There exists a large body of work by two Chilean biologists, Humberto Maturana and Francisco J. Varela, usually referred to collectively as Autopoietic theory. At the heart of this work lies the description of a process, called 'Autopoiesis'.

This body of theory concerns the dynamics of living systems, purporting to answer the question "what is the characteristic organization of living systems?" The process of Autopoiesis lies at the heart of the answer.

Maturana-the-biologist was unhappy with enumerating features of living systems to define 'life', and wanted to capture the invariant feature of living systems around which natural selection operates. He wanted to do this in a way that retained the autonomy of living systems as a central feature, and hence without recourse to referential concepts like 'purpose' or 'function'.

The process called 'Autopoiesis' purports to capture this invariant feature of living systems, which are characterised as 'living machines'.

The wider body of Autopoietic theory expresses this idea, and a number of other ideas supporting and expanding on it, representing a complete and coherent worldview. This body of theory contains novel and significant perspectives on a number of familiar phenomena and concepts, such as behaviour and observation.

It is not hard to see why this theory might appeal to a variety of other academic disciplines - the constructs involved are based on dynamics and relationships, not the particular 'stuff' or constitution of a system (except in so far as they instantiate the relevant dynamics), and hence permit their potential application to anything that can be expressed in terms of 'systems'.

Some care needs to be taken however, not to just loot the theory for promising labels and concepts, whilst abandoning some of the deeper ideas and perspectives that are required for the theory to function healthily.

For example, all of the operational features of an Autopoietic system are active and referred to solely within the bounds of that system - this concept of 'closure' is central to Maturana and Varela's interpretation of 'autonomy'. It would be odd to claim that a software Agent with explicitly represented knowledge, beliefs and so on were 'Autopoietic.' Under Autopoietic theory, features like knowledge and beliefs arise in the domain of the observer - someone watching a system interact with its environment in such a way as to prompt the use of such terms - they are not found 'in' systems. Maturana and Varela's work offers a very different way of understanding cognitive phenomena.

This is a pseudo-Ancient Greek word, formed through the conjunction of two 'proper' Ancient Greek words, 'auto' meaning 'self', and 'poiesis' meaning (roughly) 'creation' or 'production' (careers long ago usually fell under 'praxis', the path of arms, or 'poiesis', the path of letters).

'Poiesis' is pronounced 'po-E-sis'. In Maturana's Chilean accent it comes out something like 'p-yey-sis'.

In Maturana (1980), Autopoiesis (the process) is defined through a definition of 'living machines' (see above):

"An autopoietic machine is a machine organised (defined as a unity) as a network of processes of production (transformation and destruction) of components that produces the components which: (i) through their interactions and transformations continuously regenerate and realise the network of processes (relations) that produced them; and (ii) constitute it (the machine) as a concrete unity in the space in which they (the components) exist by specifying the topological domain of its realisation as such a network." (Pp. 78-79)

This is in effect an abstract cybernetic description of cell metabolism. Put *very* crudely, it reads something like: a system is Autopoietic if the bits and pieces of which it is composed interact with each other in such a way as to continually produce and maintain that set of bits and pieces and the relationships between them.

This is the defining 'organization' of Autopoietic systems, and in any particular Autopoietic system this organization is instantiated or embodied in its 'structure' (the actual bits and pieces, and the relationships between them).

The 'private world' of the Autopoietic system is based on maintaining this organization. They try to keep as stable as possible this dynamic process, which is in turn the basis for their identity - as an ongoing process, rather than a particular set of bits and pieces, which might change over time.

Systems are structure determined. That is, anything a system does at any moment in time is determined by its structure - its component bits and pieces, and the relationships between them. Maturana and Varela are at pains to take account of the perspective of the observer when talking of systems and how they behave in relation to their environment. The behaviour of a system is something ascribed to it by someone observing it in interaction with its environment. Hence behaviour is not something that is 'in' a system, and to refer to how a system relates to its environment whilst trying to understand it as an autonomous entity violates that very notion of autonomy. This is why all of the mechanics of the process of Autopoiesis as described by Maturana and Varela are kept strictly within the bounds of the Autopoietic system. This strict requirement is enforced via concepts like 'operational closure' and 'organizational closure.'

