RE: No AI for Nano/No Nano for copyloads

From: bbrown@transcient.com
Date: Fri Jul 14 2000 - 18:32:16 MDT


Robin Hanson wrote:
> Billy Brown wrote:
> >I agree. However, that process of learning and incemental improvement will
> >almost certainly take several decades if it is done using extensions of
> >current techniques. That makes the resulting brain simulation a 2030 -
2050
> >era technology, vs 2010 - 2030 for basic nanotech manufacturing, sensing
and
> >computing devices.
>
> While it can see how it might take some time to learn how local things
work,
> I think it takes much less time to compile that more abstractly.

Why would you think that?

> And since I tend to think of nanotech as more of a gradual increase in
> manufacturing abilities, I don't think specifics of nanotech progress matter

> that much here.

Nanotech also offers qualitiatively new abilities that can significantly
affect the development of other technologies. Without nanotech you aren't
going to see cheap uploads in this century - the computers will be too slow,
and it will take too long to do the optimization research.

But with nanotech, you can have neural interfaces and limited IE before you
can do an upload, based on partial understanding of particular portions of
the brain. You can build cheap computers that have the power to run uploads,
but those same machines are availible for AI research. The relative rates of
progress of these different technologies all change significantly, which gives
you a very different world when it comes time to predict what will happen when
those first uploads happen.

> I don't think neural interfaces will be very useful,

Neural interfaces would make truly immersive VR and ER environments possible,
which would in turn lead to the appearance of large new markets for virtual
goods and services. The hard distinction between things that are "inside a
computer" and "outside here in the 'real' world" would vanish, leaving society
much more willing to acknowledge that someone who lives inside a computer
might be a real person. And, of course, the fact that an upload has no
physical body would no longer constitute a significant limit on its ability to
interact with normal humans.

> and don't think substantial intelligence enhancement is that relevant -
> whatever enhancements humans have uploads can use too.

Yes, we should expect that uploads will be able to use most enhancements that
would work on humans. But that doesn't mean IE doesn't matter. What it implies
is that, as IE and related technology advances, the difference between an
upload and a 'normal' human dwindles to insignificance. If IE gets going
relatively quickly, and uploading takes longer, it is entirely possible that
uploads per se will never be an important factor.

To put it in concrete terms: Does an upload that does a hundred jobs at the
same time, some virtual and some via teleoperated drones, have any particular
competitive advantage over an enhanced 'human' who can do the same things? I
would say, only if you don't have to pay the upload - and at that point it
will be obvious to everyone that you are practicing slavery.

> Sentient AI seems very unlikely to me to
> happen before uploads, being a vastly harder problem.

On what basis do you assume that it is 'vastly harder'? I have considerable
respect for the difficulty of duplicating the human mind in software, but you
don't have to do that to make significant breakthoughts in AI.

Acheiving animal-level intelligence seems to be a much more tractible problem,
and it would enable us to build robots good enough to automate almost all
physical labor. Building specialized programs capable of superhuman
performance on narrow problems is obviously feasible, and seems likely to
become increasingly common in the near future. But these kinds of advances
would significantly affect the usefullness of upload slavery (since humans
would primarily be usefull for general-purpose decision making, and you can't
trust a slave to do that for you), and would have many of the
wealth-multiplying effects that pie-in-the-sky types usually atribute to
nanotechnology.

---------

Having written all that, I have two general observations to make about your
particular upload scenarios:

1) Slave labor is generally not profitable, because of the size of the
investment required to ensure that the slaves don't revolt (and, more
generally, that they actually obey your orders instead of just doing what they
want to do). I don't see anything about upload slavery that would make it
immune to this problem - unless you want to posit effective mind control
technology, in which case it is fairly obvious that the mind controllers will
end up either dead or in control of the entire economy.

2) Creating an upload doesn't just increase the labor supply. It also
increases demand, since the upload will want to spend his wages on something.
Upload copying is thus exactly analogous to ordinary population growth - it is
just a faster, potentially less costly way of doing the same thing. Now, it is
obvious that ordinary population growth hasn't driven wages down or led us to
enslave unpopular segments of the population, so why should the situation with
uploads be any different?

Billy Brown



This archive was generated by hypermail 2b29 : Mon Oct 02 2000 - 17:34:37 MDT