Re: Is generalisation a limit to intelligence?

From: Joaquim Almgren Gāndara (claw@lords.com)
Date: Sat Dec 02 2000 - 12:05:56 MST


> > To sum it up: Is there some way to combine the fuzzy quality that
> > intelligence relies on with the rigid quality of not making a single
> > mistake? Is generalisation a limit to intelligence?
>
> I think this is the same "memory limits intelligence" idea we've been
> tossing around, recently?

Hmm, is it? I don't know, doesn't sound like it (at least not to me). Let me try
to explain what I mean. I'm saying that nature came up with generalisation as a
means of storing data virtually, or implicitly. This in turn developed into
intelligence. I think that intelligence builds upon generalisation, and that if
you store everything (infinite memory), you'll run the risk of getting too
rigid. How do you force an AI to make generalisations when it's so much easier
to "look it up"? Are you saying that the AI should just look up the exceptions
to its generalisations whenever needed -- or rather, check if there's an
exception whenever it uses a generalisation?

> ...it could also use the raw data to continually re-evaluate its
> generalization to be sure it still worked.

But what about data that can't be properly classified? What about inconclusive
data or errors?

> Lacking infinite memory, some error is necessary due to overgeneralization
> (overfitting).

Sorry, I obviously didn't explain overfitting properly. With "overfitting", I
meant the act of taking the data too literally and not generalising at all. In
the case of too small training sets, you'll get something like "No, that can't
be an 'apple'. It doesn't look exactly like this 'apple'. This is an 'apple'."
Extremely high error values. However, with infinite memory, you can show the AI
every apple in the world, and the error will be zero. Until you take a bite out
of one of the apples, whereupon the AI will wonder where the 'apple' went.
    Basically, I'm saying that *with* generalisation, you have intelligence and
hence error. *Without* generalisation, you have no intelligence. How do you
combine the two? Can you really solve this with infinite memory, or will the AI
just turn into a walking dictionary (if indeed it walks)?

- Joaquim Gāndara
. claw@lords.com
. http://www.ite.mh.se/~joaal98
. http://games.scandit.com
. http://www.mp3.com/sdtank



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:35 MDT