of interpretation and speed, in that the CR experiment is operating at a very much slower speed than the brain operates and as such ‘understanding’ cannot be detected.  They use the analogy based on Maxwell’s theory of light being made up of electro-magnetic waves.  In their thought experiment, a man stands in a darkened room and waves a magnet up and down.  Although light is indeed made up of electro-magnetic waves, no light would be detected.  The man waving the magnet does not disprove Maxwell’s theory that light consists of electromagnetic waves.  The missing component is speed.  The Churchlands’ thought experiment slows down the waves to a range to which we humans no longer see them as light.  By trusting to our intuitions in the thought experiment, we falsely conclude that rapid waves cannot be light either.  Similarly, the Churchlands’ claim that Searle has slowed down the mental process to a range we humans no longer think of it as ‘understanding’.  Thus the Churchlands contend that the same applies to Seale’s experiment and that if we were to meet the man from the CR who seemed to converse intelligently in Chinese, but was really deploying millions of memorised rules in a fraction of a second, it is not so clear that we would deny he understood Chinese (Pinker, 1997, pgs 94-95).

In conclusion, in consideration of the narrow interpretation of Searle’s claim that the mind is not a computer program I would accept his argument to be valid and that the computer program itself does not ‘understand’ as we interpret that word.  I found Seale’s dismissal of the counter arguments to be reasonable, with exception of Dennett’s and Churchlands’ arguments. The Churchlands’ thought experiment it is based on a very simple premise and easy to understand and Dennett’s emergent properties argument compelling.  Speed and complexity could be key factors in strong AI.  What is undoubtedly true is that the ‘Turing Test’ is no longer a sufficiently subtle evaluation of artificial intelligence.  Progresses in computing have advanced to a stage whereby it is entirely possible to converse with a computer and believe you are conversing with a human being.  The key issue in achieving a true computer ‘mind’ comes back to the framing issue.  This relates to the ability to program algorithmically beyond the essentially mathematical and logical tasks to consider such areas as intuition, belief or even love.  These functions of the mind are frequently considered irrational and illogical, but they are what make us human. The ability or functionality to program ‘illogically’ is probably beyond algorithmic programming of digital computers as we know them today.  Equally Searle’s argument hinges on our understanding of the semantics of language.  What do we mean by ‘understanding’ or ‘meaning’?  Or finally, is it the limitation of English as a language which is unable to provide an adequate explanation of differences and similarities between the mind and artificial intelligence?  Regardless, Searle’s valid dismissal of strong AI will not slow the pace of computing development and the likely move from the physical limitations of silicon-based computer architectures to bio-computers which utilise biologically derived molecules to perform computational processes.  The future of ‘strong AI’ probably lies in the rapid development, dare I say growth, of such bio-computers!

References

Wilkinson, RJ (2007) ‘Chapter 3 Monism (Conclusion) and Artificial Intelligence (Beginning), Robert Wilkinson Minds and Bodies, 2007, Open University Press.

Pinker, S (1997), Chapter 2, ‘Thinking Machines’, Steven Pinker, How the Mind Works, W.W.Norton & Company Ltd., 10 Coptic Street, London. WC1A 1PU

Copyright© John Tomany 2012