Are Models Able to Forecast Collapse?

A discussion about "The Limits to Growth"

The Goddess Gaia engaged in system dynamics modeling. The model on the whiteboard is not exactly how a real SD model should be, but Grok drew a nice-looking Gaia, so it is fine.

How to Use Models Without Believing in Them

Ian Sutton - Reproduced by kind permission from the Blog “Net Zero”

Dec 01, 2025

At our Process Safety Report site we have just published the post All models are wrong, but some are useful. It turned out that, only one day earlier, my colleague Ugo Bardi also published a Substack post: Are we on the Edge of Collapse? Impressive Data from a Recalibration of World3.

Both posts provide interesting and useful insights as to the use of models when it comes to understanding complex systems such as climate change and resource depletion.

The Process Safety Forecast

A theme of the process safety post was that Large Language Models (LLMs) can provide valuable summaries of incident reports from organizations such as the Center for Chemical Process Safety (CCPS). However, LLMs may also smooth away contradictions and outliers, leading to hallucinations: confident but false statements generated when messy or missing data is ignored. Therefore, it is vital to pay careful attention to any raw, unfiltered data that does not fit neatly into a model such as the CCPS 20-element management framework. Examples are contradictory witness statements, anomalous temperature readings, and near-miss logs.

Doing so may provide unexpected insights.

Industrial Production Forecast

In his post, Bardi says,

I always say that models are not predictions; they are qualitative illustrations of what the future could be. But as the future gets closer to the present, models can start being seen as predictive tools. It is the weather/climate dichotomy, so aptly exploited to confuse matters by politically minded people in the discussion about climate. Right now, we are getting close to the point that we could forecast a collapse in the same way as we can forecast the trajectory of a tropical storm.

He discusses the projections of that remarkable and seminal report, Limits to Growth, published in the year 1972. (We have discussed that report in various posts at this site, including Limits and Beyond: No More Growth and Limits and Beyond: The Yawning Gap.) The authors of the report did not claim that it was exact, but that their model did capture system dynamics driven by feedbacks.

Bardi shows that that model, published so many years ago and well before the advent of modern computers, forecast many of our current difficulties with uncanny and rather frightening accuracy.

Bardi illustrates his point with updated predictions to do with Industrial Production, as shown in the following chart (based on a paper by Nebel et al.)

If this forecast is even close to being correct, we can anticipate industrial production declining by almost 50% just ten years from now.

Bardi goes on to say,

Do pay attention to the other curves of Nebel et al.’s paper. Agricultural collapse will be at about the same time as the industrial one. Population should start collapsing a few years later. Pollution will reach a peak around 2080 at levels some three times higher than the current ones. If this is a good prediction, we are in for a rough ride, a VERY rough ride.

Conclusions

As an engineer, I was taught to distinguish between precision and accuracy. For example, computer models predict the temperature profile in a distillation column with great precision. But, if we used incorrect vapor/liquid vapor pressures, the result could be badly inaccurate. As Large Language Models and other Artificial Intelligence tools become more prevalent, we are likely to see an increasing number of forecasts that are precise but inaccurate. It will be increasingly important to pay attention to outlying data that doesn’t seem to make sense because it does not fit the model.

Given that precaution, we need to accept that models can provide a useful forecast as to what the future may hold.

Finally, when it comes to climate change and other Age of Limits issues, we have to recognize that many models are predicting extreme social changes within the next few years. For example, if the forecast to do with industrial production is even close to being accurate then, as Bardi says, “We are in for a rough ride, a VERY rough ride”.

Models cannot save us from uncertainty, but they can warn us when the ground beneath us is shifting.