Investigations: On the Nature of Autonomous Agents
Metanexus: Views. 2002.02.17. 1487 words
Stuart Kauffman has coined the term ‘autonomous agents’ to characterize aprogram of research aimed at explaining how a system can have ‘a life of itsown.’ He is a biophysicist and complexity theorist with his own theory ofthe origin of life based on autocatalytic cycles of chemical reactions. ForKauffman, constraints play a key role in the theory of autonomous agents.Another important quality is a type of Godelian incompleteness that permitsthe system to display freedom or spontaneity in its behavior. Kauffman’sideas provide a new definition of life. They may even help us understandhow, with increasing complexity, a physical system can leap from being mereclod-like matter to being an information-rich participator in a meaningfuluniverse.
In continuation, Paul Davies also observes that in this Kauffman is bringingtogether two of Wheeler’s more prominent ideas–that of the participatoryuniverse and the idea of it from bit. And thus we continue our specialseries on the VIEWS list in anticipation of the Science & Ultimate RealitySymposium in Princeton. This symposium in honor of the 90th year of JohnArchibald Wheeler–a great physicist and teacher of physicists–runs fromMarch 15-18, 2002.
To get more information or to register for the Science & Ultimate RealitySymposium at Princeton, go to <http://www.templeton.org/ultimate_reality>.We hope to see many of you there. You can also subscribe to this listindependently of VIEWS by going to<http://listserv.metanexus.net/metanexus/archives/wheeler.html>. You canreply to this message or send a new message for distribution on theconference list to <email@example.com>. This is a moderatedemail distribution list, so all messages will be approved by Davies torestrict the quantity and maintain the quality of the discussion, and wewill be cross posting many of the messages here on Metanexus: VIEWS.
— Stacey Ake
Science & Ultimate Reality
Perhaps the most provocative of Wheeler’s ideas is that of theparticipatory universe in which observership assumes a central place inthe nature of physical reality, and presumably at some level must enter intophysical theory. But what exactly constitutes a participant/observer? Is aparticle detector enough? A living organism? An information gathering andutilizing system (IGUS)? A human being? A community of physicists?
Melding the participatory universe with it from bit reveals the keyconcept of information at the core, and moreover working both ways. On theone hand an observation involves the acquisition and recording ofinformation. On the other hand an observer, at least of the living variety,is an information processing and replicating system. In both cases it is notinformation per se that is crucial, but semantic information. An interactionin quantum mechanics becomes a true measurement only if it means somethingto somebody (made explicit in Wheeler’s meaning circuit). Similarly, theinformation in a genome is a set of instructions (say, to build a protein)requiring a molecular milieu that can recognize, decode and act upon it. Thebase-pair sequence on a strand of DNA is just so much goobledygook withoutcustomized cellular machinery to read and interpret it.
Where is there room in physics for the notion of information, not merely inthe blind thermodynamic sense, but in the active life/observation/meaningsense? How does a lofty, abstract notion like meaning or semanticinformation emerge from the blundering, purposeless antics of stupid atoms?
Part of the answer must involve the subtle concept of autonomy. Livingorganisms are recognized because they really do have a life of their own.A cell is subject to the laws of physics, but it is not a slave to them:cells harness energy and deploy forces to suit their own ends. How does thisquality of autonomy arise? Clearly the system must be open to itsenvironment: there must be a throughput of matter, energy and – crucially –
information. But more is needed. When my computer plays chess, the shapesmove around on the screen in accordance with the rules of chess. But mycomputer is also subject to the laws of physics. So are the rules of chesscontained in the laws of physics? Of course not. The chess-playingregularities are an emergent property in the computer, manifested at thehigher level of software, not in the bottom level of hardware (atoms andelectrons). Trace back how the rules of chess work in the computer and youwill discover that constraints are the answer. The physical circuitry isconstrained to embody the higher-level rules.
Stuart Kauffman has coined the term autonomous agents to characterize aprogram of research aimed at explaining how a system can have a life of itsown. He is a biophysicist and complexity theorist with his own theory ofthe origin of life based on autocatalytic cycles of chemical reactions. ForKauffman, constraints play a key role in the theory of autonomous agents.Another important quality is a type of Godelian incompleteness that permitsthe system to display freedom or spontaneity in its behavior. Kauffman’sideas provide a new definition of life. They may even help us understandhow, with increasing complexity, a physical system can leap from being mereclod-like matter to being an information-rich participator in a meaningfuluniverse.
