tskelly@ccvax.ucd.ie (11/28/90)
I don't know if the following has occurred to anyone and been dismissed or thrashed out and finished with or what but please bear with me! An implicit assumption in Searle's Chinese room seems to be that a set of symbol manipulation rules could somehow manage to produce intelligent answers to questions which merely involve symbols! When you think about it, what I'm talking about is NLP - to which, as far as I'm aware, there is no final solution. Examples never come easy but I hope the following will illustrate : The question (in English, assuming an English room again - :-) ) "Why did Napolean march on Russia?" Intelligent beings will almost always make some sort of answer to this question even if they have never heard of 'Napolean'. One would expect that the most common answer would be to the effect of "Because he wanted to command a larger empire!" However, clearly the question is open to interpretation based on the meanings of the words (symbols) used. ^^^^^^^^ For example, I could be asking "Why did he 'march' as opposed to 'saunter'?" or "Why did he 'perform a particular walking motion on land denoted as Russian' as opposed to 'moving to fight against people of Russian nationality'?" Clearly, the answer lies in the interpretation of the symbols and no one interp. is necessarily valid at a formal level (in fact symbolic manipulation does not involve 'interpretation of symbols!). So, recognising this, it should be obvious taht the English room could not answer the question! (OK?) Now, one could argue that a valid rule in the room would be "If there is no string of symbols which can be produced by manipulation of the input symbols then return 'I don't know.'" An interrogator in the TT could and should force the participants to reply to questions to which it wants to say 'I don't know.' with answers that are based on assumptions about the components of the question. There is no way the English room could make assumptions as by nature assumptions are entirely based on the meaning of the object symbolised and have nothing to do with the symbol itself! So, to distill the point :- surely a good Turing Test would involve testing for aspects of human intelligence such as 'assumptions' the Chinese room would not pass this part of the test (in the sense that the CR can't make assumptions while a normal human can.) And when I say "aspects of intelligence such as 'assumptions'" I mean to imply that this is only one of the many different attributes which are a necessary part of intelligence but which go beyond symbol manipulation!! Thinking more about Searle's chinese room, (apparently) he was proposing the assumption that the Chinese room 1. Was purely a symbol manipulator 2. Could pass the Turing Test If this is the case then, by the above argument, his assumption was self- contradictory (ie. a symbol manipulator could not pass the TT). Of course a lot depends on how the TT is performed (some of the articles under this topic debate the TT and it's definition - I don't want to repeat what's been said - but I do think there are definitions of the TT which would be seen to be unacceptable but still allowed under Turings general description) Having said all of that I would like to say that I take Searle's point that one could develop a symbol maipulation *machine* which could mimick to a certain degree some of the activities which we call 'intelligent' but which clearly could not be said to be intelligent! There is more that could be said (and no doubt already has been said) about Searle's CR and it can be used to illustrate many points other than that which it was created to do but sometimes there appears to be too much read into it! Incidently, I received some good replies to my article "Chinese room and stuff" which are well worth going into and, in fact, I am composing an article dealing with these which will appear soon ( in other words I haven't forgotten :-) ) Stephen Kelly TSKELLY@CCVAX.UCD.IE