While this paper has concentrated on the development of computational autopoiesis, per se, this is clearly related to several other, parallel, developments.
The Chemoton of Tibor Gánti , though apparently developed quite independently, is clearly related to cellular autopoiesis; indeed, this first (English) report on the chemoton concept appeared in the same journal, BioSystems, as the first presentation of Autopoiesis , but in the following year (1975). These parallels--and links to a wide variety of much earlier work--have been more recently reviewed by Gánti .
The chemoton is again a proposal for an abstract, ``minimal'' cell, consisting of a collectively autocatalytic network of reactions which is enclosed within a membrane, which is also generated and maintained by the reaction network. It differs from the minimal autopoiesis model in explicitly including a ``genetic subsystem''. It is also rather more detailed in its analysis of the required chemical dynamics (kinetics etc.), and aims at supporting self-reproduction by growth and fission even in the minimal version.
While the provision of a genetic system is significant, it should be emphasised that, in the basic chemoton, this is limited to replication of varieties of ``polymers'' which then have direct chemical effects. The polymers do not participate in a translation process; thus, this subsystem is again relatively limited in scope compared to a von Neumann style genetic architecture, or programmable constructor.
The chemoton has been subject to various computer simulation studies, particularly by Csendes . However, these were based essentially on an Ordinary Differential Equation approach, rather than being molecular or agent-based. This means that, in particular, the distributed, spatial, dynamics of membrane growth, fission, and individuation, were not substantively modelled.
Other computational systems with some connection or similarity to autopoiesis include Alchemy [8,9], the -universes , Coreworld  and Tierra . However, as already mentioned, while these do involve closed, self-sustaining, reaction networks, they do not have mechanisms for the self-generation of spatial boundaries.
Another very interesting line of work involves much more physico-chemically realistic computer models of minimal cellular structures. Rasmussen and colleagues, in particular, have recently exhibited spontaneous emergence of such hierarchical levels of organisation . This line of attack poses some scaling difficulties, as the computational demands of physico-chemical realism rise very rapidly. On the other hand, the cost of computation continues to fall, so this may well be a very fruitful domain of further research in the immediate future.
Finally, we should note that there are clear conceptual connections between the idea of autopoiesis and Robert Rosen's metabolism-repair systems and ``closure under efficient causation'' . This connection has recently been made explicit by Letelier et al. , with the concrete suggestion that Rosen's work may provide an appropriate formal framework for the understanding of autopoiesis. A similar conjecture has also been recently presented by Nomura .
This approach immediately encounters the difficulty that Rosen explicitly argued that his form of closure could not, even in principle, be embedded within a purely computational system. By contrast, computer realisation has been an explicit exemplar of autopoietic closure from the very start. As already discussed, Letelier et al.  have attempted to resolve this by interpreting  as showing that computational autopoiesis was fundamentally flawed from the beginning; but this is not an interpretation I can share. Accordingly, at this point, while I would agree that Rosen's work presents an interesting avenue for further exploration and contrast with autopoiesis, it is certainly not yet demonstrated than it can contribute a substantive, formal, theory to underpin it.
Copyright © 2004 All Rights Reserved.