X-Message-Number: 30159 From: Kennita Watson <> Subject: The dangers of AI Date: Mon, 17 Dec 2007 11:24:31 -0800 In recent discussion of the dangers (or lack of dangers) of AI, there seems to be a lot of discussion of "it", as though there would be a *single* artificial intelligence, generated like Colossus (of Colossus: the Forbin Project, in case you're unfamiliar with it) and unleashed wholesale on a completely unsuspecting and unsophisticated populace. I think, rather, that there will be thousands, even millions, of AIs, developed in parallel in offices, home offices, gaming rooms, etc. worldwide. Since they are self-modifying, they will have widely divergent motivations, and like the Transformers (I've only seen the movie; try Netflix or Wikipedia if the pop-culture reference completely loses you), not all of them will be inimical to human interests. Likewise (leaving behind the Transformers), not all of them will be similarly easily duped by a given AI's persuasive powers (note that they will have human-*level* intelligence, not necessarily human-*like* intelligence). Proto-AIs will probably be programmed to perform tasks for humans, and for any given PAI, one of those tasks may well be some version of "warn me if I'm about to do something bad for me, or if bad things are happening/about to happen" (if you had one, you'd want it to do that, right?). Myriad definitions of "bad", myriad varieties of warning, myriad capabilities, etc. will make it all but impossible that a single uber-AI will be able to take over everything/everyone. The problems of AI will be approached from many directions, because there will be so much to be gained, even from AIs with far less than human intelligence. I say: open-source AI research! Live long and prosper, Kennita -- Vote Ron Paul for President in 2008 -- Save Our Constitution! Go to RonPaul2008.com, and search "Ron Paul" on YouTube Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=30159