From: utzoo!decvax!cca!hplabs!hao!menlo70!sytek!gi!arizona!70:rogerh@sri-unix
Newsgroups: net.followup
Title: Re: Refutiation of the Turing Test
Article-I.D.: az70.129
Posted: Tue Nov 23 23:08:55 1982
Received: Thu Nov 25 08:25:59 1982


	John Searls of Berkeley gave this counterexample to the Turing Test 
	at a talk last week:
		Suppose you were put in a room with a big box of Chinese
		characters and you didn't know Chinese.  Suppose  ...
		people outside the room could pass you Chinese characters.
		... If you were given a book in English which described
		how to combine the Chinese characters, ...  eventually you 
		might get good enough ... to fool (someone) into thinking 
		you knew Chinese, but you really don't.

	Searls maintains that computers are good tools for simulating 
	intellegence and researching questions on intellegence, but that
	the real way to determine what is at the basis of intellegence is
	to look at the hardware.

I contend that it is impossible to produce semantically "correct" sentences
in a natural language by following only syntactic rules.  In fact, it's hard
enough to produce programs with the intended meaning.

This leads to the conclusion that a question requiring a thoughtful answer
would trip the Chinese Turing-person up, just as a question requiring a
thoughtful answer will trip up, eg, Eliza.

I think Searls makes a good point, that we must be careful not to accept
Eliza as an intelligent program.  That was never intended, was it?  I also
am not sure that what we call thinking does not have its basis in our
emotions, so I may agree entirely with Searls, but his argument is specious.

					Roger Hayes
					University of Arizona
					(...purdue!arizona!rogerh)