Re: Why would AI want to be friendly?

From: Eugene Leitl (eugene.leitl@lrz.uni-muenchen.de)
Date: Tue Sep 26 2000 - 01:16:42 MDT


Franklin Wayne Poley writes:

> Well, JR, my investigation of the matter tells me that the knowledge in
> the public domain alone is sufficient that a mega-project could be
> undertaken now to develop AI to surpass human equivalency and it would be
> analogous to the man-on-the-moon effort from a 1960 vantage point. In
> other words, put 10 years and x billion $ into it and we can expect to

Or one year, and 10*x billion? Or 100*x billion?

Sorry, that doesn't work that way. You can hurry certain developments
by throwing resources on it, up to a point. But some projects just
need input from too many fields (which also have to be developed) to
soak for a while before generating output. In fact some projects will
become later, if given more resources.

To start with, you won't have enough crunch by 2010 to be able to run
anything serious in a single installation. And lack of bandwidth and
too high latency between distant installations will bottleneck the
total size of what you can run. And you can't invest in hardware,
because after a few months something much better comes along.

One architecture which commodity computing hasn't produced yet is a
scalable embedded RAM based neuronal accelerator engine. But, of
course, the fastest engine will be quite useless if you're running the
wrong model. And I doubt neuroscience will find it all out in just a
decade.

You might be able to hurry up molecular circuitry development by
pumping a large percentile of the GNP of major industrial nations, but
I doubt it. Your bottleneck will be people who know what they're
doing. You can't train them fast enough, and they'll be essentially
glorified administrators, commanding hordes of less skilled people.

> succeed. Between 2000 and 2010 might require tens of billions, even
> hundreds of billions of dollars but we can plan it out now and we can see

If you want to fund a worthwhile project, which doesn't in the end
turn round and bite clean your head off, why don't you fund uploading
critters. A virtual nematode (or even a rodent) in a simulation
sandbox is unlikely to come out and devastate the countryside.



This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 15:31:13 MST