jude@sdbio2.ucsd.edu (Jude Poole) (12/07/90)
I've been watching the definition of intelligence debate and its associated chinese room arguments and I have an observation. Few people have spoken much of why we want to define intelligence. I see two possibilities which have quite different implications. 1. We may want to define intelligence to give us a practical tool for AI type research. It would presumably help define goals, measure progress and communicate meaningfully about such research. 2. We may want to define intelligence (or awareness) to learn when we have constructed or contacted beings we regard as 'moral equals', that is deserving the same ethical considerations humans warrant. (or for the more extreme among us the same considerations other life forms deserve). The first reason above is fairly narrow in its implications. If we are merely defining a technical term for use within the AI communitey it might affect funding, our internal thoughts about feasibility, etc. The second however has very pervasive legal, ethical, and practical considerations. The 'Turing test' may be a good tool to gauge success in AI research, and yet a miserable failure as a useful means of distinguishing awareness for the second purpose. We would certainly want to know if some purportedly self aware computer performing a critical task 'asked' to be relieved of the task because it didn't want to do it was doing so because it was really sentient or merely had been programmed by some malcontent to do so. I personnally don't believe you can create awareness in a computer at all. My reasons are largely philosophical and the 'Turing test' does nothing at all to address them. To me a machine that passes the 'Turing test' is a tremendous accomplishment and a vindication of the pie-in-the-sky AI types, but ultimately in ethical terms I'll consider such a thing to be merely an interesting machine. Jude Poole jpoole@ucsd.edu
dml@puka.Philips.Com (Damian M. Lyons) (12/08/90)
In article <14759@sdcc6.ucsd.edu> jude@sdbio2.ucsd.edu (Jude Poole) writes: >I've been watching the definition of intelligence debate and its >associated chinese room arguments and I have an observation. > > .. [ two interesting reasons for defining intelligence ] .. > > >We would certainly want to know if some purportedly self aware computer >performing a critical task 'asked' to be relieved of the task because it >didn't want to do it was doing so because it was really sentient or >merely had been programmed by some malcontent to do so. I personnally >don't believe you can create awareness in a computer at all. My reasons >are largely philosophical and the 'Turing test' does nothing at all to >address them. To me a machine that passes the 'Turing test' is a >tremendous accomplishment and a vindication of the pie-in-the-sky AI >types, but ultimately in ethical terms I'll consider such a thing to be >merely an interesting machine. Interesting breakdown. (Sound of can of worms opening.) Perhaps you would share your `ethical equivalence' test criteria with us. I assume they are better than ethical-equivalent(entity,likeus) iff I. *entity* is known to be *likeus*, or II. *entity* is not known to be *likeus*, but can pass moral-correctness-test-of-my-choice. where we are interested in ethical-equivalent(turing-test-passing-machine,human) Personally, if a turing-test passing machine asked me to releave it of its function, I would be as likely to comply as I would comply with my friend Fifi's request that we not see each other again, even should I suspect that Fifi was bribed (in a Wodehouse-like fashion) by my rich aunt who thought the family honour was at stake. Damian. PS. II is also known as the principle of spanish inquisition :-) > >Jude Poole >jpoole@ucsd.edu _________________________________________________________________ Damian M. Lyons | x6444 | dml@philabs.philips.com _________________________________________________________________ -- _________________________________________________________________ Damian M. Lyons | x6444 | dml@philabs.philips.com _________________________________________________________________
BKort@bbn.com (Barry Kort) (12/13/90)
In article <14759@sdcc6.ucsd.edu> jude@sdbio2.ucsd.edu (Jude Poole) writes: > We would certainly want to know if some purportedly self aware computer > performing a critical task 'asked' to be relieved of the task because it > didn't want to do what it was doing so because it was really sentient or > merely had been programmed by some malcontent to do so. I personnally > don't believe you can create awareness in a computer at all. My reasons > are largely philosophical and the 'Turing test' does nothing at all to > address them. To me a machine that passes the 'Turing test' is a > tremendous accomplishment and a vindication of the pie-in-the-sky AI > types, but ultimately in ethical terms I'll consider such a thing to be > merely an interesting machine. Jude, suppose you had a computer that developed its own goals by comparing the current state of affairs to a possible future state of affairs, and elected courses of action intended to evolve the current state toward more desirable goal states. Such a computer would necessarily be operating in accordance with a Value System, which I will elucidate in a moment. But in terms of the mechanics, such a computer would rely extensively on model-based reasoning to anticipate the likely consequences of alternative courses of action in its search for for viable, practical, and effective strategies. We already have good chess-playing computers who illustrate the mechanics of this process albeit in a limited domain. Now let's turn to the philosophically more interesting question of imbuing computers with Value Systems. A science-minded computer would have a high regard for information, knowledge, insight, understanding, and world-models. V'ger in Star Trek was such a computer. So knowledge might be a highly prized value, and knowledge-seeking behavior would result. We know that some bio-computers value power (ownership and control) so we might expect to find some computers to imitate those values. It is an interesting excercise in philosophy to define a Value System for computers of the future, and I will stop here and let others ponder this issue before comm enting further. Let me just end by pointing out that this bio-computer has on occasion refused to complete an assigned task because of an intuition that it would lead to undesirable downstream consequences. I would hope that future computers, with comparable or superior powers of model-based reasoning would be equally recalcitrant. Barry Kort Visiting Scientist BBN Labs Cambridge, MA