This is also something which has baffled me for a long time. I've been mucking around with AI and related software since I was a kid, and as the years went by it gradually dawned upon me what the real problems were which needed to be overcome if any significant progress was going to be made. One of the really important things in my opinion is just being able to see the world at a reasonably high fidelity and level of confidence, so that you can then begin to make some decisions about what to do and how to move around.
There's been very little done, both in the academic and industrial arenas, on trying to produce a good all-purpose vision system which you can just bolt onto the side of some machine and have it be able to see what's in front of it. I think many people have shied away from such a task because it's traditionally regarded as difficult, and instead there have been many narrow-AI type vision systems employing an assortment of dodgy heuristics to pick out limited aspects of the visual scene.
However, producing a system which just reverse engineers camera images and produces good 3D models isn't impossible. In fact, it's quite straightforward, and doesn't involve having to resort to any particularly exotic algorithms - just regular geometry and probability theory. People such as Moravec have been working on this kind of stuff for years, and it's surprising how little attention this approach has received.
However, I think the good news is that soon there will be very generic vision systems of the bolt-on kind, which will work within a wide range of environments. I see this as a critical component which will facilitate the eventual realisation of AGIs capable of operating in the real world.
To go back to the original question I think the answer is that if you want to be a good researcher, keep paying the bills and work your way through the ranks you should avoid difficult or unconventional subjects. Stick to stuff that's easy and don't try to rock the boat and you'll have an easier life. Also there's the peer pressure factor. If some notable professor X comes out and says that general purpose vision is a problem to which there are no solutions on the horizon, then lesser mortals are likely to assume he knows what he's talking about and avoid getting involved with a subject which is unlikely to yield results.
- Bob
On 13/09/06, Joshua Fox <[EMAIL PROTECTED]> wrote:
I'd like to raise a FAQ: Why is so little AGI research and development being done?The answers of Goertzel, Moravec, Kurzweil, Voss, and others all agree on this (no need to repeat them here), and I've read Are We Spiritual Machines, but I come away unsatisfied. (Still, if there is nothing more to say on this question, please do the AGIRI-equivalent of sniping this thread immediately.)
I respect existing AGI researchers, but I am surprised that more members of the "establishment" are not on board. I just can't believe that , for example, almost all leading computer-science/cognitive-science professors are herd-following closed-minded stuck-in-the-muds. The leading universities do have their share of creative, free-thinking, inquisitive people, and the same goes for other parts of the "establishment".
To clarify what I am looking for, I should describe a recent conversation. I spoke to an open-minded and intelligent friend who has a PhD from, and does research in, a top university. The research is in exactly the sort of technologies used in brain-scanning. I asked him about Kurzweil's trends on the accelerating advance of human-brain-scanning technologies. He did not agree with Kurzweil's conclusions, and explained why.
Likewise, I'm looking for input from a open-minded, intelligent, computer/cognitive scientist (who does not strongly support AGI research) on the above question. I don't know where to find them, so perhaps someone on this list could role-play one.
What would s/he say if I asked "Why do you not pursue or support AGI research? Even if you believe that implementation is a long way off, surely academia can study, and has studied for thousands of years, impractical but interesting pie-in-the-sky topics, including human cognition? And AGI, if nothing else, models (however partially and imperfectly with our contemporary technology) essential aspects of some philosophically very important problems."Thanks,
Joshua
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]
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]
