Re: AI architectures in the 2000s

From: Ben Goertzel (ben@goertzel.org)
Date: Wed Aug 09 2006 - 08:12:00 MDT


>It seems people are not putting as much emphasis on finding
> the One True Unifying Principle to create AI.
>
> What do you think?

And, as was probably implicit in my prior reply, I do not think any
such One True Unifying Principle exists; and I think that the search
for such a Principle is misleading, has slowed down AI in the past,
and will probably continue to slow it down in the future...

A mind consists of a general-purpose pattern recognition engine
coupled with a host of appropriately-chosen special-purpose
pattern-recognition engines. The general-purpose pattern recognition
engine must be given the capability to introspect (recognize patterns
in itself) and also to modify itself with some reasonable flexibility.
 A knowledge representation and operational framework must be created
allowing these different pattern-recognition engines to productively
interact and contribute to the same evolving knowledge store. There
must also be a pragmatic "action execution framework" so that patterns
involving which actions to carry out in which situations may be not
only learned but executed. And, the operational framework must be
linked in with a specialized pattern-recognition engine whose goal is
to intelligently allocate resources among the various
pattern-recognition engines...

In all this, there is no central magic trick. There are plenty of
clever tricks needed to make the various components work, along with
highly competent integrative design and engineering.

While I have expressed the above in my own language, I suppose that a
similar intuition underlies the work of others who are working on
"integrative" approaches to AGI.

Phil, do you disagree with this overall intuition regarding the nature
of intelligence? If so, could you elaborate? What role do you think
the "special magic trick" at the core of intelligence, if any, plays?
Do you think it underlies both generalized and specialized pattern
recognition ability, or do you think it's a special trick for
generalized pattern recognition? Or would you reject this language
for discussing things altogether?

-- Ben G



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:57 MDT