Thanks to Raphael Arar’s suggestion, I recently read Luke Munn and Liam Magee’s “Other Worlds: Using AI to Revisit Cybersyn and Rethink Economic Futures” paper (see it here). Their approach is certainly innovative and interesting. However, I couldn’t help but think of two alternative paths that could be derived from their paper.

On the one hand, I wonder about the potential of AI to revisit Project Cybersyn but in a quantitative way. The authors used fine-tuning to develop chatbots of former Chilean president Salvador Allende and British cybernetician Stafford Beer (whom they consider the architects of the project), and from dialogues with these chatbots they speculate about a historical path for Chile’s socialism in which Project Cybersyn succeeded.

My question, however, is whether – instead of using the so to speak semantic space offered by these chatbots (the authors give examples of the dialogues they had with them) – it would be interesting to use AI to reconstruct the data that Project Cybersyn was supposed to use.

Here I’m thinking in particular of the economic simulator that Beer simply called Futuro, but that the rest of the team called CHECO. Below is a picture of one of the first Futuro simulations ran in Chile (from September 1972)

DYNAMO plotted output of the “Simulation Model of the Chilean Economy.” Source: CHECO team 1972, Stafford Beer Archive, Liverpool John Moores University Special Collections & Archives.

The graph shows a simulation of the Chilean economy out to 10 years with a focus on inflation. The simulation suggests that the efforts to boost production work until year 6. After that, prices skyrocket and everything becomes a mess.

Medina suggests that the main problem with Futuro was the lack of adequate data to run successful simulations. My own research suggests that this was not the case, and that Futuro’s failure was most likely due to internal frictions between government agencies. You can read more about this in chapter five of my dissertation (here the link).

In any case, I couldn’t find the actual datasets they used, but only written confirmations of what they used. So wouldn’t it be interesting to use AI to reconstruct the economic data of that time, and then rebuild Futuro to run different, more robust simulations? Munn and Magee’s “socialist infrastructuring” approach suggests that this is entirely feasible.

On the other hand, Munn and Magee’s paper made me think of Leif Weatherby’s “ChatGPT is an Ideology Machine” piece for Jacobin (read it here).

Munn and Magee acknowledged that despite feeding their model with Allende’s political discourse and Beer’s work on management cybernetics, their simulations seemed co-opted by the language of capitalist economics – probably, they suggest, due to the model’s training data.

The discussion section of this paper seems to suggest that their experiments were victims of what Bender et al. call “stochastic parrots” – in this case, capitalistic stochastic parrots – or the nascent machinic ideology that Weatherby describes. I wonder, however, if this is not also due to the limitations of the fine-tuning process in general, or the modest amount of data used for this purpose in Munn and Magee’s experiments.

Two assumptions by the authors may have also played a role here. First, they consider Allende to be one of the architects of Project Cybersyn. This is simply not the case. Allende was, of course, the articulator of Chile’s socialism, but his connection with Project Cybersyn was always distant, to say the least. His political ideas did influence the development of the project, but they were always mediated through Beer’s management cybernetics – Medina clearly established this in 2011.

Here is the second point. Most revisits of Project Cybersyn tend to take the role of cybernetics for granted – they offer brief summaries and attempt to explain the scope of the project only through socio-technical, cultural-theoretical, or political analyses. On the contrary, cybernetics was the core of the project, and any effort to grasp its potential requires tracing and problematizing the cybernetic thinking that informed it.

Therefore, I would argue that if there is a way to counter the machinic ideology of which Munn and Magee’s experiments would have been victims, it would be to feed (fine-tune) these LLMs with all the cybernetic thinking (socialist and otherwise) that set Project Cyberyn in motion. After all, to paraphrase Beer himself, only variety defeats variety.