Annual Report 2025 - Report - Page 35
High-profile panel on AI in Chemistry: Moderator Derek Muller, Francisco Martins, Michael Levitt, Animashree Anandkumar,
Joachim Frank, Danielle Belgrave, and John M. Jumper
it really means in physical terms.” His remarks underlined a central theme of the discussion – that AI’s value
depends both on its output, and on scientists’ ability to
question and validate it.
Danielle Belgrave, Vice President of AI and Machine
Learning at GSK and 2011 Lindau Alumna, echoed this
view. In drug development, she said, AI is not a simple
pipeline but part of an iterative process: “We see a lot
of heterogeneity in patient responses. It’s very much a
measurement science: understanding what to measure
and how to capture that variability.” Her insights illustrated how AI can enhance understanding rather than replace expertise, helping researchers manage the growing
complexity of biomedical data.
Large language models also entered the discussion.
Levitt noted that he had asked them around 50,000 questions, finding them “sometimes really stupid and sometimes incredibly clever.” Used wisely, they can be valuable
for administrative or preparatory work in science, such as
drafting documentation or patent applications. However,
he and others warned of their limitations. Frank voiced
concerns about AI systems that both generate and analyse data, arguing that such circular use could distort
results. Jumper also underlined the need for caution: “An
AI tool where the user misunderstands the reliability of
the tool can be really challenging. It’s better for wrong
answers to look really wrong than for all answers to look
plausible but some to be very wrong.”
Throughout the discussion, it became evident that AI
holds extraordinary potential. Young Scientist Francisco
Martins stressed AI’s ability in “helping to put together
ideas to understand data”, resulting in faster and more
logical ways to formulate new structures. The panelists
did however agree that the power of AI must remain anchored in human judgement and scientific principles. It
can speed up discovery, reveal hidden patterns, and handle vast datasets, but it cannot replace critical thinking or
accountability.
Looking ahead, the panelists highlighted a shared
conviction: AI is a transformative scientific instrument,
but it is only as reliable as the expertise guiding it. When
used thoughtfully – with transparency, responsibility,
and an awareness of its limitations – artificial intelligence can become both a computational aid, and a catalyst for discovery across chemistry and beyond.
33
