Tackling Climate Change with Machine Learning — the Jevons paradox

By Fredrik Edin

Recently I wrote a blog post for “We don’t have time”, a social medium for climate change, where I reviewed and discussed a recent paper by Rolnick et al. on how machine learning can be applied to tackle climate change. The paper not only reviews the field but is a call for collaborations between stakeholders that are required for actions on climate change to be efficient. Associated with the paper is a platform, Climate Change AI, where one can partner up to try to work on climate change issues. I hope you read the post, or even better, the paper, and visit the platform.

In the end of the blog post, I mention the Jevons paradox, a counter-intuitive phenomenon whereby actions on climate change actually can lead to increases in climate gas emissions. Due to space limitations, the phenomenon could not be described in the original article. Instead I describe it here.

Coal-burning factories in 19th-century Manchester, England, taken from the Wikipedia article on the Jevons paradox.

The Jevons paradox

The Jevons paradox is an interesting phenomenon in economics that applies to climate change, and it is an illustration of the sometimes non-straight-forward nature of the area. The paradox was first described in 1865 by British economist William Stanley Jevons as an example of how more efficient steam engines led to increased coal consumption since although each engine consumed less coal, with better engines the market expanded and more engines were produced. Thus one cannot assume that a solution that lowers emissions in e.g. a factory will lead to an overall reduction in emissions. Of course, this should not be an excuse for inaction, but rather be a reminder of the importance of listening to advice from experts when trying to work on tackling climate change. Since this is well established in environmental research it may be that many of readers know about it already, but I try to illustrate it here as many readers who stumble upon this blog post may be non-experts.

To illustrate the paradox, we imagine a product where the cost of production basically follows emissions. If we assume that the company settles for a small profit of let’s say 2%, then you can see that the emissions per unit will follow the price. We imagine that if the product requires 1 unit of CO₂ emissions to produce, the cost will be 0.98 and the price will be 1.

In Fig. 1 below you see the fictive demand curve of this imaginary product. As the price goes down, demand goes up. Thus, if you are successful in improving production efficiency and thereby lower CO₂ emissions, this leads to an increase in demand.

Figure 1. Demand as a function of price. As price goes down, demand goes up. If the price is 100 demand will be 0, and from there demand increases until 100 when the price is 0. The blue dot shows that if the price is 90, demand will be 10.

Note that although this demand curve is fictive, all products have a demand curve and many follow the basic assumption encapsulated in my imaginary demand curve, namely that if the price goes down, demand will rise. So the conclusions drawn below will generalize to many situations where the demand curve does not look exactly like this one (but one counter-example will be “snob goods”, i.e. goods that follow a Veblen demand curve).

In Fig. 2 below you can see how the demand curve determines the effect of reducing the emissions per produced product on total emissions. This shape arises directly from the demand curve in Fig. 1 — it simply shows [demand given a price] * [emissions / unit at that price]. In the beginning of the technical development of a product, the cost of producing will be too high and there is no demand for the product. However, as processes improve, the price can be brought down below 100, and then there will be demand. If you enter the company at the blue dot and work to improve production efficiency, lowering the emissions / product, then you will only cause total emissions to increase. This will happen until the price reaches 50 (green diamond), which is the optimum price given the above demand curve. After this point, total revenue will start to decrease, and the company will not produce more even if the costs go down. Rather it will keep the price to increase its profit margin if it cannot change the demand curve, which it will want to do.

Figure 2. Total emissions = emissions/unit * units produced. This shape arises directly from the demand curve in Fig. 1. Blue square: emissions are 90 / unit and 10 products will be sold (see Fig. 1). Total emissions are then 90*10 = 900. Green diamond: point of maximal emissions. Emissions are 50 / unit and 50 products will be sold. Total emissions are then 50*50 = 2 500.

Thus, if we want to help fighting climate change, we should ensure that decreasing the CO₂ emissions per product does not lead to total increased demand. If demand is tightly linked to CO₂ emissions (e.g. when price is tightly linked to CO₂ emissions), then there is a risk for the Jevons paradox to occur.

Sources:

[1] Rolnick D., et al. Tackling Climate Change with Machine Learning, arXiv:1906.05433 [cs.CY]

About the author, Fredrik Edin

Fredrik Edin is a senior data scientist and co-founder of Codon Consulting and holds a PhD in computational neuroscience. Fredrik is working on deep learning and other types of machine learning and data science. In addition, he has dev-ops and cloud computing skills. Apart from machine learning, Fredrik also works in close collaboration with management to help companies establish their data architecture and team. He has previously worked in the financial services and big pharma sectors with quantitative analysis, business development and management.