This Week in Green AI #2
European Union AI Act, DAIR & Council of Europe
A weekly update on three major stories from the world of artificial intelligence and sustainability.
1. Slovenian Presidency shares Green AI proposal for EU AI Act
This week, the Slovenian Presidency of the Council of the European Union has circulated a compromise text on the proposed AI Act. One of the main changes to the Commission’s proposal includes the addition of environmental protection to the high-risk areas, which carry “specific obligations in terms of health, safety and fundamental rights”. The inclusion of digital infrastructure meant to protect the environment, namely
AI systems intended to be used to control emissions and pollution
represents an important addition to the high-risk areas of the EU AI Act. It reflects the Slovenian Presidency’s conviction that the trustworthy and ethical use of AI will play a significant role in combating climate change.
2. Timnit Gebru launches independent AI research institute DAIR
Google’s former AI ethics co-lead has initiated the Distributed Artificial Intelligence Research Institute (DAIR), which will focus on the harms of AI on marginalized groups. The announcement comes a year after Dr. Gebru was fired from Google for a research paper, which identified a variety of environmental costs and risks associated with ever larger LM models:
It is past time for researchers to prioritize energy efficiency and cost to reduce negative environmental impact and inequitable access to resources — both of which disproportionately affect people who are already in marginalized positions.
Green AI will surely be one of the focus area of DAIR: one of its two advisory committee members is Ciira wa Maina, whose research focuses on biodiversity assessment and environmental monitoring. Google has since published its own paper (co-authored by its chief AI scientist Jeff Dean) on the carbon impact of NLP, arguing that
The choice of DNN, datacenter, and processor can reduce the carbon footprint up to ~100–1000X.
and that misunderstandings about the full model lifecycle can lead to wrong CO2e estimates. Nevertheless, the paper also recommends more researchers to measure and report energy usage and CO2e for computationally intensive projects, so an independent research institute like DAIR can hopefully increase transparency and inclusion in the field of Green AI.
3. The Council of Europe recommends the “World’s First AI Treaty”
The Ad hoc Committee on Artificial Intelligence (CAHAI), instructed by the Council of Europe, has put forward its first recommendations on a legally binding treaty on AI to promote democracy, human rights, and the rule of law. Once ratified, this treaty would represent the first multilateral treaty on artificial intelligence. CAHAI has been supported by the 47 members, which include Russia and Turkey, as well as the United States, Canada, Japan, and Mexico.
While the full text of the recommendations is not yet public, it is hoped that CAHAI will include Green AI in the treaty. In its publication of December 2020, it had marked a “thematic underrepresentation of sustainability”, with “massive computational resources requiring “high energy consumption”. However, the report also noted the “possibility of harnessing AI for the benefits of ecosystems and the entire bioshphere”. The world’s first AI treaty certainly has the potential to rectify this underrepresentation and put Green AI on the global agenda.
Please feel free to reach out to me on LinkedIn or Twitter to exchange ideas and insights about Green AI.
