6 Conclusion

The substantive criticisms of (Varela et al. 1974) presented in this report can be summarised as follows:

The presented algorithm implements the model chemistry in a way which is obscure and counter-intuitive, particularly in relation to the bonding reaction.
The algorithm itself is unclear and ambiguous in places.
The presented experimental data are not consistent with the algorithm.
In fact, it seems likely that the experimental data actually derived from a program closely related to that of EXP29.FOR, and thus quite different in substantial detail from the presented algorithm.
By critical consideration of the experimental results and of the EXP29.FOR code, an important--perhaps crucial--constraint on the bonding reaction has been re-discovered: namely that bonding of free L particles be inhibited in the immediate vicinity of a chain of such particles. This interaction can now be clearly recognised as important to the intended autopoietic phenomena. However, this interaction has not, to my knowledge, been explicitly mentioned in any previous description of the model.

All these defects arise at least in part from the fact that the target paper dates from the very early days of this kind of biological modeling. Indeed, the idea of using this particular form of computational model should be seen as highly original for its time, and the authors deserve considerable credit for that. I regard the concept of autopoiesis, although still somewhat vague, to be a very useful one in clarifying the conditions for discussing the origin of life. To this extent, the concept, and its computational modeling or realisation, is potentially very significant for the development of the field of Artificial Life (Langton 1989).

The criticisms raised here should not be regarded as detracting from these genuine and important achievements.

Nonetheless: from the point of view of continuing or developing the particular model presented in the target paper, these criticisms represent real and serious obstacles, which need to be overcome. In particular, I make the following concluding points:

The discrepancies between the experimental runs and the given algorithm mean that the demonstration of basic autopoietic organisation in computational media claimed by the target paper must be regarded as weak.
There are published descriptions of independent implementations of autopoietic organisation in computational media (Zeleny 1977, Zelany & Pierre 1976). However, I have found the corresponding algorithm descriptions even more difficult to follow than the original presented in (Varela et al. 1974). The corresponding code was made available at the time, but I understand it is no longer so.[15]
As a general principle, the publication of adequate information to allow independent assessment and testing of any claims made is a crucial distinguishing feature of the scientific enterprise. At the time of original publication of (Varela et al. 1974) it was neither technically nor economically feasible to include full source code with the publication (in machine readable format). Therefore the authors opted to include a semi-formal algorithm. This was surely the best practise possible at the time. However, as we have seen, it suffers from serious practical deficiencies, and does not actually meet the desired standard of allowing independent assessment and testing. Critically, it seems, in this particular case, that there was an important interaction present in the code but not mentioned at all in the description; and this may not be all that exceptional an occurrence.
The technology is now available to augment the use of a published summary, or algorithm, of model code. At a basic level, the World Wide Web allows full source code to be made easily available to any interested researcher, in machine readable format. At a more advanced level, the emerging availability of standardised modeling environments, such as the Swarm simulation system[16], should significantly ease the burden involved in independent examination and assessment of such published models.
I suggest that it would thus be beneficial to re-implement the original algorithm--and various related (ideally, simpler) systems modeling the same qualitative chemistry--to establish how robust, or otherwise, the autopoietic phenomenology is in these systems; and it may be especially beneficial to do this using the Swarm simulation system, and make full source code of the implementation available via the World Wide Web. This is the focus of a continuing research effort, and is the subject of a planned companion report.

References

Computational Autopoiesis: The Original Algorithm

5 External Review

Document: Computational Autopoiesis: The Original Algorithm

mcmullin@eeng.dcu.ie