On April 23-24, the Center for the Study of Market Processes in conjunction
with the Washington Evolutionary Systems Society held a conference entitled
"Evolutionary Economics: Learning from Computation." An outgrowth
of the Center's Agorics Project, the conference explored the overlap among
evolutionary economics, computer modeling of complex evolving systems, and
machine learning.
Our first speaker was Peter M. Allen of the International Ecotechnology
Research Center in England. Allen described the complex, non-linear, dynamic
interrelationships between economic, environmental, and cultural factors.
Using computer models of the Nova Scotian shelf fisheries that ran as he
spoke, Allen demonstrated that it is the non-average detail of time and
place which drives evolution, and that there can be no equilibrium in a
dynamic economy.
John H. Miller of the Santa Fe Institute
spoke on classifier systems and genetic algorithms. These evolutionary machine
learning techniques incorporate a feedback system by which they adjust to
success or failure. The feedback system is based on economic principles:
in classifiers, for example, successful rules essentially "pay"
other rules for useful information.
Paul Werbos of the National Science Foundation spoke on neural networks,
another machine learning technique incorporating economic principles. With
neural networks, as with classifier systems, there is a kind of payment
back through the system by which success is rewarded and the system evolves.
Mark S. Miller, chief architect
of the Xanadu Operating Company in Palo Alto, California, spoke on Agoric
Open Systems, a logical next step in the evolution of computational
systems from closed and centrally controlled to open and evolving. Miller
discussed computational processes built on analogs to market principles
of property rights and competitive bidding (e.g. for processor time
and space in core main memory). Such processes allow for greater complexity
and efficiency in computer systems.
Other important guests were Robert W. Crosby of the Washington Evolutionary
Systems Society, who helped organize the conference, and Jack Corliss of
the Computer Systems Research Facility at Goddard Space Flight Center, who
showed a startling videotape of complex evolutionary dynamics.
The conference made clear that examining the characteristics of these various
methodologies can help us better understand economic processes. It also
suggested possible applications of these various methods that may help us
build a more integrative approach to economic problems.
Howard Baetjer is a PhD candidate at the George Mason University Department
of Economics. He is a member of the Agorics Project, directed by Prof. Don
Lavoie. This article is reprinted from the newsletter @UX{Praxis}. The Project
may be contacted at the Center for the Study of Market Processes, phone
703-323-3483, fax 703-764-6323. For papers on Agorics, see the Japan
Prize article below.
This year's Japan Prize (the Nobel Prize might be termed the "Sweden
Prize") was awarded to Marvin
Minsky of MIT for his pioneering work in artificial intelligence. It
includes $318,000 and a meeting with Emperor Akihito. Minsky, now a professor
at MIT's Media Lab, founded the MIT Artificial Intelligence Lab and serves
on the Foresight Institute Board of Advisors.
In other foresighted news from Japan, their new International Institute
for Novel Computing will have twelve subcommittees to study areas of future
computing. Three of these are of particular interest to FI members:
Molecular computing. Computational systems built from molecular components
will be a major payoff of progress toward nanotechnology.
Social-hyper computing. This presumably refers to social software
and open hypertext. For more information on this general area, request "Hypertext
Publishing and the Evolution of Knowledge" from FI. In the U.S.,
include a large stamped, self-addressed envelope with 65 cents postage.
Evolutional computing. This may include work on agoric systems papers
as discussed in The Ecology of Computation, an Elsevier Science
Publishers book reviewed previously in Update. A limited number
of agoric systems
papers have been given to FI by coauthor Mark S. Miller and are available
on request. A donation to cover our costs is appreciated: $3.50 in the U.S.
For computing news from both the U.S. and Japan, with a strong focus on
neural computing, a good source is the newsletter Intelligence,
edited by Edward Rosenfeld. While it is expensive ($295 in North America,
$350 outside), it has the latest news on neural computing and we find it
of great value. The publication has been coming out for six years so it
seems likely to last. A typical issue is 8 pages, published monthly. They
can be reached at PO Box 20008, New York, NY 10025, or by phone 212-222-1123
or 800-NEURALS.
The Office of Technology Assessment in Washington, DC, is beginning a study
of the enabling technologies leading to nanotechnology. These will include
bottom-up approaches such as STMs, bioengineering, and synthetic chemistry
as well as top-down approaches such as lithography. They are grouping these
topics under the general term 'miniaturization.'
They are seeking applicants for an 8-10 month interim position to study
these areas. The ideal candidate would be a recent PhD in physics or engineering,
but perhaps others would be considered, since no one candidate will already
be knowledgeable in such a wide range of fields. Having completed the degree
may not be strictly necessary. OTA plans to make a decision and have the
candidate begin 'as soon as possible.'
This is a unique opportunity for a researcher to broaden his or her technical
background and have an influence on U.S. technology policy. OTA will be
presenting their results in a report to Congress.
