thom@dewey.soe.berkeley.edu (Thom Gillespie) (03/03/89)
Dear Jim, Eliza worked because it tried to simulate a Rogerian shrink, there was no domain , a wonderful area for an expert system wouldn't you say Jim? Are you trying to carry on the tradition? Read "computer Power and Human Reason : from Judgement to Calculation" by Weizenbaum. Better yet , why not write an expert program to understand it for you. Thom Gillespie while you and I have lips and voices which are for kissing and to sing with who cares if a one-eyed son of a bitch invents an instrument to measure spring with - - ee cummings
curry@hplchm.HP.COM (Bo Curry) (03/07/89)
>/ hplchm:comp.ai / geddis@polya.Stanford.EDU (Donald F. Geddis) / 12:47 am Mar 4, 1989 / >True enough, but then you are defining the "computer" to be the dumb processor >that interprets the rules. No one claimed that the processor (by itself) >*did* understand. I still haven't heard a satisfactory rebuttal to the >"Systems Reply", namely that (Searle + Rules) understands, whereas just >(Searle) doesn't. [To use your analogy: (Computer Processor + Symbolic >Rules) understands, but just (Computer Processor) doesn't.] > OK, let's follow up on this. Let me see if I can paraphrase Harnad interpreting Searle: 1. Searle does not understand Chinese. I agree, because *not only* does he so claim, *but also* he fails to pass the Chinese TT. Although neither of these facts alone would suffice (since the first would not convince me, and the second would fail to convince Harnad), together they prove the case. 2. The Room (Searle + rules) passes the TT and *seems* to understand Chinese. I agree. 3. However, the Room doesn't *really* understand, since "Searle is doing everything [the Room] is doing, and *he* claims not to understand" (!). I don't see how 3. follows at all from the rest of the argument. It obviously has a powerful intuitive appeal to Harnad and others. Let me try to show how intuition is misleading these deep thinkers. The foundation of the intuition that Searle's opinion about the Room's understanding is definitive seems to be: 4. Other than Searle, there is *nothing* in the Room (except for blackboard, chalk, and a few slips of paper upon which the rules are inscribed). Since we all know, from our experience, that blackboards and slips of paper don't understand anything, nothing is left except Searle himself. Searle has suggested that even these meager props would be unnecessary, since he could *memorize* the rules! Now, let's look at this intuitive argument in a bit more detail, to test its plausibility. Perhaps, in his mind's eye, Searle sees the rules as consisting of a Chinese dictionary, a Grammar, perhaps a Thesaurus and a few syntactic rules of grammar such as "Never split an infinitive", or some such. In any case, Searle's (envisioned) set of rules is clearly compact enough that he imagines *memorization* of the rules to be possible. This idea (or even the idea that the rules, written on paper, might fit into a room) is, I would submit, ludicrous to anyone who has actually attempted to design a program to understand natural language. Consider what it would require for a Room to be able to pass the TT. It must be able to use language well enough to convince a native speaker. It must therefore know the denotations, connotations, and normal (physical, cultural, etymological, etc) associations for a large subset of Chinese words, phrases, proverbs, etc. In order to carry on a conversation about, say, riots in Tibet, the Room would have to understand Tibetan Buddhism, its relationship to Chinese Buddhism and to Confucianism, the political and historical relations between Tibet and China, and thousands of other facts and relationships. All this knowledge, for thousands of possible topics of conversation, must be represented in the rules before the Room can hope to satisfy Searle's premise. I submit that this knowledge base is essentially isomorphic to a large part of Searle's brain, and that it would be clearly impossible for him to "internalize" it (as explicit memorized rules). Think about the size and complexity of these rules. Will they fit in a room? If punched out on Hollerith cards and laid end to end, would they reach from Earth to Sirius? How many centuries will Searle require, while interpreting these rules sequentially, to respond to the simplest question posed to the Room in Chinese? Once the complexity of the hypothesized rule set is fully grasped, it becomes clear that intuitions about the "obvious" lack of understanding embodied in "a few slips of paper" are seriously misleading. Admit that this intuition may be mistaken, and Searle's (and Harnad's) argument disappears. Bo (still waiting for the PC version) Curry curry%hplchm@hplabs.HP.COM
gilbert@cs.glasgow.ac.uk (Gilbert Cockton) (03/13/89)
In article <280003@hplchm.HP.COM> curry@hplchm.HP.COM (Bo Curry) writes: > Admit that this intuition may be mistaken, and Searle's > (and Harnad's) argument disappears. Why should I? Admit that gravity may not exist, and then what disappears? I think you should argue your case. If people's "intuitions" say the room/rules have no understanding, then why doubt them? What's the gain? If your AI systems "work", all well and good. But don't demand that people call black white in the process. If AI folk spent less time trying to redefine everyday language, people might trust them more. There is no quicker way to lose people's trust than to abuse language. -- Gilbert Cockton, Department of Computing Science, The University, Glasgow gilbert@uk.ac.glasgow.cs <europe>!ukc!glasgow!gilbert
throopw@agarn.dg.com (Wayne A. Throop) (03/21/89)
> gilbert@cs.glasgow.ac.uk (Gilbert Cockton) > If people's "intuitions" say the room/rules have no understanding, > then why doubt them? What's the gain? Well... many "people's intuitions" say that Eliza already understands. Why doubt *them*? The gain in doubting people that think Eliza understands is that we don't cheapen what we mean by "understanding". The gain in doubting that people that think the CR shows that the room/rules have no understanding even in principle is that we don't arbitrarily anthopomorphize what we will accept as an "understanding entity". > If your AI systems "work", all well and good. But don't demand that > people call black white in the process. If AI folk spent less time > trying to redefine everyday language, people might trust them more. This situation doesn't arise in the CR. In fact, the CR's premise is that "people's intuition" from outside the room leads them to think the room understands, and "people's intuition" once they've seen inside the room leads them to think otherwise. So, we aren't asking to call black white. We are asking whether black should be defined functionally (in terms of the light it reflects) or structurally (in terms of which pigments it is constructed of). -- "Who would be fighting with the weather like this?" "Only a lunatic." "So you think D'Artagnian is involved?" --- Porthos, Athos, and Aramis. -- Wayne Throop <the-known-world>!mcnc!rti!xyzzy!throopw