Re: Darwinian Extropy

Robin Hanson (hanson@hss.caltech.edu)
Mon, 23 Sep 96 21:57:45 PDT


Dan Clemmensen writes:
>> >I believe that an SI with a computer component will be
>> >able to design and build ever more powerful hardware and software,
>> >thus increasing its own capabilities. I think that this is likely to
>> >lead not just to an improvement, but to a rapid feedback process.
>... Our understanding of computer hardware and software is
>considerably more complete than is our understanding of the world
>>economy, and we have demonstrated the ability to continue to increase
>the capabilities of both hardware and software enormously since the
>development of the computer. Furthermore, computers are already being
>used to assist in the further development of computing. Good changes
>are not hard to find, and 18 month's worth of development results in a
>doubling of capability. Yes, I do "imagine" that there are a long
>series of "big wins". I base this on the recent 30-year trend.
>... Instead, we will develop
>an intelligent entity that has a computer as one component. This
>entity will be smart enough to develop new hardware and software to
>augment itself. This the fundamentally new factor. ...
>We've actually already developed the primitive precursors of this
>entity. a computor development organization that uses its own
>computers to develop the next generation computer, or a software tools
>shop that uses its own tools to develop its next-generation tools, is
>such an entity. However, these primitive examples are not yet focused
>on self-augmentation, and are not tightly-integrated enough to
>precipitate a runaway fast-feedback of self-augmentation.

I agree that the 18 month doubling rate of capacity is likely to
continue. However, I'm not convinced that this rate depends much on
the "IQ" of the researchers involved.

If this doubling rate were very sensitive to the total number or
researchers, lots more researchers would be hired to speed up the
process. If it depended sensitively on the IQ of these researchers,
a higher premium would be placed on hiring high IQ researchers.

To the contrary, there are lots of good ideas, and the big expense is
doing the grunge work of trying them out. Raw smarts helps a little
in picking the winners, but not that much. How much grunge work gets
done depends on the size of the global market for computers.

Tightly integrated self-augmentation groups *have* been tried, with
poor success. The fact is that progress depends much more on lots of
little improvements across the whole economy than many people want to
admit.

So uploads and artificial intelligences will improve their hardware
and software, increasing the productivity of mental labor at a faster
rate than has been possible so far. But that doesn't imply a flash
of super-accelerating growth - just a faster growth rate.

Robin D. Hanson hanson@hss.caltech.edu http://hss.caltech.edu/~hanson/