The advent of AI: an opportunity for the climate?

With their traditional sense of proportion, the markets have taken note of the gigantic potential of AI (Artificial Intelligence) and ML (Machine Learning) technologies. Even their most enthusiastic critics are struggling to establish the limits of what it will be possible to do with these tools.

Every economic agent is now called upon to consider the impact that AI will have on their business. But the question also arises in the area of climate change, especially as there are real differences of opinion at the moment. At this stage, some see AI as more of a threat than an opportunity.

In our view, the impact of AI on the climate will be seen in, at least, 3 dimensions:

  1. Firstly, there is the direct carbon footprint of AI and ML themselves, taking into account the entire value chain (from the manufacture of the hardware to the electricity consumption required for calculations);
  2. Secondly, each AI application will have its own effects (direct or indirect), positive or negative, on the climate;
  3. Finally, the systemic changes in our lifestyles brought about by these technologies could restrict our ability to achieve our climate objectives.

Assessing the impact of each of these three issues is particularly complex.

AI’s carbon footprint: difficult to estimate, but growing

At present, the carbon footprint of many activities can be estimated. However, for AI/ML we still lack tangible data that would enable a reliable estimate. Because of its relative “youth”, a lack of transparency (most companies protect data concerning the quantity and nature of the calculations they carry out) and the absence of standards, it is still risky to put a figure on the carbon impact of these new technologies.

According to the main studies, the IT sector accounts for 2 to 3% of global emissions. Counter-intuitively, this percentage has remained fairly stable over the past decade, with progress in energy efficiency linked in particular to Moore’s Law ‘offsetting’ ever-increasing use.

AI and ML represent only a fraction of this percentage, which is still difficult to quantify. The only certainty is that, with their now exponential deployment, they will have an increasing weight in total emissions, particularly in countries where the electricity mix remains highly carbon-intensive.

Short-term applications of AI: not necessarily without risks

Many scientists are already working on AI applications in areas such as ecology, climate research and sustainability. In some applications, AI promises to make a positive contribution. In others, AI could do harm or, at the very least, disappoint. Here are 3 examples from a growing list of applications.

In the energy sector, AI should considerably improve forecasting models for both electricity demand and production. At a time when the increasing integration of intermittent renewable production capacities is making electricity grids more difficult to control, AI algorithms will play a considerable role in the management and optimisation of electricity grids. AI will play a positive role here.

In the same way that it will help to optimise the operation of an electricity network, AI will also help oil companies to analyse seismic data more effectively and thus increase their production of hydrocarbons. For this application, the consequences for the climate could be very negative.

When it comes to modelling the climate and assessing the changes underway, AI could prove disappointing. AI climate models analyse historical data to produce their forecasts. However, current climatic phenomena seem to have no connection with the past, either in terms of their frequency or their severity. In this context, AI forecasts could turn out to be mediocre and fail to alert us before major disasters.

In the longer term, the AI revolution will be what we make of it

Let’s take just one example: the automotive sector. The rise of AI/ML brings us closer every day to the arrival of autonomous vehicles. This is a perfect example of the potential revolutions brought about by these technologies, but also an illustration of the ambiguity of their contribution to the climate. Will autonomous vehicles be the missing link in meeting the challenge of multimodal mobility? Or will they simply be a new development in a system based on individual vehicle ownership?

A blockbuster this summer delivered a Hollywood version of Robert Oppenheimer’s story. Nuclear power is the historic and probably unsurpassable example of the abysmal gap between ‘good’ and ‘bad’ use of a technology. We have no doubt that AI has the potential to cause a real ecological disaster. We therefore believe that it is vital to provide a framework for the development of this young industry, through constraint and regulation if necessary, in order to bring it into line with our ambitions in the fight against global warming.