Foresight Update 7
page 3
A publication of the Foresight Institute
 AI
and Scaling: One Argument Three Ways
by Jeffrey Soreff
If one regards a full, Turing-test-passing artificial
intelligence (AI) as essentially allowing a one-for-one
substitution of computer power for human employees, I believe
that one underestimates the effectiveness that such a development
will have. At least three arguments suggest this:
The argument from co-workers:
In many projects, the ideal procedure would be to have the
chief designer do the top level work, then "xerox"
copies of that person to perform more detailed work, then
"xerox" more copies to do yet more detailed work, and
so on until project completion. In any human enterprise, the time
and effort required to communicate subtasks to those who execute
them is a large cost. Consider how this might work in an
organization with a series of design projects to complete and a
pool of AI-capable computers. The "xeroxing" of the
designers can occur merely by loading the memory of one currently
unused computer from the memory of the one containing the
designer. As long as there is a spare machine being freed up by
one terminating project somewhere, one can use it to duplicate exactly
those skills that the organization curently needs most, rather
those that are available through hiring and training. One need
only assume that machines continue to be built in such a way that
their internal state can be dumped and reloaded.
The argument from subroutines:
Many programs, including AI ones, have economies of scale not
found in human organizations. No organizational tricks seem to
let one aggregate the short term memory of the participants in a
meeting so that the group can keep track of a more complex agenda
than any individual can. The reverse effect seems more typical.
With computers there are well-known techniques (albeit with some
overhead) for allowing a set of communicating machines to
simulate a single, more powerful one; many AI processes (solution
space search, application of large numbers of heuristics, search
for "fuzzy" matches of templates, signal processing in
computer perception, simulation, etc.) can benefit from the
increased power. In general the problems that can be solved scale
up more slowly than linearly with the processing power, but human
organizations often do not solve the equivalent problems at all,
typically doing a less thorough optimization or entirely ignoring
some class of problems.
The argument from raw bandwidth:
There can be considerable dispute on the true bandwidth of
information going into a human. The apparent raw input
capacity of the human visual system looks very large, on the
order of megabits per second. Obviously this is severely filtered
at later stages of processing, but it is difficult to nail the
real numbers down. What is not subject to the same
dispute, however, is the tiny rate at which humans can emit
information. Given the physical limits on the rates of speech and
muscle movement--both in speed and accuracy--no speaker, typist,
or graphic artist can emit more than 100 bits per second or so.
Even today's fiber optic links can easily exceed this by over six
orders of magnitude. AIs with hardware links will be more like
connected lobes of the brain than like humans in an organization.
|
The power of an AI installation
will grow more rapidly with size
than does a human organization |
|
As a consequence of this close coupling, the power of an AI
installation will grow much more rapidly with size than does the
power of an organization with an equivalent number of humans. If
you have one piece of hardware which acts as the equivalent of
one engineer, you have the economic equivalent of one engineer.
If you have ten such pieces of hardware, you have much more than
the equivalent of ten engineers. For similar reasons, advances in
the performance of computing hardware are also of more
consequence than one would expect from a one-for-one
substitution. This tends to speed up the effects of achieving
full AI beyond the usual extrapolations.
Jeffrey Soreff works for IBM and has a background in physics.
In his spare time he is calculating the dynamic friction of two
atomically smooth surfaces sliding past one another, an important
parameter of a class of nanomechanisms termed "van der Waals
bearings" by Drexler.
Conference
Thanks
by K. Eric Drexler, Chairman
Many people worked hard to make the First Foresight
Conference on Nanotechnology a success beyond our
expectations. Our cosponsor, Global Business Network, made a big
difference: Stewart Brand (both a GBN principal and Foresight
Advisor) convinced us to have the meeting and encouraged GBN to
help; Peter Schwartz (President of GBN) approved GBN's
participation and chaired a conference session; and most of all
Danica Remy, whose organizational talent ensured that the meeting
went smoothly despite unexpected problems--such as a major
earthquake not long before the event.
Other help of a sponsorship nature came from Prof. Nils Nilsson,
Chairman of the Dept. of Computer Science, which hosted the
meeting, thus giving us access to Stanford facilities (i.e.
those Stanford facilities still available after the earthquake).
Mark Pearson, head of Molecular Biology at Du Pont, helped with a
generous grant. Ed Niehaus of Niehaus Public Relations donated a
great deal of time and office help to ensure the press's needs
were met. Peter Toldger of Tam Systems donated financial
assistance in renting a fax machine, without which no
international conference can be held.
Many companies donated demonstrations of their products: Biosym,
Digital Instruments, Silicon Graphics, Stardent, Sun, Tektronix,
and Tripos. Being able to see these state-of-the-art products
enabled attendees to understand how rapidly progress is being
made. Special thanks to Michael Pique of Scripps Clinic, who
arranged for these demonstrations. Roy Hovey of Stanford
Bookstore did a super job of finding relevant books and making
them available to conferees (a list of these is available from
Foresight).
