X-Message-Number: 11703 Date: Sat, 08 May 1999 14:53:14 -0400 From: Daniel Crevier <> Subject: consciousness in simulated characters, again. References: <> Mike Perry answered the following to Tom Donaldson, in message #11691, about what characteristics make a virtual object in a computer program aware or not. >I can't give necessary and sufficient conditions, guaranteed valid, but >it's not hard to give a hypothetical example, invoking as usual some >future possibilities. Think of a robot that can see and hear and speak, >who responds like a human, *and* whose internal processing corresponds >isomorphically to the processing in our own brains when we are >conscious and doing things. Such a construct has a strong claim to >being conscious in my view, and even at a human level. I think it's possible to be a little more specific. The classical position in philosophy holds that consciousness implies the ability to represent and reason about one's own mental states. For example, a conscious being should be able to explain the reasons for its actions. A classical example of this in computer science is the program SHRDLU, written by Patrick Winston at MIT in the early 1970's. It manipulated simulated geometric objects at the request of a human user, and could answer questions about its motivations, as the following dialogue excerpt illustrates. (The program's statements are capitalized). Why did you pick up [the green pyramid]? TO GET RID OF IT. Why did you do that? TO CLEAR OFF THE RED CUBE. Why did you clear off the red cube? TO PUT IT ON A LARGE GREEN CUBE. Why did you do that? TO STACK UP THE RED CUBE, A LARGE RED BLOCK AND A LARGE GREEN CUBE. Why did you do that? BECAUSE YOU ASKED ME TO. SHRDLU may thus have been the first conscious computer program, even though the details of its programming were definitely not homo- morphic to the human brain. We can however surmise that it was much less conscious than a human being because its internal states and the kind of reasoning it could perform about them were much less complex, by many orders of magnitudes, than our own. By the way, most so-called 'intelligent agents' that are all the rage in AI nowadays are not conscious by this measure. For example programs of the Eliza type, which aim at passing a simplified version of the Turing test (see http://www.loebner.net/Prizef/loebner-prize.html) are just bags of tricks with no representation of their internal states, even though they can at times make halfway sensible conversation. Daniel Crevier Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=11703