Re: New AI Prize

Fred Flintstone <[EMAIL PROTECTED]> wrote in message
news:[EMAIL PROTECTED]
>
> An AI can only be a complex as its environment, an artificial world will
be
> useful for getting an AI up and running but then it would need to be
brought
> into the real world (robot) to really learn and do anything useful. After
> all, you could put a human child on a far off planet by themselves and
they
> would be lucky to grow up to be any better than a monkey.

That's true, but an AI inhabiting an artificial world doesn't have to be
on its own. There can be as many AI's as you like, of as many different
types as you like, all interacting via a shared artificial world, so that
the
whole is implemented in software which could be sold on DVD's for
entertainment purposes. Like The Sims, or Creatures, or any number of
existing war strategy games. Except I think the artificial universe would
have to be several orders of magnitude larger and more complex than
what is found in any existing game, if it is to support the development of
genuinely intelligent entities.

Note that a world doesn't have to be specified in detail in order
to be large and detailed and complex. Consider e.g. a Mandelbrot
set, in which an intricately detailed pattern is generated using a very
simple rule.Using equally simple  rules, A large and complex 3-D
universe with a realistic landcape and physics, suitable for interactive
gaming, can be specified in remarkably few lines of code.

> We'll always be
> potentially much brighter than monkeys, but what really makes us smart is
> our ability to learn from others and use the tools that others have
created.

True, but do those others necessarily have to be human?  If AI's are
developed which can learn from other AI's,  and which can use tools
which exist only inside an artificial world and which have been created
there by other AI's, why should there be a ceiling on the smartness
that they can attain?

> With the knowledge I have now I could never build a computer myself, I
> wouldn't have a bloody clue but this doesn't prevent me from using a
> computer to try and create something as ambitious as AI.

You can try, but how would you rate your chances of succeeding,
realistically? Computers are tricky but I know that it is possible to
build one because if it wasn't possible then I wouldn't be sat at a
computer keyboard now. But can we be sure that intelligent sotware,
on a par with human intellect,  is even possible?

To make a computer, you could just follow the procedure established
by somebody else who has successfully built a computer, but to build
an AI on a par with human intelligence where would you begin? What
steps would you go through? How would you know whether or not
you were heading in the right direction? If building a computer is too
complex a task for you, how can you realistically expect to put together
sumething which is vastly more complex than any existing artificial
computer?

>
> The reason why creating AI in artificial worlds is such an attractive idea
> for AI programmers is because they just can't afford the time and money or
> don't have the knowledge to build robots themselves.

That's one reason, but we shouldn't have to build robots ourselves
in order to program them, any more than we should have to build a
car ourselves if we want to drive.And even if we had the knack,
and could build lots of programmable robots and buy lots more off
the shelf at Toys'R'Us for a very reasonable price, we'd still have
a big problem programming them to behave intelligently. We would
need some criterion for deciding which of them are acting most
intelligently, e.g. we could place them all on a high table and scrap
or reprogram the ones that walk off the edge, but then we'd just
end up mostly with robots that are so cautious that they don't move
at all. Set them a task which is too easy and they all pass, or make
it too easy and they all fail. So how can you ensure that they are
suitably challenged at all times? By pitting them against one another,
that's how. Evolution by survival of the fittest. It works in nature,
but there it needs billions of years and a similar number of individuals.
The only way to achieve it in a reasonable timespan artificially is
to use an artificial world, in which the main constraints are those
imposed by processor speed and storage capacity.

> Maybe the programmer
> that eventually creates AI could pass it onto the engineers at Honda so
they
> can upload it into their robots. Their robots will be able to really dance
> then.

Yes. But it won't happen the other way round. The Honda
robots won't evolve intelligence which can then be downloaded
into the internet.

>
>
> The human race is smart but a human being is stupid. (Sounds like
something
> a character in a kung fu movie would say)

Whereas I would say the opposite. The human race is stupid but
a human being is (or can be) smart. Think e.g. of all the stupid things
that are done by governments, even though practically all the people
under them can see through the hype. The people  know that what's
happening is stupid, but what can they do about it?