From: Jochen.Fromm@t-online.de
>
> You are aware, I hope, that their machines aren't idle? Google is
currently
> using all those boxes to run its search service, and bring in $1B/yr in
> revenue. That isn't a resource available for a new purpose.
>
> While more computer power is always better, computation resources isn't
the
> critical bottleneck for AI. In many cases, there is simply a lack of
> sufficient theory to understand what is going on in intelligence.
>
Of course the Google cluster is not idle. The question is if
huge increase in computational power would facilitate further
development in AI ? You say we need also more sophisticated
theories. That is true.
Yet more computational power, for example in form of large clusters,
allow you to perform experiments which were not possible before.
These experiments in turn could guide the development of new theories.
Nils J. Nilsson argued in "Eye on the Prize" (*), AI Magazine Vol. 16
No. 2 (1995) 9-17:
"Is general intelligence dependent on just a few weak methods
(some still to be discovered) plus lots and lots of commonsense
knowledge? Does it depend on perhaps hundreds or thousands
of specialized minicompetences in a heterarchical society of
mind? No one knows the answers to questions such as these,
and only experiments and trials will provide these answers."
What kind of experiments ? It's simple: create intelligent agents
in a complex, life-like virtual environment, a normal 3D or
multidimensional environment we know from daily life.
If it is so easy, where is the problem ?
a) At the universities, most scientists work with Unix, Linux,
Java and C++. You can not program a fast virtual 3D environment
with JAVA, and the development of large programs with C++ without
a visual studio or an integrated development environment is very
complicated and difficult, and takes a lot of time.
Games with complex 3D worlds are usually written in Assembler,
C and C++. But every software game company has it's
own 3D graphics or game engine. There is no standard.
b) to simulate a comlex environment is inherently complicated,
as the attribute "complex" says. An agent which is able
to understand such a world has a similar complexity.
c) every AI research group tries to find the last AI-mystery alone.
Instead of working together in a large scale project, which would
be comparable to the Apollo program or the LHC Collider project
at CERN, every research group wants to tackle the big problems
alone.
Consider Denmark. It is a small country which would never be
able to build a rocket. Yet as member of the ESA, it is able
to take part in the European space program. 2003 the ESA budget
was € 2700 million. The ESA (European Space Agency)
would not work if each country would try to build it's own
rocket. If every country (Austria, Belgium, Denmark, Finland, France,
Germany, Ireland, Italy, the Netherlands, Norway, Portugal, Spain,
Sweden, Switzerland and the United Kingdom) would want to
build the control panel of the rocket, and no one would want to
build the _suitable_ engine, if there would be no common
strategy or space policy.
The AI community needs a common strategy or policy, or a
large scale project which really forces the scientists to work
together.
(*) see
http://www.aaai.org/Library/Magazine/Vol16/16-02/vol16-02.html
[ comp.ai is moderated. To submit, just post and be patient, or if ]
[ that fails mail your article to , and ]
[ ask your news administrator to fix the problems with your system. ]
--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)
|