The consequences of this perspective are not always obvious. A good example however, is the immune system's ability to distinguish between self and non-self. Varela has been pointing out for some time that this is an observed behaviour, produced by the operational dynamics of the immune system in its environment, and that it is wrong to look for some discriminatory recognition mechanism within the immune system. Attention should be focused on the internal dynamics of the immune system, and how this is affected by and affects its environment of operation in such a way as to give rise to the behaviour observed. A similar approach is taken to the nervous system.

Autopoietic theory of course recognises that systems exist within environments, relate to them, and at low enough material level are entirely open to them.

This construct allows for both the total operational autonomy of Autopoietic systems (in terms of their closure, or self-referentiality), and for intimate and mutually affecting relationships between such systems and their environments.

The result of structural coupling is an autonomous and strictly bounded system, that has nevertheless been shaped extensively by its interactions with its environment over time, just as the environment has been shaped by its interactions with the system. Note that this is not the 'fitting' of a system to its environment, but rather a congruence between systems and environment, arising from the changes that each prompts in the other.

Structural coupling in biological systems arises as a result of the plasticity of their structures, and the plasticity of the structure of the environment. As suggested above, systems are structure determined - how they respond to environmental events, and indeed what events they respond to is something which is determined by their structure at a given moment in time (the same goes for the environment). A plastic structure is one that can be affected by outside events - it can be 'perturbed'. So, structure determines what a system 'does', and if the structure of a system changes, what it 'does' is likely to change - in a manner determined by the system's structure. The same goes for the environment. Over time, the structure of both system and environment change as a result of mutual non-destructive perturbations, and we are likely to see a system responding to events in its environment in an 'appropriate manner' - minimally, the system is not destroyed. Maturana and Varela define 'cognition' in terms of this kind of basic ability to respond to environmental events. Figure 1 is a typical diagrammatic representation of this process.

Things get really interesting with structural coupling when two Autopoietic systems perturb one another over a period of time. This is the basis for language, characterised by Maturana and Varela as behavioural coordination through mutual and recursive structural change. An interesting thought is that under this theory, language necessarily arises given certain sorts of agents. 'Behavioural coordination' is used to reflect the way in which two mutually perturbing systems affect one another's structure, which in turn affects the sort of behaviour that they manifest. 'Recursive' is used to reflect an ongoing cycle, where each structural state of each of the systems arises as a result of the cumulative effects of all the previous interactions. Maturana and Varela use the term 'languaging' to capture the nature of language as an ongoing and situated activity.

This perspective on language actually fits neatly into a long and well-heeled tradition of understanding language as situated practice used to coordinate actions and behaviour. This is the basis, for example of Ludwig Wittgenstein's well-established philosophy of language based on 'language-games' - the 'things we do' with language as a basis for their semantic significance. This sort of view of language is grounded in connotation rather than denotation, and has been put to successful practical use in robotics -see, for example, the work of Luc Steels with the Origins of Language Group at the AI department of the Free University of Brussels, and the doctoral research of Aude Billard, working in the Mobile Robots Group at the University of Edinburgh's AI department. See also more formal theoretical work by Ezequiel Di Paolo at the School of Cognitive Science (COGS) in Sussex.

A number of people working in social systems theory are interested in the idea of using the process of Autopoiesis to understand social systems themselves, as self-contained and self-maintaining dynamic systems of conventions and so forth. This might be an interesting area for study using agents, not least because one could escape the need to produce agents that are themselves Autopoietic, and instead concentrate on the interactions and relationships between them.

