Re: AI and statistics

Abe Kohen wrote:
> Actually I had no __prior__ knowledge that the NYSE doubled the Utility
> Index, as I have never traded the Utility Index. What I had was prior
> knowledge about doubling and halving of Indices, and of stock splits and
> reverse stock splits. So my "pattern matcher" brain deduced that the NYSE
> must have done something similar. Then I googled to verify my "hunch."

If you were interested I could give an example of what could be done to
extract this type of knowledge from you. I don't know your background
so you may be full trained in knowledge acquisition/engineering
already.

> Perhaps what I meant to say is that I do not know how to program an
> "adequate" amount of domain knowledge into an expert or rule-based system,
> but I do know how to program some very specific and very useful rule-based
> trading systems, which do not rely on external domain knowledge.

I guess that indirectly you're saying that there are certain problems
where the return/effort ratio is good, while others have a much poorer
ratio.

> >
> > > Be that as it may, it does not preclude building specific rule-based
> > > applications, much as is currently the situation in finance vis-a-vis
> > > "algorithmic" trading.
> >
> > Provided of course there is sufficient financial and other will to
> > build these systems, as they are not easy nor cheap to build.
>
> Nope, they are not easy for __most__ people, and if they were cheap I would
> be making ONLY two cents. ;-)

Hmm.... But there are problems that are incredibly difficult for
anybody. If your work is in finance, then you can pick and choose
problem based on return/effort. But in theory AI as a discipline should
be addressing or at least planning to address all types of problems
that humans can solve.

Cheers,

Ross-c

.