[wplug] OT: AI

[Bill Moran]

Patrick Weber wrote:
> You may find that the topic of Genetic Algorithms may interest you.  They
> have long been a topic of AI research and have some fun aspects to them.
> Basically what GA's do is take a series of parameters and "mutate" them to
> use the biological term.  Then it "breeds" these parameters to produce
> outcomes.  The outcomes that most closely resemble what you want are then
> selected, mutated and bred again.  What this does is "evolve" your
> parameters for your data in a similar way that monkeys evolve into people.

<humor>
If humans evolved from monkeys, why are there still monkeys?
</humor>

It's an interesting concept ... you could "breed" a result the same way
you would breed dogs (cull the undesirables, and breed the successful)
until you ended up with what you wanted.

I guess you could also breed without the culling (like humans do ;) and
just see what happens.

> For your situation you could use your set of parameters and run it through a
> genetic algorithm until it meets the regressions that you currently have.
> Then you can use what evolved to make predictions... This is just a thought.
> 
> AI is a broad topic of computer science research that has yeilded some
> interesting algorithms.  AI is not a thing unto itself however.  My personal
> definition: "Non-deterministic code that behaves intelligently within the
> domain for which is was intended"

That's better than mine, except for two things:
1) Does the code _have_ to be non-determinstic?  Can deterministic code work?
    A chess program that looks 10 moves ahead to determine the best move could
    never be AI?  (If so, I don't think any chess programs that qualify as AI)
2) What do you mean by intelligently?  Could a better term be used there, such
    as saying "... code that makes appropriate choices within the domain ..."?

Of course, to the layman, AI means "a computer that acts like a human" ... which
of course implies that humans are intelligent.  It's intersting that you could
dredge up a number of chat-bots that are perfectly capable of keeping a real
human engrossed in conversation for a considerable amount of time, thus arguing
that they act like human chatters and are thus AI software.  The only problem is
I quite often meet human chatters who show no signs of intelligence ...

> -Pat Weber
> 
> ----- Original Message ----- 
> From: "Chris" <vze2f6h6 at verizon.net>
> To: "'General user list'" <wplug at wplug.org>
> Sent: Thursday, April 29, 2004 9:04 AM
> Subject: RE: [wplug] OT: AI
> 
> 
> 
>>-----BEGIN PGP SIGNED MESSAGE-----
>>Hash: SHA1
>>
>>
>>>-----Original Message-----
>>>Behalf Of Bill Moran
>>>
>>>Chris wrote:
>>>
>>>>I have been trying to find out more about AI the past week
>>>
>>>but haven't
>>>
>>>>been really able to find what I am looking for.  I am hoping that
>>>>someone on this list might have at least some experience
>>>
>>>with it.  I
>>>
>>>>think that the problem is that I am not sure that AI is what I am
>>>>looking for.  I want an app that will be able to look at a couple
>>>>different regression lines, look for patterns, and make predictions
>>>>based on those patterns.  I haven't been able to decide if that's a
>>>>good fit for AI or just some good old probability and stats.  Does
>>>>anyone have some insight for me?  Feel free to contact me off-list
>>>>because this is OT.  Chris.romano [at] verizon [dot] net
>>>
>>>IMHO, the biggest problem with AI is that the term "AI"
>>>doesn't really mean anything.
>>>
>>>So, the real question (if you ask me) is more involved with
>>>what you really need the code to do, and whether or not it's
>>>been done before.  If there are existing algorithms to do the
>>>interpolation you need, use them.  If not, you'll likely have
>>>to develope it yourself, and if it's impressive enough,
>>>someone will probably call it AI.
>>>
>>>Just my extremely biased opinion.
>>>
>>
>>Well, I don't know if there are existing algorithms or not.  I want to do
>>this as a little project for myself.  I am sure that in the beginning that
>>it will be a simple algorithm (if I am developing it), but I guess it will
>>need to some what learn and change based on new patterns and the addition
>>of other factors that will influence it.  In my previous example that
>>would be things like the money supply and interest rates.  I guess that I
>>have a lot of reading a head of me.

-- 
Bill Moran
Potential Technologies
http://www.potentialtech.com