Veni grant for research on how AI is transforming science
Artificial Intelligence is changing many fields of science. The project ‘Beyond “Ground Truth”: An Anthropology of AI at the Limits of Scientific Knowledge’, by anthropologist Rodrigo Ochigame, explores how researchers are redefining what counts as scientific evidence with new AI methods for deduction, observation and prediction. With the Veni grant, Ochigame aims to help scientists think more clearly and critically about the ongoing transformations in their fields.
The project investigates how researchers are redefining what counts as acceptable evidence in three main fields: pure mathematics, particle physics and cosmology, and climate science. Ochigame explains: ‘This question is important not just for scholars of anthropology and science and technology studies like myself. It matters for scientists in many fields, who urgently need to know which AI methods to trust, for what purposes and why.’
Deduction in mathematics
In mathematics, AI is enabling novel forms of deduction. Mathematicians increasingly try to discover and prove theorems with the help of AI techniques. But what if a computer-generated sequence of deductions is so long and complex that no one can understand it? Should it be accepted as a ‘proof’? Some mathematicians accept computer-assisted demonstrations as proofs because computers supposedly don’t make mistakes when checking them. Others disagree and say that such a result is pointless, arguing that research should help us understand mathematical ideas better, not just produce more proofs in a mechanical way.
Observation in physics
In physics, AI methods such as machine learning are raising controversies about what counts as observational evidence. Physicists debate the acceptable uses of AI in their research, for example in searching for new particles or in imaging black holes. A key issue is that AI-assisted results aren’t necessarily ‘direct’ observations that simply happen to agree with theory. Rather, the results are often produced by algorithms that already encode some theoretical assumptions.
Prediction in climate science
In climate science, Ochigame is interested in how researchers make difficult predictions. When building computational models to predict the future of complex ecosystems like Amazonia, scientists must consider ‘extreme’ scenarios that have never happened before. So these scientists cannot merely extrapolate from patterns in an existing dataset. In the terminology of AI researchers, there is no ‘ground truth’ data to rely on. This absence or limited availability of ‘ground truth’ data is a major concern in all of Ochigame’s cases.
Since deduction, observation, and prediction are fundamental to diverse fields of science, Ochigame hopes to contribute to a critical understanding of how AI is transforming scientific knowledge production more generally.