Path: utzoo!attcan!uunet!mcvax!ukc!strath-cs!glasgow!gilbert From: gilbert@cs.glasgow.ac.uk (Gilbert Cockton) Newsgroups: comp.ai Subject: Re: Question on Chinese Room Argument Message-ID: <2483@crete.cs.glasgow.ac.uk> Date: 28 Feb 89 12:14:01 GMT References: <4298@pt.cs.cmu.edu> <8174@netnews.upenn.edu> <51123@yale-celray.yale.UUCP> Reply-To: gilbert@cs.glasgow.ac.uk (Gilbert Cockton) Organization: Comp Sci, Glasgow Univ, Scotland Lines: 58 In article <51123@yale-celray.yale.UUCP> engelson@cs.yale.edu (Sean Engelson) writes: >(Searle + rules) understanding of Chinese? It seems that to >demonstrate or refute the position of understanding being demonstrable >purely through I/O behavior, one must have an effective definition of >understanding. By effective I mean one that does not beg the >question, i.e. by defining understanding to be symbol-processing, or >conversely, to be that which humans do. Sorry, but your constraints are a little weird. Understanding *IS* what humans do. It *MAY* involve symbol processing. What question is begged? What is an effectiver definition? Look how well physicists manage with "force", "charge", "gravity". You cannot ask commentators on humanity for "definitions" that are any less (fast and) loose than those used by commentators on nature. Understanding involves more than lexicography. Let's just define "understanding", no constraints. Stevan Harnad has already pointed to two senses a) the feeling of understanding b) the attribution of understanding. For some domain where something can be right (a) involves thinking that you know what "right" is (b) involves someone else deciding that you know what "right" is (a) is not wholly like pain, but like pain, its perception is a wholly internal event. Where objective tests exist, understanding (a) can only be wrong in the sense of the content of the understanding, as can understanding (b). In both cases, the experience of understanding does not wither away in the face of a failed test. Understanding is monotonic in this sense. Once asserted, the act of assertion is unchangeable, and years after (as we've seen from postings) we can remember just how we felt. (a) is also accompanied by mood changes (elation, nausea etc.). These are probably measurable in some physiological sense. Such measures will be orthogonal to performance on objective behavioural tests. Explain that one. On Searle's room, Searle would not understand Chinese, but neither can the system, since it only "knows" how to understand problems about Chinese and how to output it. There is nothing in (Searle + rules) which asserts "I understand" in response to each problem put to it. The rules just run, and no honest user of the English language would ever attribute understanding to a bunch of rules. As far as more effective computer systems are concerned, it doesn't matter either. The point is one of intellectual honesty, and the distaste felt when groups of supposed academics in a liberal culture fall under the control of a shallow ideology. The question for the strong AI brigade is: "Given the normal usage of understanding, what grounds are there for attributing it to computers, and why bother anyway" While we're at it, what about those halucinogenic thermostats with beliefs. Whatever it was, don't eat it again :-) -- Gilbert Cockton, Department of Computing Science, The University, Glasgow gilbert@uk.ac.glasgow.cs !ukc!glasgow!gilbert