This Week in Green AI #3
DeepMind, ESG Book & New York Times
A weekly update on three major stories from the world of artificial intelligence and sustainability.
1. DeepMind releases paper citing environmental risks of AI
DeepMind released three major papers about their interdisciplinary research on language modeling. This included training models up to a size of 280 billion parameters — the largest of which was aptly named Gopher. While Gopher exceeded existing models in some tasks such as Massive Multitask Language Understanding (MMLU), it also revealed low improvements in logical reasoning and common sense tasks, hinting at diminishing returns to scale for certain ML tasks.
The second paper looked at ethical and social risks from large language models, including “automation, access and environmental harm”. Citing “the potential of LMs to create significant environmental costs”, the authors draw attention to the significant harm of operating LMs compared to just training them. They also mention the ambiguity of increasing efficiency to decrease environmental costs, which can lead to the adverse effect of ever larger models increasing the overall energy consumption (“Jevon’s Paradox). Many of these ethical and environmental risks will not be easy to solve for AI researchers, and will force companies like DeepMind to make trade-offs between performance and sustainability.
2. ESG data platform launched amidst concerns over corporate greenwashing
Last week, an alliance of financial institutions, investors and corporations launched ESG Book, an open data platform for sustainability information. Developed by asset manager Arabesque, the platform is supported by partners such as Allianz, HSBC, and the World Bank. ESG Book claims to follow the UN Global Compact and aims to “democratize” ESG data in order to accelerate the global sustainability transformation.
ESG finance has been called a “dangerous placebo that harms the public interest” by non other than Blackrock’s former head of sustainable investing. While some researchers have called for AI to be leveraged for more efficient extraction and interpretation of ESG-related data, other would question the very notion of ESG for the good of the planet. In The ESG Mirage, Bloomberg describes how these rating simply measure the impact of the climate crisis on the company, turning “the very notion of sustainable investing on its head”. The ESG book might be a genuine attempt of the financial services industry to democratize access to data, but for some such an initiative is based on a flawed premise from the very start.
3. The New York Times publishes dramatic climate S.O.S project powered by AI
A team of journalists at the New York Times have published a dramatic and spectacular collection of 193 stories about the real impact of climate change in every single country of the planet. They showcase how the climate crisis has already transformed the lives of millions of people, and tell a story about the “most existential issue facing the planet today”.
In many cases, the stories are enabled, enhanced, and powered through the use of AI-based tools and services. Citing just two examples, Climate Engine leverages Google’s geospatial AI to showcase the drought threatening Lake Sevan in Armenia, while deadly clashes between Kyrgyzstan and its neighbors over water shortages are explained using TerraMetrics ML-powered platform that detects changes in soil and microclimate. The New York Times project showcases how AI can be used for good to support essential journalism on the climate crisis and environmental degregation.