I have a PhD in AI, but I got it in 1994 and promptly left academia, so my knowledge is historical rather than contemporary. I'd like to hear what's going on in AI if anyone's up on it, and I'm happy to share my (dated) knowledge on AI.

I'm not up on the recent stuff either, but one thing that struck me as interesting was that Computer Go has got significantly better recently:

http://teytaud.over-blog.com/article-35709049.html

(anybody remember the "Ing" prize for Computer Go?)

It also looks like the same algorithm may have the potential to improve AI in other more complex board games and even computer strategy games:

http://sander.landofsand.com/publica...or_Game_AI.pdf

Juk

I got a degree in it in 2007 but that's also probably out of date now to some extent. It was mainly focussed on machine learning, and the thing I was taught with most emphasis is how only recently (note within the last 5-10 years) computers are fast enough to actually put a lot of the theoretical algorithms into practise.

Genetic algorithms were described to me as incredibly important in insurance and engineering, and these interested me greatly, and have continued to be interested in these. It's the one thing I took away from my course that I truly valued.

I'd love to hear about what you studied on your PHD, it's such a fast moving area of computing, I've probably forgotten a lot of what I learnt but it is fascinating. Odd sods I've read about is Scrabble is can beat every human now apparently (no source, but I used to be very much into scrabble).

I was working on second-gen knowledge-based systems, and I was particularly interested in modelling uncertainty. I didn't feel any of the main methods around then (Bayesian, Certainty Factors, fuzzy logic etc) really replicated human reasoning of uncertainty, and came up with a more qualitative approach, that used words to express uncertainty rather some numeric method. (PS. Genetic algorithms were around in the early 90s by the way )

The idea was you could state facts with a level of uncertainty, and they could be combined with other facts in a chain of reasoning, which would lead to a lower level of certainty the longer the chain got, and depending on what qualitative certainties were in the chain. The idea was a chain is only as strong as its weakest link.


A means you're fairly certain of B, B means you're not very certain of C, so A means you're not very certain of C is a very simple example.

In addition, there could be a number of different chains of reasoning leading to a certain point, and the more chains leading to a point which have strong certainties could combine to make you more certain of the end point. Another example:

X makes you reasonably certain of A
Y makes you reasonably certain of A
Z makes you reasonable certain of A

X+Y+Z makes you feel A is strongly certain.



I've explained it in simple terms, and hope I get it across.

I called the reduction in certainty as your individual chain of reasoning got longer 'knowledge senescence', and the increase in certainty as your individual chains of reasoning combined was 'knowledge renascence'.


There were other benefits found in looking at the machines output. It was able to match a heuristics-based first gen KBS in terms of correctness compared to a human expert (both were right about 80% of the time compared to the human experts opinions, but they varied in which cases they found right by some way!) but also in terms of self-explanation - the machine could explain in plain English how it reached certain conclusions by laying out the individual facts in order, and I found this proved educationally comparable to a hand-drafted (by a human expert) explanation of reasoning getting to a certain point.


(I got very peeved with the slow speed of academia on completion, and moved into industry, btw)

That sounds pretty interesting! Do you have your thesis to download by any chance?

What do you actually think AI is? What do you think intelligence is? I always thought intelligence was the ability to make genuine mistakes (as in, computers will never get 2+2 wrong, but humans can therefore computers aren't intelligent), but I'm a mere peasant when it comes to philosophy.

No it was original typed up in WordStar. I don't have a copy of it electronically.


I'll get back to you on the other questions.

anyone work with JGAP at all to do genetic programs or genetic algorithms?

didn't want to start a new thread when i found this in the search feature.

This popped up on HN today:

http://aima.cs.berkeley.edu/books.html

Wow, that list on uncertainty makes it look like the field of uncertainty in AI is really stagnant. My work was way ahead of 90%+ of that list, and I finished it 15 years ago.

Maybe I should have stayed in it, I'd be like the world expert by now.

I would love to hear/discuss what AI actually is! Interested in your opinion