Re: The engine of intelligence.

From: Heartland (velvethum@hotmail.com)
Date: Wed Aug 30 2006 - 16:24:23 MDT


Hi Michael,

>> Humans, like a car as an example of *a* thing that transports people, are just
>> one
>> of many possible implementation of intelligence. If someone wanted to build a
>> machine whose purpose would be to have intelligence, what would its "engine",
>> "the
>> mechanism for transferring motion", and the "structure that houses all the parts
>> of
>> machine" be?

Michael:
> Mainly, sequence prediction:
>
> http://scholar.google.com/scholar?q=%22sequence+prediction%22&ie=UTF-8&oe=UTF-8&hl=en&btnG=Search
>
> Coupled with a utility function, perceptual apparatus, and a set of
> heuristics for preference ordering.

And, as Tennessee Leeuwenburg points out, memory is also a necessary requirement.
While I may agree that these things are associated with intelligence, unfortunately
they don't really explain how intelligence works.

In the internal combustion engine, when the compressed gases inside combustion
chambers are ignited, the expanding gas pushes the pistons down while the
crankshaft transfers that motion to the rest of the system which transfers the
motion further to the car's wheels. By far, this is the most important mechanism
that causes cars to serve their purpose. Without it, and without motion, a car is
just a weird outdoor furniture.

What would be an analogous explanation for how the "engine" of intelligence works?
Obviously, I'm aware of the likely possibility that intelligence doesn't reduce to
a single equation. The explanation might be far more complicated than the
explanation for how the mechanical engines produce force. The point is, I would
like to see one if one is available.

It's often the case that when people explain intelligence they tend to explain it
in terms of what it does, not in term of what it is. Unfortunately, intelligence
cannot be understood and built based only on what it does. You have to know exactly
what it is and how it works. So, I would classify sequence prediction as something
that intelligence does, not what it is. What I'm interested in is an explanation of
how intelligence achieves things like sequence prediction.

A human-level intelligence is able to read this message and understand it. We, as
intelligent beings, instantly know what that means, but what does "understand" mean
exactly? How does it work? What is the step-by-step process that intelligence uses
from the moment it perceives characters on the screen to the point when it
understands the meaning of the text?

Michael:
> If you want the intelligence to be capable of self-modification, you
> also need a reflective decision theory.

Yes, of course. A need for a reflective decision theory is a problem in the field
of FAI, not AI, just like car safety is not a problem in the field of engine
design, correct?

S.



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