One potential caveat here is that Maturana is unhappy about ascribing the process of Autopoiesis to social systems. Rather, he sees social systems and their qualities as arising or emerging as a result of the ongoing Autopoiesis of the individual biological components of those systems, not as things that are themselves Autopoietic.

John Mingers has written an introductory overview of Autopoiesis, which also highlights its application in a number of other academic areas:

J. Mingers, "An introduction to Autopoiesis - implications and applications," Systems Practice, vol. 2 (2), pp. 159-180, 1989.

There is also a much more comprehensive book:

J. Mingers, Self-producing Systems: Implications and Applications of Autopoiesis. New York: Plenum Publishing, 1994.

Both of these are very comprehensible.

There is also a good basic introduction to the idea of Autopoiesis in:

T. Winograd and F. Flores, Understanding Computers and Cognition. Reading, MA: Addison-Wesley, 1986.

The 'core' work by Maturana and Varela (certainly the most frequently referenced) is:

H. R. Maturana and F. J. Varela, Autopoiesis and Cognition. Dordrecht, Holland: D. Reidel, 1980.

See also:

F. J. Varela, Principles of Biological Autonomy. New York: North Holland, 1979.

However, these (particularly the first) are pretty heavy going for the uninitiated. A textbook style introduction by Maturana and Varela is available in the form of:

H. R. Maturana and F. J. Varela, The Tree of Knowledge. Boston, MA: Shambhala Publications, 1987.

Many more publications are available - see 'Links', below. It is worth noting that Maturana and Varela went off in separate directions after their early work together. In particular, Varela has worked towards integrating the ideas and perspectives of Autopoiesis into the scientific mainstream. More recent works that might be of interest are:

H. Maturana, J. Mpodozis, and J. C. Letelier, "Brain, language and the origin of human mental functions," Biological Research, vol. 28, pp. 15-26, 1995.

F. J. Varela, E. Thompson, and E. Rosch, The Embodied Mind. Cambridge, MA: MIT Press, 1991.

F. J. Varela, "Patterns of life: Intertwining identity and cognition," Brain and Cognition, vol. 34, pp. 72-87, 1997.

By far the most useful online resource is Randall Whitaker's Observer Web. In particular, there is an extensive bibliography (with links to referenced works that are available online), and the Encyclopaedia Autopoietica, containing extensive detail on most aspects of Autopoietic theory, with reference to Maturana and Varela's published work. Whitaker has also written an introduction to Autopoiesis for the ACM (Office Information Systems SIG).

Other online papers by Maturana can be found here.

Maturana's official homepage has some links to some more recent papers that are not likely to be of so much interest to the Agents community, and ... not an awful lot else besides. There is a slightly bizarre 'Ask Dr. Maturana' section.

Those not averse to a Philosophical level of analysis might want to look at Kent Palmer's Thinknet pages, particularly the Autopoiesis section, and the links to Autopoiesis related websites.

Barry McMullin at Dublin's City University has modelled Autopoiesis as a formal process using the Santa Fe Institute's SWARM simulation system, extending work by Varela in the mid-70s. Many of his papers are available online, via his publications page.

George Kampis has written a number of papers relating to Autopoiesis and computation. In addition to the one cited in the Observer Web bibliography see also:

G. Kampis, "The inside and outside views of life," in Proceedings of the Third European Conference on Artificial Life, pp. 95-104, 1995.

G. Kampis, "Life-like computing beyond the machine metaphor," in Computing with Biological Metaphors, R. Paton, Ed. London: Chapman and Hall, 1994.

People working on consensus, behavioural coordination or related and potentially 'Autopoiesis-friendly' phenomena include: Ezequiel Di Paolo, Aude Billard, Luc Steels, Luis Rocha and Alexander Riegler.

See also the school of thought known as Constructivism, and this resource-site for non-Cartesian cognitive science (Cartesian in the sense of reasoning as an abstracted, explicit process, removed from the material world - derived from Descartes' sharp mind-body dualism that has been such an influence on Modernism).