Mike,
You are talking about two different occurrences of a computational
explosion here, so we need to distinguish them.
One is a computational explosion that occurs at design time: this is
when a researcher gets an algorithm to do something on a "toy" problem,
but then they figure out how the algorithm scales when it is scaled up
to a full size problem and discover that it will just need too much
computing power. This explosion doesn't happen in the AGI, it happens
in the calculations done by the AGI designer.
The second type of explosion might occur in an actual working system
(although strictly speaking this would not be called a "computational
explosion" so much as a "screw up"). If some AGI designer inserts an
algorithm that, say, requires the system to engage in an (almost)
infinitely long calculation to make a decision at some point, and if the
programmer allows the system to start this calculation and then wait for
it to end, then the system will hang.
AI and Cog Sci have not been "obsessed" with computational explosions:
it is just a fact that any model that suffers from one is dumb, and
there are many that do.
They have no connection to "rational" algorithms. Can happen in any
kind of systems. (Happens in Microsoft Windows all the time, and if
that's rational I'll eat the entire town of Redmond, WA.)
it is certainly true that some style of computation are more prone to
hanging that others. But really it is pretty straightforward matter to
write algorithms in such a way that this is not a problem: it may slow
some algorithms down a bit, but that is not a fundamental issue.
For what it's worth, my system does indeed stay well away from
situations in which it might get locked up. It is always happy to stop
what it's doing and go for a drink.
But remember, all this is about "hanging" or "livelock", not about the
design problem.
Richard Loosemore
Mike Tintner wrote:
Thanks. But one way and another, although there are different
variations, cog sci and AI have been obsessed with computational
explosions? Ultimately, it seems to me, these are all the problems of
algorithms - of a rigid, rational approach and system - which inevitably
get stuck in dealing with real world situations, that don't fit or are
too computationally demanding for their models. (And can you *guarantee*
that your particular "complex" approach isn't going to run into its own
explosions?)
These explosions never occur, surely, in the human brain. For at least
two reasons.
Crucially, the brain has a self which can stop any computation or train
of thought and say: bugger this - what's the point? - I'm off for a
drink. An essential function. In all seriousness.
Secondly, the brain doesn't follow closed algorithms, anyway, as we were
discussing. And it doesn't have a single but rather always has
conflicting models. (I can't remember whether it was John or s.o. else
recently who said "I've learned that I can live with conflicting
models/worldviews").
Richard: Mike Tintner wrote:
Essentially, Richard & others are replaying the same old problems of
computational explosions - see "computational complexity" in this
history of cog. sci. review - no?
No: this is a misunderstanding of "complexity" unfortunately (cf the
footnote on p1 of my AGIRI paper): computational complexity refers to
how computations scale up, which is not at all the same as the
"complexity" issue, which is about whether or not a particular system
can be explained.
To see the difference, imagine an algorithm that was good enough to be
intelligent, but scaling it up to the size necessary for human-level
intelligence would require a computer the size of a galaxy. Nothing
wrong with the algorithm, and maybe with a quantum computer it would
actually work. This algorithm would be suffering from a computational
complexity problem.
By contrast, there might be proposed algorithms for iimplementing a
human-level intelligence which will never work, no matter how much they
are scaled up (indeed, they may actually deteriorate as they are scaled
up). If this was happening because the designers were not appreciating
that they needed to make subtle and completely non-obvious changes in
the algorithm, to get its high-level behavior to be what they wanted it
to be, and if this were because intelligence requires
complexity-generating processes inside the system, then this would be a
complex systems problem.
Two completely different issues.
Richard Loosemore
-----
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?member_id=8660244&id_secret=74586829-bb45d1
