As already explained, a natural research programme in computational autopoiesis is to attempt to realise Darwinian evolution among lineages of autopoietic cells. This requires--among other things--the realisation of cells that are at least capable of self-reproduction through simple growth and fission.
In pursuing this programme, an initial problem which arises then is individuation.
My own exploration of this problem  begins with a conceptual comparison of autopoiesis and Stuart Kauffman's idea of a ``collectively autocatalytic set'' . I conclude that there is a critical distinction between these precisely in that autopoiesis requires self-generated ``individuation''. This is classically captured by stating that the autopoietic network of processes must give rise to a boundary; however, it turns out that it is not entirely clear what should qualify, in general, as a ``boundary''. In  I suggest a heuristic test for this, which essentially asks whether two putatively individual cells, in direct contact with each other, can reliably maintain their separate identities.
This test is, of course, motivated directly by the issue of self-reproduction. In the situation of self-reproduction of an autopoietic cell by growth and fission, then the parent and offspring will clearly initially be in direct proximity to each other. If they fail to maintain their separate identities--if they can as easily merge back into one cell--this would hardly qualify as ``self-reproduction'' in the functional sense required for evolution, as such merging would undermine the very Malthusian population growth required for Darwinian selection.3 This then is precisely the acid test for individuation: whether an entity can maintain its self-identity when it interacts with another entity of exactly the same material components and organisation.
The test was first applied to a variety of ``classical'' artificial life systems: the -universes , Alchemy [8,9], and Tierra . In these cases, the conclusion was as expected: it served to make more precise the a priori judgement that none of these should be considered as realising computational autopoiesis.
The surprise came when the test was applied to the original (re-implemented) model of computational autopoiesis [46,33]. Though experimental results were not presented in detail, it appears that the chain-based bond inhibition reaction, while essential to the self-repair of a single isolated cell, also has the unintended side effect of inhibiting maintenance of the bounding membrane when two membranes are adjacent to each other. This means that, if anything, adjacent agents tend positively to merge rather than to maintain their individuality. In turn, if the proposed heuristic test is accepted as an operational test for autopoietic individuation, we are forced to the somewhat controversial conclusion that the original minimal model, which was intended precisely to be an exemplar of the concept, is not in fact properly autopoietic.
However, to pre-empt mis-interpretation, I would emphasise that this result still does not argue for any obstacle in principle to computational autopoiesis. It merely demonstrates that robust autopoietic organisation is not easy or trivial to achieve.
Copyright © 2004 All Rights Reserved.