Path: utzoo!censor!jeff From: jeff@censor.UUCP (Jeff Hunter) Newsgroups: comp.ai Subject: Re: Question on Chinese Room Argument Summary: some new thought experiment Message-ID: <387@censor.UUCP> Date: 6 Mar 89 08:23:26 GMT References: <4298@pt.cs.cmu.edu> Organization: Bell Canada, Business Development, Toronto Lines: 62 In article , harnad@elbereth.rutgers.edu (Stevan Harnad) writes: > Counterargument: To ascertain (beyond reasonable doubt) that a system > CANNOT understand, you don't need a theory. Searle's argument is a case > in point: If Searle (or you, or me) does exactly what the computer does > but does not understand, then the computer does not understand. I'd like to try another thought experiment to see if I can find out why Mr Harnad and I disagree. An assumption of the Chinese room argument is that a set of rules that manipulate symbols can be built that passes the (Linguistic) Turing Test in Chinese. (ie. it can convince Chinese speakers that it too is a human Chinese speaker.) I make the further assumption that modelling a human on an atom-by-atom level would be sufficient to reproduce that human's verbal behaviour. (Gilbert Cockton will probably disagree with me on this :-), but I don't think that Steven Harnad does.) Of course to keep the simulation on track it must get the same input (from a real or simulated environment) as the real human gets. So, to start off this experiment we take Mr. Harnad into a quiet, dark room (to minimise the environmental factors), sit him in a chair, and let him get comfortable. We then take the Thought Experiment Matter Mapper (tm) and record the positions of his atoms to the precision allowed by Quantum Mechanics. We then ask "Do you understand English?" to several systems derived from this map. a) the original Mr. Harnad b) a duplicate Harnad made from new atoms in the same pattern c) a "functionally duplicate" android Harnad made using some other chemistry such as silicon molecules d) a special purpose computer hardwired into the shape of Harnad's neurons, etc e) a general purpose computer simulating Harnad's atoms moving according to the laws of physics f) an English speaker simulating e) above g) a Chinese speaker simulating e) above h) Mr. Harnad simulating e) with pencil marks on paper i) Mr. Harnad simulating e) after having memorised the entire program and database (shades of the halting problem :-) Presumably a) will answer "yes", and if the others are good simulations they will also answer "yes" as well. (To forestall the posting that there are quantum/chaos/rounding error limits to the quality of the simulation I point out that we're only running for a few (simulated) seconds, and that we can always run b) thru i) several thousand times and check for averages.) Now would Mr. Harnad be kind enough to state which of these systems really understands English (according to his definition). It seems clear that it is impossible for the humans in f) thru i) to understand the symbols they are manipulating due to sheer volume. Each would be justified in saying that they did not understand what the program thay were simulating was doing. e) thru i) are clearly only processing symbols. By my reading of Harnad's posting to date he believes they do not understand. But then, to paraphrase him, "If a computer does what Searle (or you, or me) does, but does not understand, then we do not understand". -- ___ __ __ {utzoo,lsuc}!censor!jeff (416-595-2705) / / /) / ) -- my opinions -- -/ _ -/- /- The first cup of coffee recapitulates phylogeny... (__/ (/_/ _/_ Barry Workman