Re: The Frame Problem

From: Harnad, Stevan (
Date: Sun Mar 09 1997 - 12:00:12 GMT

> From: Michelle Smolarek <>
> What makes the human brain any different from a computer?

No matter how much you believe the controversial hypothesis that
cognition is computation, surely there are far many more differences
between people and computers than there are similarities.

Besides, the hardware (the physical implementation) of the computer is
irrelevant. Only the software (the symbol-manipulation rules = the
programme/algorithms) is relevant. So the question is: Are our minds
just the physical implementations of the right algorithms?

> We still have input, only in a different way to a computer.

Right, but so does a toaster or an automobile...

> All messages have to be stored in the brain and this isn't done by
> images but must be done in the same terms of symbols and rules.

It is only the computationalists who argue that images MUST be
stored as symbols and rules. An image could be stored (and processed)
in analog form too. See the passages about "mental rotation" in Green et
al. (and the other Student Skywriting files on my Web page).

> And then all these memories need to be reached when needed. The same
> process with a computer only we need to provide it with the symbols and
> rules and not the image (only having said that it is now slowly
> possible with scanners)

Yes, both computers and brains must store and retrieve information;
but symbols and rules are not the only way to do it. In mental rotation
tasks it might be more practical to recover the 3-D shape of an
object from the 2-D shadows on our retinas, and then do the analog
of a rotation on this analog shapes. (See the other Skywriting files
for information about mental rotation.)

> I think every thing you learn is subjective to experience, learning
> basic principles and then being able to expand on them, for example the
> idea that if you learn throughout you life that every-thing with one
> leg/stump is a tree, and one day you saw a stork standing on one leg
> would you call it a tree?

We often have to revise our internal representations of things,
including the features of their shadows that we use to recognise,
categorise and name them correctly. But this sort of learning and
revision is done better by neural nets than by symbol systems (with
their Symbol Grounding and Frame Problems).

> We don't remember a million things but that doesn't mean that we
> haven't learnt them, they're stored in our memory only we don't know
> how to recall what we've learnt.

To be able to categorise, a system needs to selectively forget or ignore
information, otherwise it's just a lifetime series of unique images,
as in the (fictitious) case of Borges's "Funes the Memorious" and
in Luria's patient "S" described in "The Mind of a Mnemonist". (See

and other courses' Skywriting to hear more about Funes and S. Go to google and enter "funes harnad" and also "mnemonist harnad" in a separate search.)

This archive was generated by hypermail 2b30 : Tue Feb 13 2001 - 16:23:51 GMT