Bill Joy, Please please don't anthropomorphize AIs Bill joy's essay "why the future doesn't need us" invokes various mistaken assumptions about the nature of AIs. Eliezer Yudkowsly has a presentation up on SIAI which makes for a good antidote...
phane2 votesWednesday, February 27, 2008 at 12:42 PM
I think it might still be anthropomorphizing too much to even talk about 'minds' at all. We will simply have software patterns that can accomplish tasks that used to require human intellectual work. It will not come in discrete personalities the way humans do, unless we decide to design it that way.
I agree that there will be a spectrum of different AIs in all different types and shapes. But I do feel compelled to point out the possibility of another sequence of events that could lead to a "race" of AIs (or a tribe as you describe it). Take the human race for example. In the long history of evolution many different kinds of intelligences emerged and continued to exist together until the human intelligence emerged and quickly took over the earth due to it's superior intelligence and desire to stay in control. Now, I think it's possible that as we build many different types of AIs that at some point a certain type of AI is built (or emerges) that has superior intelligence to ours along with the desire to spread it's own kind and be in control. If that happens, then it's a possibility that that kind of AI will replicate itself as fast as it can and try to stop anything that stands in it's way. So, my point is, Bill Joy's ideas are not completely off the mark, and the future might indeed not need us unless we take the right steps in the present to ensure our survival.
rmijic1 votesThursday, February 28, 2008 at 3:12 PM
"at some point a certain type of AI is built (or emerges) that has superior intelligence to ours along with the desire to spread it's own kind and be in control. If that happens, then it's a possibility that that kind of AI will replicate itself as fast as it can and try to stop anything that stands in it's way."
- true. But I don't think that this is what we should worry about most. It seems more likely that uFAI (unfriendly AI) will come from an AI doing what we tell it to do, rather than what we actually want it to do. Eliezer has, of course, written on this.
this particular failure mode (the kind of "survival of the meanest" failure mode, that is) is more likely to befall iterated human enhancement methinks.
rmijic1 votesThursday, February 28, 2008 at 3:15 PM
but, in any case, joy should be very clear that not all AIs will be the same, "a priori", and I think he is in quite a muddle about it. The essay is clearly very popular, as a critic of transhumanism I met in the college canteen sent the link to me.
I feel that Joy's essay needs a thorough rebuttal writing. I may do this over the summer. Has it already been done?
I don't know if a rebuttal is the right course. How about an explanation of why Joy's fears are unfounded, or a "proposed plan" to avoid the potential risks of self-advancing AI.
ok, I can see where you're coming from - maybe a rebuttal would be seem too argumentative. Anyway, I'm really pushed for time at the moment, what with exams coming, and applications to make for next year,... so it definitely won't happen for another couple of months!
The proposed plan would be that we really, REALLY need to do more research into the problem, and that more institutions like SIAI should be set up.
philosophistry1 votesWednesday, March 05, 2008 at 1:24 PM
We're also a natural class because we've evolved that way. We've evolved to give a damn about each other. Some of us don't care about our entire species even; their only tribe their family. Technology isn't undergoing the same "care about things like me" evolution that we had gone through
If anything, consider the opposite construct, that technology is so linked to our survival. Most technology created only exists because it benefits human interests.