Finally, thanks go to those on the Foresight team who worked on
the conference. Chris Peterson, Foresight Editor and Board
member, worked with Danica Remy to make the meeting happen,
especially on registration and general coordination. Ralph Merkle
recruited speakers and sponsors. Russ Mills and Dave Kilbridge
produced conference materials. Jim Lewis is editing the
conference proceedings. Others who volunteered either before or
at the meeting include Mike Butler, Kurt Bohan, Stan Hutchings,
Dave Lindbergh, Nat Stitt, and Dave Wilson.
Based on this success, we plan to hold similar events and to
sponsor other, larger events to enable participation by more
Foresight supporters.
Interdisciplinary
Center
by Christine Peterson
It has increasingly become clear that progress on major new
technologies such as nanotechnology depends on interdisciplinary
efforts. In this area, the U.S. is weak relative to Japan, where
an intense focus on new products tends to topple disciplinary
walls. We are often asked to name universities or other
institutions where work is being done toward nanotechnology; this
is the first in a planned series looking at relevant
interdisciplinary groups.
Center for Biopolymers at Interfaces
Formed in 1986, CBI is a joint university-industry cooperative
research center with a focus on medical applications of
biopolymers. This product-orientation ranges from artificial
organs to "targeted drugs" and protein-based sensors.
Nineteen faculty members drawn from the University of Utah's
Colleges of Engineering, Science, and Pharmacy and its School of
Medicine pursue studies on a broad range of biopolymers such as
proteins and DNA, using techniques including molecular graphics,
scanning tunneling microscopy, and atomic force microscopy.
Utah's product focus has paid off well in the past: it is
surpassed only by Caltech and MIT in the number of spinoff
companies it has produced. The University shows its commitment to
the Center by exempting its industrial membership revenues from
the University's overhead charges. (I don't know what percentage
Utah normally charges, but a 50% burden is not uncommon in
universities.)
|
In Japan, an intense focus on new
products tends to topple
disciplinary walls |
|
CBI's industrial members can both influence and directly
benefit from the Center's research: they vote on which projects
should be pursued, have right of first refusal in licensing
inventions, and can sponsor fellowships to obtain custom research
results. Industrial members pay $15,000 per year and currently
include Du Pont, Eli Lilly, Kodak, Johnson and Johnson, Biosym,
and Silicon Graphics.
CBI may not be the perfect place to pursue nanotechnology work,
due to its focus on biopolymers only, but its interdisciplinary
breadth is an improvement over a standard academic department,
and the industrial participation ensures that work there has
practical results. Students, researchers, and companies
interested in participating can contact the Center at the
University of Utah's Department of Bioengineering, (801)
581-3867.
Thanks to Foresight member Joseph Andrade, Chairman of the
Department of Bioengineering and Co-director of CBI, for sending
us information about the Center. Information on other
interdisciplinary efforts is welcome.
MIT
Nanotechnology Symposium
This fourth symposium on nanotechnology to be held at MIT,
entitled "Nanotechnology: Molecular Engineering and its
Implications," will be sponsored by the MIT Nanotechnology
Study Group. Plans are not yet final, so contact the NSG or the
Foresight office in early January for the correct schedule. The
location will be MIT Room 66-110, in Building 66 (the Chemical
Engineering building) on Ames St. Confirmed speakers and topics
to date:
- Howard C. Berg, Harvard University, on the world's
smallest rotary motor.
- K. Eric Drexler, Stanford University, on nanotechnology
and technical foundations of molecular engineering.
- Bruce Gelin, Polygen Corp., on molecular modeling.
- Gary T. Marx, MIT, on privacy and security issues posed
by molecular engineering.
- Gary Tibbets, General Motors Research Laboratories, on
the growth of nanometer-scale carbon tubes.
- Abraham Ulman, Eastman Kodak Research Laboratories, on
engineering of molecular monolayers.
- Kevin Ulmer, seQ, Ltd., on mosaic tiling with proteins.
[Webmaster's note: See Update 8 for a report on this
symposium.]
In an earlier event at MIT, on November 21 the NSG screened the
nanotechnology documentary "Little by Little" made by
InCA for the British television series Equinox,
similar to the US Nova. The show features Nobel
winning chemist Jean-Marie Lehn, Eric Drexler, John Foster of IBM
Almaden, and others working on the path to nanotechnology. It is
to be shown in the US on the Discovery channel; we will announce
the date if we are notified in time.
Future
Calendar
Foresight members with a taste for space development and
science fiction may enjoy the "Calendar for the Year
2001" (also good for 1990). Twelve speculative scenarios are
included for various years through 2103, some with
nanotechnology. For more information or to order call LaGrange
Publishing at (708) 482-4321.
From Foresight Update 7, originally
published 15 December 1989.
Foresight thanks Dave Kilbridge for converting Update 7 to
html for this web page.
|