X-Message-Number: 8249 Date: Tue, 27 May 97 14:08:01 From: Mike Perry <> Subject: Artificial Consciousness The discussion continues on consciousness, and in particular, whether it could reside in an artificial device or needs an organic substrate. The doubters that things we have done so far, such as a programmed robot or a computer program, exhibit the slightest degree of consciousness or feeling, seem once again to be overlooking the forest for the trees. Just because a machine crunches bits at the lowest level in no way (in my view) precludes its having feeling and consciousness, or more properly, the system it is running might have feeling and consciousness. There is an amusing discussion along these lines in *Godel, Escher, Bach* by Hofstadter, pp. 314-333, about Aunt Hillary, an ant colony which is intelligent and can understand things at the human level even though the ants that make her up cannot. This is very similar to Searle's Chinese room, if I understand it right. The system as a whole can have properties not possessed by its parts. The idea of trying to build "feeling" into a computer system would make a great topic for research, in my estimation (speaking from the standpoint of having a PhD in computer science that is, though I haven't been too active in the field in recent years), provided, of course, we aren't worried too much about the "civil rights" issue. (This is a non-trivial issue, though, if we are going to take seriously the prospect that our computer programs might be able to experience pleasure or pain.) I remember that, some years ago, Terry Winograd did a dissertation on a program that conversed in English. With great difficulty, it was possible to design a program that could converse reasonably well about a "blocks world" type of toy universe. But this creation could not be extended in any easy way--we do not today (a quarter-century later) see programs that converse in fluent English about the real world, and seem just like normal humans (i.e. pass the Turing Test). As far as I can tell, Winograd approached his project as one involving *language* directly and not a system that *uses* language because of wants or needs it has. That latter approach would be quite fruitful, I think, and may have been attempted a few times already, but probably we could do better if we tried harder, without having to wait for more advanced technology. As a start, I would advocate thinking about goal-seeking (or avoiding) systems. These can be created artificially, and they certainly bear a resemblance to systems in the natural world (biological organisms) that we credit with consciousness. We might try to create a reasonably general theory of goal-seeking systems. Some questions then come to mind. What are the kinds of "goals" we would study? For instance, we might focus on the outside world, but that is a big place and, I think, the wrong place to start looking if we want to study consciousness. We should instead look inside our prospective conscious device to see what is happening there. What is "goal-seeking" from an internal standpoint? Is it reducible to maximizing a number stored in a certain, designated place, for instance? If not what additional features should we consider, if we used that idea as a starting point? etc. Again, I'm sure these thoughts have occurred to other people, and I'd guess there is already a substantial literature. In fact, I'd be interested in references anybody may have. Mike Perry Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=8249