Alicja Ostrowska defends her PhD thesis, “Life and AI at NASA: An Ethnography of How Scientists and Engineers Make Tools to Explore Other Worlds” on 30 January at Chalmers University of Technology.
Abstract
Artificial intelligence (AI) is increasingly introduced into scientific practices, including NASA’s missions that explore conditions for life and habitability on other planets and moons. How does the development of new AI tools within these missions transform scientific knowledge production?
Drawing on theories from Science and Technology Studies (STS), this dissertation analyzes science as a cultural practice. It is based on ethnographic research conducted at NASA and within the wider community of planetary scientists and astrobiologists, including interviews and documentary materials.
The dissertation demonstrates how efforts to realize visions of autonomous science beyond Earth already reshape the everyday work of scientists on the ground. It shows how AI is shaped by organizational structures, knowledge infrastructures, and scientific cultures at NASA, while simultaneously feeding back into these dimensions. Boundary work to sustain the legitimacy of planetary missions influences the purposes for which AI can be developed – to identify organic molecules, to explore habitability and potential biosignatures.
The study further shows how field sites, laboratories, and national databases together constitute a knowledge infrastructure that shapes AI by determining which data are available for training. Choices of field sites are influenced by accessibility and symbolic value, rendering some places more popular than others, which skews knowledge production. Digital databases and AI training datasets serve as libraries of knowns against which the unknown is identified. Decisions about anomalies, artifacts, and novelty in data are central to both AI design and scientific discovery. The study highlights the limits of performance metrics and the importance of negotiations with domain experts, particularly in the emerging use of synthetic data.
Although AI remains at an early stage of development in the cases studied, it already reshapes power relations in scientific knowledge production by introducing new ideals of epistemic order and altering who determines the value and usability of data.
By providing an empirical account of AI development in one of the most impactful scientific institutions, this dissertation contributes to discussions about data-driven solutions in science, and the epistemic consequences of using AI in science on Earth and beyond.
Supervisor
Francis Lee, main supervisor
Shai Mulinari, co-supervisor
Opponent
Marisa Cohn, IT University of Copenhagen, Denmark
