[extropy-chat] Re: Any progress towards AI at all?

[J. Andrew Rogers]

On 10/10/05 2:11 AM, "Eugen Leitl" <eugen at leitl.org> wrote:
> On Sun, Oct 09, 2005 at 10:11:59PM -0700, J. Andrew Rogers wrote:
>> To an outsider, little has changed in the last decade in AI.  To an insider,
>> theoretical progress in the last five years or so has been hyperactive and
>> qualitative.  This sea change just is not something that is easily seen or
> 
> How do we measure that progress, however? Do we have benchmarks on practical
> systems? I don't see much progress in practice, but I'm admittedly not looking
> very hard (it would be a full-time job).


For almost the entire history of AI research, there has been no coherent
general theoretical model for intelligent systems.  In the last few years,
something that looks very much like a comprehensive theoretical description
and framework for intelligent systems in the abstract has been coalescing
very nicely.  The net effect is that it is starting to transition from
looking like a really hard theoretical problem to a really hard engineering
problem.  And the engineering problems that follow do not look like
classical AI, so we know that we have something interesting and new rather
than rehashing an old concept with a fresh coat of paint.  To the man on the
street those two states look the same, but for someone in the trenches that
looks like real progress for a change.

That is the basis on which I am saying AI has been making real progress.
Solving a problem that one cannot even describe in the theoretical abstract
is nigh impossible, but engineering problems are something humans are
reasonably good at brute-forcing in many cases. There are still legions of
academics and clueless researchers following their One True Religion and
living off grant mills that prefer minor variations of very old (and dead)
ideas, but these people have never defined progress in the field and a lot
of really good work is being done by multiple research groups on various
aspects of core theory.

The main thing to take away from this is that AI research is not nearly as
stalled as a cursory examination of its history might suggest.  For the
first time in its history, it is starting to look interesting at something
other than an academic level.


J. Andrew Rogers