Re: answers I'd like, part 2

From: Adam Safron (asafron@gmail.com)
Date: Fri Nov 16 2007 - 06:31:14 MST


Whoops. I think I was using some terminology in an idiosyncratic
fashion. If I continue to do so, please point it out to me. You are
absolutely right that we would have no chance of making our emulated
brain into a genius–or even a sane person–if its lifeworld constituted
a box. By brain-in-a-box, I meant a simulated brain that was
restricted in its allowed outputs. Instead of a box, it would have a
simulated embodiment in a simulated world. But this would all be
restricted to "the box"–not necessarily a box, but some sort of semi-
closed system–of a given set of computational elements.

If we find ourselves with sufficient computational resources that we
could build an emulated brain that functions on a par with human-brain
time-scales, then it would necessarily surpass all humans in terms of
speed of cognition. Our neurons are fixed in terms of how fast they
can sum/fire action potentials. The limits of digital emulation have
no clear boundary, though potentially sufficiently difficult to
advance that this method is rendered irrelevant by other methods. If
you could make the brain as smart as the smartest human, and then
speed up the rate of computation, you would have surpassed the
greatest human intelligence. Over-clocked digital Einstein (to pick a
well-known genius) could think more thoughts in a given period of time
than meat-brain Einstein. Mistakes/failures are often the result of
failing to give proper thought to a problem. By giving digital
Einstein more time to work on any given problem, you would get better
results than meat-brain Einstein. To someone interacting with a sped-
up digital Einstein using a flesher-brain, he would seem impossibly
intelligent. Since the information constituting the intelligence
would be digitized, you could then make multiple copies of this
intelligence–not sure of the ethics of this–and have them work on
problems in parallel. Though this may not get us to the "knee-of-the-
curve" in terms of progress towards super-intelligence, it would get
us much closer than we are now. Ensuring "friendliness" still seems
problematic. But once again, this method could be rendered obsolete
by the time these technologies become feasible.

In terms of ethics, emulated neurons seem no different than
experimenting with cognitive models based on completely different
principles (e.g. some sort of hierarchical/probabilistic world-
modeler). The one difference I can see is that an emulated brain will
necessarily have qualitative subjective experiences like ours and an
AI based on other computational principles may not. As long as we're
not actively non-benevolent to our digital-genius-brains, then it
might be possible to justify giving them special-case status in terms
of making them part of our moral communities.

-a

On Nov 15, 2007, at 8:48 PM, Bryan Bishop wrote:

> On Thursday 15 November 2007 09:56, Adam Safron wrote:
>> A brain-in-a-box would not know how to improve its own
>> functioning. Emulation is not understanding.
>
> The problem with self-improvement-in-a-box is that it is a _box_ and
> that is no environment in which complexity can grow. A baby needs the
> womb. It is like trying to draw a 128-sided polygon in a pyramid,
> which
> clearly does not have enough points or 'complexity' to support the
> struct.
>
> But in general I question whether or not the actual functioning of the
> brain has to be improved at the moment to reach seed-AI levels. What
> we
> need is more brain, not necessarily better brain, although both
> avenues
> of research should be pursued. Instead of emulating we could be adding
> more raw brain power and trying to interface it with the readily
> present neurons.
>
> I say this because we _are_ AI ourselves and so far not yet seed AI.
> So
> the way to solve this problem seems to be to add more brain, which
> seems to be the special component, not necessarily better brain,
> something that is as of yet impossible. You can cite geniuses, yes,
> but
> you cannot yet cite ways to replicate genius, so scientifically I am
> not sure if genius counts as an argument.
>
> Interesting post. Thank you. Please continue. :)
>
> - Bryan
>



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:01:00 MDT