Re: Re: What are useful for a phd?

From: Philip Goetz (philgoetz@gmail.com)
Date: Fri Dec 22 2006 - 13:26:06 MST


On 12/7/06, Anthony Mak <anthony.mak@iname.com> wrote:
> What about logic/symbolic reasoning? I seem to get a mix opinion on this.
> Is logic not useful for FAI? If not, why not? For example,
> wouldn't these have some relevant non-monotonic
> Reasoning, belief revision, ontology, commonsense reasoning,
> deontic logic, operations research, constraint programming...etc?

Why not? Because people have banged their heads against it for
decades without accomplishing much.

Logic is useful for engineering. In your job, it would be good to be
able to use predicate logic. Ontologies provide powerful data in many
ways, but you can get a lot with just the "isa" + "instance-of"
predicates, which is hardly logic. Belief revision is important.
Constraint programming may be helpful. Deontic logic is pretty
specialized, and I wouldn't study it unless I had a particular need
for it.

But I doubt the value of learning logic beyond what you need to do
proofs and to program in Prolog, and then fuzzy logic and maybe
statistical/probabilistic logic.

I don't know how much is useful, but clearly there are a lot of people
who go way too far. Any time you see someone trying to prove that
their logic is complete and consistent, you know they've gone too far.
 Any time someone is worrying over the implicational closure of belief
in their logic, they're going too far. Whatever logic the brain uses
is neither complete nor consistent, nor does anyone work with the
logical closure of their beliefs.

> As for machine learning, besides bayesian and GA, how relevant do any
> of you think the following are? SVM, reinforcement learning,
> gaussian process, Kolmogorov/Solomonoff etc. And, what other ML areas
> can I watch out for?

Reinforcement learning apparenly occurs in the brain, so study that.
SVMs are powerful.

> I haven't heard recommendation for AI planning, so I am a bit surprised.
> Shouldn't a strong AI needs some sorts of planning?

Planning is fun. Sure, you should study planning.

> Eliezer: by Evolutionary psychology is it generally subsumed by a cognitive
> science course?

You mean, an Intro to Cognitive Science course? Probably not.
It should be sufficient if you study psychology, and you study evolution.



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:57 MDT