Breathing and ‘toys’: these two humanities research projects receive an ERC Grant

Our planet suffers from human-caused crises: how can we expose both the environmental and the social impact of these challenges? And what are the pros and cons of using AI in creating new (scientific) knowledge? The projects seeking answers to these questions, led by Magdalena Górska and Emily Sullivan respectively, have been awarded an ERC Starting Grant by the European Research Council.

The link between environmental issues and social inequality

In her project RESPIRE: Planetary Breathing in Asphyxiating Times, Magdalena Górska introduces the idea of ‘planetary breathing’. “You could say your breathing is never simply ‘your own’,” she explains. “With every breath, we share the air with, for example, other humans, animals, plants, microbes, bodies of water, and soil.” To her, planetary breathing is not a metaphor. “We all are co-respirators, as everything is literally connected through the air, and therefore through breathing.”

But although we all need air, there are significant differences in its quality, Górska says. She argues that this inequality in planetary breathing is related to social processes, such as racialisation, migration, classism, and geopolitical power relations. In RESPIRE, she will work with natural scientists, organisations, and artists to develop this new field of scholarly research.

Environmental problems go hand in hand with social issues

To develop a new, comprehensive scientific approach, the RESPIRE team will focus on three planetary ‘lungs’ and the social and environmental problems they suffer: forests (deforestation), oceans (dead zones), and soil (peatlands). “Traditionally, environmental subjects like these were mainly studied by natural scientists,” Górska says. “Now, in RESPIRE, we are connecting these environmental problems with their social dimensions, because environmental and social toxicity go hand in hand.”

“Traditionally, environmental subjects like these were mainly studied by natural scientists.”

According to Górska, the added value of the humanities lies in the ability to develop a conceptual understanding of the world we live in. In RESPIRE, she also aims to utilise interdisciplinary and intersectional feminist perspectives. “These can challenge norms and expose social power dynamics,” she explains. “We will combine the different approaches with insights from natural sciences to understand how environmental issues and social problems are deeply intertwined, and to develop new ways of understanding and thinking about planetary breathing and suffocation.”

As part of the project, Górska will set up research hub RESPIRATORIUM. Here, she will bring together scholars, artists, and activists whose work addresses the challenges she and her team are studying.

The impact of AI on science and society

Ever more scientific fields are turning to artificial intelligence (AI) in the hope of solving problems or making discoveries. “In medical science, for example, AlphaFold recently made headlines by solving a longstanding problem in predicting protein structures,” Emily Sullivan illustrates. “This discovery could boost research in fields like antibiotics, cancer therapy, and resilient crops.”

At the same time, AI is also used across society to inform and provide knowledge. “The Dutch tax authority used AI to detect welfare fraud, but this model turned out to be discriminatory and caused the toeslagenschandaal. And large language models, or LLM, have disrupted society overnight, with millions relying on ChatGPT for information despite its known ‘hallucinations’.”

Deep learning models as ‘toy models’

“Expectations surrounding AI are high,” Sullivan says. “So we urgently need to evaluate its potentials – and dangers – across science as well as society.” In her project Machine Learning in Science and Society: A Dangerous Toy? (TOY), Sullivan focusses on one certain type of AI: deep learning (DL). “DL involves computer systems that are trained to identify patterns and to make decisions. Our project will help understand the promises and limitations of DL for knowledge production.”

“We urgently need to evaluate its potentials – and dangers – across science as well as society.”

The project team will examine DL models as if they are toy models, highly idealised representations that simplify and distort complex real-world processes. “Every scientific domain has their own that they use to ‘play around’ with and to gain insight into these complex phenomena,” Sullivan explains. “TOY will reveal the nature and value of toy models and idealisation, and we will better understand when the use of simplified AI models is beneficial and when it poses risks.”

“We hope that treating DL models as toy models will help us better understand both their surprising successes – like AlphaFold – and their frequent failures – like the AI models that should diagnose COVID-19,” Sullivan concludes. “And, importantly, we also explore how we might address these failures.”