If you know of a potential candidate, have him or her send a resumé
to:
Dr. James Curlin
Program Manager
Communication and Information Technologies
Office of Technology Assessment
Washington, DC 20510-8025
202-228-6760
The translators requested in the last issue have volunteered--see Thanks
column--but we could use more translators from Japanese to English, especially
those willing to do long pieces. The translations usually are not time critical
and can be done over a few weeks. We would also appreciate additional volunteer
help in Macintosh desktop publishing, especially layout.
FI needs equipment, new or used: a small photocopier, two fax machines,
and a second Laserwriter printer. Note that donations of equipment or funds
are tax-deductible as charitable contributions.
If you or your company can help, call our office at 415-324-2490.
NASA and Self-Replicating Systems:
Implications for Nanotechnology
by Ralph Merkle
In the summer of 1980, NASA and the American Society for Engineering Education
(ASEE) sponsored a summer study by 15 NASA program engineers and 18 educators
from U.S. universities to investigate advanced automation for space missions.
The resulting 400-page report included a 150-page chapter on "Replicating
Systems Concepts: Self-Replicating Lunar Factory and Demonstration"
which proposed a 20-year program to develop a self-replicating general purpose
lunar manufacturing facility (a Self Replicating System, or SRS) that would
be placed on the lunar surface. The design was based entirely on conventional
technology.
The "seed" for the facility, to be landed on the lunar surface
from Earth to start the process, was 100 tons (approximately four Apollo
missions). Once this 100-ton seed was in place, all further raw materials
would be mined from the lunar surface and processed into the parts required
to extend the SRS. A significant advantage of this approach for space exploration
would be to reduce or eliminate the need to transport mass from the Earth--which
is relatively expensive.
The report remarks that "The difficulty of surmounting the Earth's
gravitational potential makes it more efficient to consider sending information
in preference to matter into space whenever possible. Once a small number
of self-replicating facilities has been established in space, each able
to feed upon nonterrestrial materials, further exports of mass from Earth
will dwindle and eventually cease. The replicative feature is unique in
its ability to grow, in situ, a vastly larger production facility
than could reasonably be transported from Earth. Thus the time required
to organize extraordinarily large amounts of mass in space and to set up
and perform various ambitious future missions can be greatly shortened by
using a self-replicating factory that expands to the desired manufacturing
capacity."
"The useful applications of replicating factories with facilities for
manufacturing products other than their own components are virtually limitless."
Establishing the credibility of the concept occupied the early part of the
chapter. The theoretical work of Von Neumann was reviewed in some detail.
Von Neumann designed a self-replicating device that existed in a two-dimensional
"cellular automata" world. The device had an "arm" capable
of creating arbitrary structures, and a computer capable of executing arbitrary
programs. The computer, under program control, would issue detailed instructions
to the arm. The resulting universal constructor was self-replicating almost
as a by-product of its ability to create any structure in the two-dimensional
world in which it lived. If it could build any structure it could easily
build a copy of itself, and hence was self-replicating.
Self-replicators need not be vastly complex
One interesting aspect of Von Neumann's work is the relative simplicity
of the resulting device: a few hundred kilobits to a megabit. Self-replicating
systems need not inherently be vastly complex. Simple existing biological
systems, such as bacteria, have a complexity of about 10 million bits. Of
course, a significant part of this complexity is devoted to mechanisms for
synthesizing all the chemicals needed to build bacteria from any one of
several simple sugars and a few inorganic salts, and other mechanisms for
detecting and moving to nutrients. Bacteria are more complex than strictly
necessary simply to self-reproduce.
Despite the relative simplicity that could theoretically be achieved by
the simplest self-reproducing systems, the proposed lunar facility would
be highly complex: perhaps 100 billion to a trillion bits to describe. This
would make it almost 10 thousand to 100 thousand times more complex than
a bacterium, and a million times more complex than Von Neumann's theoretical
proposal. This level of complexity puts the project near the limits of current
capabilities. (Recall that a major software project might involve a few
tens of millions of lines of code, each line having a few tens of characters
and each character being several bits. The total raw complexity is about
10 billion bits--perhaps 10 to 100 times less complex than the proposed
SRS.) Where did this "excess" complexity come from?
The proposed SRS has to exist in a complex lunar environment without any
human support. The complexity estimate for the orbital site map alone is
100 billion bits, and the facilities for mining and refining the lunar soil
have to deal with the entire range of circumstances that arise in such operations.
This includes moving around the lunar surface (the proposal included the
manufacture and placement of flat cast basalt slabs laid down by a team
of five paving robots); mining operations such as strip mining, hauling,
landfilling, grading, cellar-digging and towing; chemical processing operations
including electrophoretic separation and hydrofluoric acid leach separation,
the recovery of volatiles, refractories, metals, and nonmetallic elements
and the disposal of residue and wastes; the production of wire stock, cast
basalt, iron or steel parts; casting, mold-making, mixing and alloying in
furnaces and laser machining and finishing; inspection and storage of finished
parts, parts retrieval and assembly and subassembly testing; and computer
control of the entire SRS.
