J. Storrs Hall wrote:
On Monday 04 December 2006 07:55, Brian Atkins wrote:
Also, what is really the difference between an Einstein/Feynman brain, and
someone with an 80 IQ?
I think there's very likely a significant structural difference and the IQ80
one is *not* learning universal in my sense.
So there is some group of humans you would say don't pass your learning
universal test. Now, of the group that does pass, how big is that group roughly?
The majority of humans? (IQ 100 and above) Whatever the size of that group, do
you claim that any of these learning universalists would be capable of coming up
with Einstein-class (or take your pick) ideas if they had been in his shoes
during his lifetime? In other words, if they had access to his experiences,
education, etc.
I would say no. I'm not saying that Einstein is the sole human who could have
come up with his ideas, but I'm also saying that it's unlikely that someone with
an IQ of 110 would be able to do so even if given every help. I would say there
are yet more differences in human minds beyond your learning universal idea
which separate us, which make the difference for example between a 110 IQ and 140.
For instance, let's say I want to design a new microprocessor. As part of
that
process I may need to design a multitude of different circuits, test them,
and
then integrate them. To humans, this is not a task that can run on
autopilot....
What if though I find doing this job kinda boring after a while and wish I
could
split off a largish chunk of my cognitive resources to chew away on it at a
somewhat slower speed unconsciously in the background and get the work done
while the conscious part of my mind surfs the web? Humans can't do this, but
an
AGI likely could.
At any given level, a mind will have some tasks that require all its attention
and resources. If the task is simple enough that it can be done with a
fraction of the resources, (e.g. driving) we learn to turn it into a habit /
skill and to it more or less subconsciously. An AI might do that faster but
we're assuming it could lots of things faster. On the other hand, it would
still have to pay attention to tasks that require all its resources.
This isn't completely addressing my particular scenario, where let's say we have
a roughly human level AGI, it has to work on a semi-repetitive design task, the
kind of thing a human is forced to stare a monitor at yet doesn't take their
full absolute maximum brainpower. The AGI should theoretically be able to divide
its resources in such a way that the design task can be done unconsciously in
the background, while it can use what resources remain to do other stuff at the
same time.
The point being although this task takes only part of the human's max abilities,
by their nature they can't split it off, automate it, or otherwise escape
letting some brain cycles go to "waste". The human mind is too monolithic in
such cases which go beyond simple habits, yet are below max output.
Again, aiming off the bullseye. Attempting to explain to someone about the
particular clouds you saw yesterday, the particular colors of the sunrise,
etc.
you can of course not transfer the full information to them. A second
example
would be with skills, which could easily be shared among AGIs but cannot be
shared between humans.
Actually the ability to copy skills is the key item, imho, that separates
humans from the previous smart animals. It made us a memetic substrate. In
terms of the animal kingdom, we do it very, very well. I'm sure that AIs
will be able to as well, but probably it's not quite as simple as simply
copying a subroutine library from one computer to another.
The reason is learning. If you keep the simple-copy semantics, no learning
happens when skills are transferred. In humans, a learning step is forced,
contributing to the memetic evolution of the skill.
IMO, AGIs plausibly could actually transfer full, complete skills including
whatever learning is part of it. It's all computer bits sitting somewhere, and
they should be transferable and then integrable on the other end.
If so, this is far more powerful, new, and distinct than a newbie tennis player
watching a pro, and trying to learn how to serve that well over a period of
years, or a math student trying to learn calculus. Even aside from the dramatic
time scale difference, humans can never transfer their skills fully exactly in a
lossless-esque fashion.
Currently all I see is a very large and rapidly growing very insecure
network of
rapidly improving computers out there ripe for the picking by the first
smart
enough AGI.
A major architectural feature of both the brain and existing supercomputers is
that the majority of the structure/cost is in the communications fabric, not
the processors themselves. A botnet using residential internet connections
would be immensely hobbled. (It would be different, of course, if it took
over someone's Blue Gene...)
I haven't examined the figures closely lately, but my guess would be no matter
which particular bottleneck you want to focus on, that bottleneck is already
large enough to allow for interesting things, and of course these things are all
improving quickly. I think the aggregate backbone capacity of the internet has
been doubling faster than Moore's Law's rate, but I don't have the figures handy
to back that up.
What comes after that I think is unprovable currently. We do know of
course that relatively small tweaks in brain design led from apes to us, a
rather large difference in capabilities that did not apparently require too
many
more atoms in the skull. Your handwaving regarding universality aside, this
also
worries me.
There we will just have to disgree. I see one quantum jump in intelligence
from Neanderthals to us. Existing AI programs are on the Neanderthal side of
it. We AGIers are hoping to figure out what the Good Trick is and copy it,
making our computers as innovative and adaptable as we are, relative to
whatever processing power they may have.
I don't see any reason to assume there is a whole nother Good Trick waiting
for them after that (and it doesn't seem likely they'll need it!)
Minsky claims you could do an AI on a 486. I think he's wrong: the thing that
makes us have to have brains with the computing power of a $30M supercomputer
is learning, which is computationally expensive. So I don't think an AI will
be able to improve itself all that much faster than a human can until it has
substantially more processing power than we do. That's at least a few decades
out.
Mmm hmm. Yes, well we do currently have a disagreement here. I on the other hand
find it very unlikely that humans who have just barely peeked over the hill of
real smarts will turn out to have anything like the best learning, creating,
etc. setup. I don't see a lot of convincing evidence that your viewpoint is
likely correct, but perhaps this will be in your book? In the meantime because
of the potential existential risk I must err more towards caution along the
kinds of arguments Bostrom presents here:
http://www.nickbostrom.com/astronomical/waste.html
Much of your argumentation seems to rely on groups of AGIs forming,
interacting
with society for the long term, etc., but it seems to completely dismiss the
idea of an initial singleton grabbing significant computing resources and
then
going further. The problem I have with this is this "story" you want to put
across relies on multiple things that would have to go just right in order
for
the overall story to turn out just so. This is highly unlikely of course.
It would be if it required everybody working on AI to conform to some specific
plan or set of constraints, because no such thing will happen. The proper
course for any of us is to see what we can do *in spite* of the fact that
there will be a lot of (corporate, military) AIs out there that are NOT built
according to our recipe. The obvious thing to start with is the social and
moral constructs that humans have developed under essentially the same
constraints. Morality works: that's why it could evolve.
While I sort of agree with the idea of trying to get out in front of others who
may not be playing nice with their AGI designs (AGI arms race anyone?), I don't
see how this gets back to answering the original discussion point you snipped:
why "SuperAI takes over" isn't possible at all. Yes, you've got your particular
ideas on how to keep your one single AGI nice and friendly. But as you point out
above, there could be one that arrives before yours, or in your "AGI society"
story there will be a lot of competing AGIs. Why again can't one of them
silently root a bunch of boxes and rapidly outgrow the others?
--
Brian Atkins
Singularity Institute for Artificial Intelligence
http://www.singinst.org/
-----
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/?list_id=303
