Document: Modelling Autopoiesis: Harder Than It May Seem!

By a wonderful and utterly unpredictable chance, just as I contacted Francisco Varela again, to pass on my assessment that there was some crucial lacuna in the original published descriptions of this model, he received a shipment of papers from Santiago, dating from the general period when this work was carried out, and which he had thought to have been completely lost. Searching through these materials, he found a single hardcopy printout of a single version of the original simulation program, together with a short textual description of the model, which he believed to be at least roughly contemporary with the program listing! These materials are now permanently archived in (McMullin, 1997b).

In context, of course, this was the intellectual equivalent of striking gold. If there was ever to be any possibility of definitively resolving what was really going on in the original program, then access to that program`s code was the absolute sine qua non ; and here, against all rational expectations, was, precisely, a snapshot of that program code. Granted, a rather grainy snapshot, in black and white; but vastly more than I had expected to have available to me!

Notwithstanding all that, I hesitated long and hard before committing to a detailed examination of this code. It was written in the long obsolete language, FORTRAN-IV. It was only sparsely commented, and such comments as there were, were in Spanish. Variable names was generally less than four characters long and had little obvious mnemonic content. The program was structured as a single monolithic block of code, rather than using procedural decomposition (i.e. subroutines). There was no guarantee that this particular version of the program even worked. Nor could I assess its likely relationship to the version actually used to generate the published results. Trying to get it working would require manual re-keying, which is intrinsically error prone. H owever: while all these were legitimate causes for concern, I had really exhausted all other avenues of attack, so that it seemed I had little to lose by attempting to decipher this program.

My strategy was to first rekey the program and attempt to get it ``working'' in some minimal sense, without making any serious effort to understand its detailed functioning; only if I could achieve this, and if the program exhibited the crucial phenomenon (i.e. that premature bonding was somehow avoided) would I invest the effort to analyse the program's mechanisms in detail.

Somewhat to my surprise, this strategy worked amazingly well. Only a very small number of keying errors occurred, most of which were identifiable from compiler diagnostics. There was only one ``external'' fragment of code missing, and its function could be reasonably inferred from the code to hand (it was a pseudo random number generator) so that it was possible to design a functional replacement. This did result in a program that could be executed; and, against my more pessimistic expectations, it did reproduce a variety of characteristic phenomena of the original published results (encouraging me to believe that this was, at the very least, a closely related version); and, most crucially of all, it exhibited the avoidance of premature bonding that had plagued all my attempts at reproducing the full-blown autopoietic phenomena.

Now knowing that the answer to my question was surely concealed somewhere in this program, I threw myself into a sustained attempt to understand it. A key factor here was that it was not necessary to grasp the program in its entirity, but only the implementation of one particular interaction. Thus a relatively cursory examination of the program quickly allowed me to narrow my search down to one relatively small section--and to identify exactly what was built into the program that was preventing the premature bonding. This turned out to be a special inhibition effect that only came into play in certain special--but critical--circumstances. This effect had not been mentioned in any of the published descriptions of the model, but seemed to be responsible for the key difference between the original program and all my attempts are reproducing it.

So--finally--I was ready to make the critical, scientific , test: I incorporated the appropriate analog of this new effect into my own, otherwise entirely independent, implementation of the model. I had to experimentally adjust other parameters of the model to take account of this, but in rather less than half an hour (and after only five years of trying!) I finally saw for myself, the full blown autopoietic phenomena reported in the original 1974 paper!

5 Epilogue

Modelling Autopoiesis: Harder Than It May Seem!

3.2 The Problems

Document: Modelling Autopoiesis: Harder Than It May Seem!

mcmullin@eeng.dcu.ie

4 The Resolution