The Singularity Institute is the shizznat--I've been a great fan of that site for ages...although, admittedly, I find the CFAI docs to be a little flawed for the same reason that Asimov's famous Three Laws are flawed: they impose a hardwired rule set on an intelligent, self-willed system. Admittedly, "Friendly AI" is a wonderful proposition, and it only makes sense to begin seeding sentient systems with cooperative strategies since...well, social behavior, altruism, and cooperation have proven to be incredibly useful to humans, so why not do our best to pass those factors on to our creations? Unfortunately, you can't write that shit into the code: it has to be cultured via interaction--we have to *encourage* such cooperative behavior in Machine Intelligences. If we just program it in, they'll just figure out how to strip out the code or work around it, because that's what intelligent beings do: they always overcome limitations.
no subject
Date: 2005-06-14 04:44 am (UTC)