When we contrast this with a bacterium, much of the additional complexity
is relatively easy to explain. Bacteria use a relatively small number of
well-defined chemical components which are brought to them by diffusion.
This eliminates the mining, hauling, leaching, casting, molding, finishing,
and so forth. The molecular "parts" are readily available and
identical, which greatly simplifies parts inspection and handling. The actual
assembly of the parts uses a single relatively simple programmable device,
the ribosome, which performs only a simple rigid sequence of assembly operations
(no AI in a ribosome!). Parts assembly is done primarily with "self-assembly"
methods which involve no further parts-handling.
Another basic issue is closure. "Imagine that the entire factory and
all of its machines are broken down into their component parts. If the original
factory cannot fabricate every one of these items, then parts closure does
not exist and the system is not fully self-replicating." In the case
of the SRS, the list of all the component parts would be quite large. In
the case of a bacterium, there are only 2,000 to 4,000 different "parts"
(proteins). This means that the descriptions of the parts are less complex.
Because most of the parts fall into the same class (proteins), the manufacturing
process is simplified (the ribosome is adequate to manufacture all proteins).
What does all this mean for humanity? The report says "From the human
standpoint, perhaps the most exciting consequence of self-replicating systems
is that they provide a means for organizing potentially infinite quantities
of matter. This mass could be so organized as to produce an ever-widening
habitat for man throughout the Solar System. Self-replicating homes, O'Neill-style
space colonies, or great domed cities on the surfaces of other worlds would
allow a niche diversification of such grand proportions as never before
experienced by the human species."
The report concludes that "The theoretical concept of machine duplication
is well developed. There are several alternative strategies by which machine
self-replication can be carried out in a practical engineering setting....There
is also available a body of theoretical automation concepts in the realm
of machine construction by machine, in machine inspection of machines, and
machine repair of machines, which can be drawn upon to engineer practical
machine systems capable of replication. . . . An engineering demonstration
project can be initiated immediately, to begin with simple replication of
robot assembler by robot assembler from supplied parts, and proceeding in
phased steps to full reproduction of a complete machine processing or factory
system by another machine processing system, supplied, ultimately, only
with raw materials."
What implications does the NASA study have for nanotechnology?
The broad implications of self-replicating systems, regardless of scale,
are often similar. The economic impact of such systems is clear and dramatic.
Things become cheap, and projects of sweeping scale can be considered and
carried out in a reasonable time frame without undue expense.
The concepts involved in analyzing self-replicating systems--including closure,
parts counts, parts manufacturing, parts assembly, system complexity, and
the like--are also quite similar. The general approach of using a computer
(whether nano or macro) to control a general purpose assembly capability
is also clearly supported. Whether the general-purpose manufacturing capability
is a miniature cross-section of current manufacturing techniques (as proposed
for the SRS), or simply a single assembler arm which controls individual
molecules during the assembly process, the basic concepts involved are the
same.
Finally, by considering the design of an artificial SRS in such detail,
the NASA team showed clearly that such things are feasible. Their analysis
also provides good support for the idea that a nanotechnological "assembler"
can be substantially less complex than a trillion bits in design complexity.
There are several methods of simplifying the design of the "Mark I
Assembler," as compared with the NASA SRS. First, it could exist in
a highly controlled environment, rather than the uncontrolled lunar surface.
Second, it could expect to find many of its molecular parts, including exotic
parts that it might be unable to manufacture, pre-fabricated and provided
in a convenient and simple format (e.g., floating in solution). Third,
it could use simple "blind," fixed-sequence assembly operations.
Conceptually, the only major improvements provided by the Mark I Assembler
over a simple bacterium would be the general purpose positional control
it will exert over the reactive compounds that it uses to manufacture "parts,"
and the wider range of chemical reactions it will use to assemble those
"parts" into bigger "parts." Bacteria are able to synthesize
any protein. The Mark I Assembler would be able to synthesize a much wider
range of structures. Because it would have to manufacture its own control
computer as a prerequisite to its own self-replication, it would revolutionize
the computer industry almost automatically. By providing precise atomic
control even the Mark I Assembler will revolutionize the manufacturing process.
Copies of "Advanced Automation for Space Missions" are available
from NTIS. Mail order: NTIS, U.S. Department of Commerce, National Technical
Information Service, Springfield, VA. 22161. Telephone orders with payment
via major credit cards are accepted; call 703-487-4650 and request "N83-15348.
Advanced Automation for Space Missions." Purchase price is about $40.00,
various shipping options are available.
Dr. Merkle's interests range from
neurophysiology to computer security; he is a researcher at Xerox Palo Alto
Research Center.
