GenAI solutions for biodiversity. This is the name of a recent hackathon in Spain where technologists and scientists tried to design innovative algorithms that help identify and contain the human impact on nature. Winners receive an economic incentive to develop the models, along with a mentorship from Accenture and Microsoft as well as access to the latter’s generative models (a striking contradiction to begin with, sustainability-wise, one might argue).

The initiative is part of the Spanish Ministry for Digital Transformation and Civil Service’s National Green Algorithms Program which includes an investment volume of 257 million euros for sustainable AI projects. A similar initiative took place earlier this year at the AI Action Summit in Paris. The “Frugal AI” challenge, organized by the French Ministry of Ecological Transition, the Data For Good association and Hugging Face, aimed at incentivizing the development of high-performance models with minimal energy consumption. The three projects that were selected are designed to detect climate disinformation, regions at risk of wildfires and illegal deforestation.

Terminology

There is a curious ambiguity in the use of the term “green.” “Green” algorithms can be understood as systems that are designed to be sustainable, i.e. algorithms that do not contribute to heavy carbon emissions and aggravate the climate crisis. However, many of the initiatives that use the label are geared towards climate measures or developing environmentally friendly projects with the help of automation.

AlgorithmWatch’s SustAIn project addressed this issue as early as 2021 by laying out guidelines for developers. It also recommended not to focus on the sustainability of the models alone, but also on their infrastructure, such as data centers, hardware and energy sources. In line with Hugging Face’s Sasha Luccioni, Signal’s Meredith Whittaker or quantum physicist and neuroscientist Gaël Varoquaux, the project challenged the dominant “bigger is better” mentality in AI that equals larger models and more parameters with better performance, which is not necessarily correct and definitely not sustainable.

Sustainability of tech or via tech

Chris Adams, executive director of the Green Web Foundation, points out that most of such “green” programs focus on technology for sustainability as their output is more marketable: “If […] you’re a profit-making company, you’re incentivized to double the thing that is profitable rather than the one that is not profitable.” This would also be the reason why oil and gas companies invest heavily in AI to speed up fossil fuel extraction, and not in making renewable energy adoption easier, he told me.

In a recent post, Adams laments “a huge gap in terms of the data required for a meaningful data-informed discussion about AI.” This information gap is prevalent in the entire field of AI. Developers and providers are not too keen on sharing data on their systems' energy consumption. They still often claim that the obligation of measuring the environmental impacts of AI systems would be too complicated and too great a burden. But one of SustAIn’s key points was that easy-to-use measurement methods already exist for monitoring energy or water consumption and emissions.

Who keeps track?

Government-funded initiatives will often involve big tech companies and consultancies – think McKinsey, Accenture or Deloitte – to explore new technology’s potential, but rarely follow through on monitoring actual implementation or real impact. Sustainability specialist Vlad Coroamă refers to this as “Chronic Potentialitis.” He exposes how cheap predictions about positive environmental impacts of new digital applications are as long as validation is not happening.

Policymakers invest in GenAI projects for sustainability, while GenAI generates more emissions than other AI systems. I asked the Spanish ministry why they decided to specifically fund GenAI (opposed to other potentially more environmentally-friendly technologies), and how they planned to monitor the environmental footprint of the awarded projects. They declined to comment.

This is an excerpt from the Automated Society newsletter, a bi-weekly round up of news in automated decision-making in Europe. Subscribe here.