Title: Investigations: On the Nature of Autonomous Agents
Author: Stuart Kauffman
Investigations, my third book, is the strangest intellectual adventure ofmy life. It began as a notebook in December 1994. I sensed that manystrands, each too large to yet be visible, were waiting for exploration.After a year, the notebook was long, and eventually became the book.
I begin with a central image. Consider a bacterium swimming up a glucosegradient. We all readily say, without attributing consciousness, that thebacterium is going to get food. That is, the bacterium is acting on its ownbehalf in an environment. I will call a system that can act on its ownbehalf in an environment an autonomous agent. But the bacterium is justa physical system. So my question became: What must a physical system be toconstitute an autonomous agent?
I had not expected even the outlines of the answer I would be led to, nor tothe expanding web of questions I would be led to explore.
In brief summary, I define an autonomous agent as a system that is able toreproduce and also able to carry out at least one thermodynamic work cycle.Importantly, all free- living organisms fulfill this definition. In tryingto find a definition for an autonomous agent, I may have stumbled upon anadequate definition of life itself, but I will not insist on it.
It is interesting that Schrodinger, in his famous What is Life?, missesthe work cycle. Indeed, Schrodinger answered a different initial question,namely, what is the source of order in organisms, and achieved his brilliantinsight into the genetic material as an aperiodic crystal that would carry amicrocode specifying how to build an organism from its genes. YetSchrodinger did not answer the question What is Life? I suspect thatautonomous agents may answer that question. It is, after all, anastonishing fact that autonomous agents do persistently act on the universeon their own behalf. Physics and chemistry will have to contend with thetruth of this fact, and lift themselves to that level that allows thesefields to talk about life.
Investigations is, at least in part, obviously science. In particular, I amled to propose a class of open thermodynamic chemical reaction networks thatare both autocatalytic and carry out work cycles. At a minimum this is anew class of reaction networks that are clearly worthy of study, and mightform the basis of a new technology: self-reproducing chemical robots able tobuild things.
But Investigations took me further, to a critique of the concept of work,the use of Atkin’s definition of work as the constrained release of energy,coupled with the puzzled realization that it typically takes work to createconstraints, and constraints to create work, to an attempt to portraypropagating work, to the almost certainly true realization that we cannotfinitely prestate all the possible Darwinian pre-adaptations that can arise,hence that we cannot prestate the, typically collective, variables thatconstitute such adaptations, to the realization that, at the level ofcomplex molecules and above, the universe is grossly non-ergodic, to adefinition of the adjacent possible, and a candidate law for biospheresanywhere in the universe: such biospheres may, as a secular trend, tend toexpand into the adjacent possible such that the diversity of what can happennext increases, on average, as rapidly as it can.
I find Investigations both deeply interesting, yet deeply puzzling. One ofthe puzzles is how we can find a mathematical foundation for its basicconcepts when we apparently cannot say ahead of time what the variables, theDarwinian pre-adaptations, of a biosphere will be.
This publication is hosted by Metanexus Online http://www.metanexus.net. The views expressed here do not necessarily reflect those of Metanexus or its sponsors.
Metanexus welcomes submissions between 1000 to 3000 words of essays and book reviews that seek to explore and interpret science and religion in original and insightful ways for a general educated audience. Previous columns give a good indication of the topical range and tone for acceptable essays. Please send all inquiries and submissions to. Metanexus consists of a number of topically focused forums (Anthropos, Bios, Cogito, Cosmos, Salus, Sophia, and Techne) and periodic HTML enriched composite digests from each of the lists.
Copyright notice: Except when otherwise noted, articles may be forwarded, quoted, or republished in full with attribution to the author of the column and Metanexus: The Online Forum on Religion and Science. Republication for commercial purposes in print or electronic format requires the permission of the author. Copyright 1998, 1999, 2000, 2001, 2002, 2003, 2004 by Metanexus Institute.