On 9/14/06, Ben Goertzel <[EMAIL PROTECTED]> wrote:
However, I am not so sure this is the most sensible approach to take.... The details of my own personal Friendliness criterion are not that important (nor are the details of *anyone*'s particular Friendliness criterion). It may be more sensible to create an AI with a more abstract top-level goal representing more abstract and general values....
Hi Ben, After reading KnowabilityOfFAI and perhaps coming to an Awful Realization, it seems Friendliness is plausible with strict criteria for an optimization target. It also seems an optimization target is necessary, regardless, with more or less strict criteria. What I think would be illuminating are some characterizations of these optimization targets from the ascendant researchers. These characterizations might show which optimization target seems Friendlier than the others and if it accordingly requires the strictest criteria. Incidentally, it recently sort of dawned on me that SIAI refers to developing an 'AI' rather than an 'AGI'. This seems to suggest the plan to build a relatively narrow AI, albeit one toward powerful optimization. I genuinely wonder if this is an approach worth considering more deeply, to the extent it's possible. I'll understand if this would be too much to get into sufficiently, but I do wish to accentuate my interest in having a little better idea about some of these planned optimization targets. Best, Nate ----- This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
