From: justin corwin (firstname.lastname@example.org)
Date: Wed Jul 13 2005 - 14:59:36 MDT
On 7/13/05, Chris Capel <email@example.com> wrote:
> Sorry to fork the thread, but this really interests me. This seems
> unlikely, but what if the limited intelligence of humans is due not
> mainly to the firing speed limitations of neurons, but due to some
> architectural limitation in the human brain that would persist even in
> a super-fast thinking AI? Just because an AI has thousands of
> subjective years to think in every minute of our time doesn't mean...
It is an interesting subject, and well worth discussing, quite aside
from my line of reasoning.
It's important here to define the playing field and terms. For
example, note that there is currently only theory, and weak theory
predicting what AI will actually look like. Anyone who says anything
is either expressing opinions based on prescientific AI theory (non
functional, non experimental incomplete stuff) or projective opinions
based on cognitive science (or more likely, personally extended
cognitive science). So keep where people are coming from in mind.
It's common knowledge that AI is different from human intelligence.
It's not a simple comparison. But there are things we know will be
different. AI's will be parallel computing, but by human standards,
they are serial speed demons. Most people interpret this as AIs
running at great subjective speed compared to humans, but that's not
quite true. It's valid to say that under most AI designs I've ever
heard of, the entity would be quite fast at evaluating abstract
concepts, orders of magnitude above humans. But this doesn't hold true
for all things. How long would it take the AI to recognize something
it has seen before?(dependent search through memory) What about how
long will it take to set actuator outputs?(serially considered
values?) What does it's 'native timescale' actually look like, and how
much does translation to physical timescales cost?(learned waits?
preset actions? what about processes that happen faster than it
My answers to these kinds of questions will be different than another
AI theorist, and depend not only on proposed AI architecture, but on
interactions within architectures we don't actually know about, like
what the 'intelligence' of a system with human-like architecture but
perfect memory looks like, or what a completely alien statistical
intelligence looks like with a poor perceptual model, versus a deep
modality? <big shrug> I'm not confident in my understanding of these
concepts within my own proposed design, much less the space of
human-possible designs. That's why I say more work is necessary, or
somebody needs to share.
> I suppose this is much to speculative to be useful (except in the
> context of justin's original post), unless someone has any data that
> would actually support these ideas. If we do eventually come across
> something like this limiting progress, we'll deal with it then, when
> we've understood enough to even duplicate human intelligence, and
> certainly not before then. In the meantime, even a pessimistic
> estimate of the intelligence of an uploaded human (10-50 times their
> intelligence before they were uploaded) might make them quite
> dangerous to the world if malicious. So I'll stop wasting everyone's
> time now.
It's true that it's speculative, but it's important to establish some
kind of possibility space so that people can make informed judgements
about relevant factors and probability.
Uploaded humans are of course another question entirely. I tend to be
of the opinion that they are dangerous, but in predictable ways, and
not immediately existentially terminal, although pretty suboptimal.
Assuming they don't kill themselves or change themselves into
something broken or nonhuman very quickly(which is also a big
AIs can be dangerous because they're different, and uploads are
dangerous because they are the same. pithy, no?
-- Justin Corwin firstname.lastname@example.org http://outlawpoet.blogspot.com http://www.adaptiveai.com
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:51 MDT