AI4Science #5 in Berlin

AI4Science #5 in Berlin

AI4Science #5 is in the books. Thank you to everyone who joined us at Merantix AI Campus during Berlin Science Week.

Our co-founder Nasim Rahaman moderated the evening in two halves: when AI thinks, and when AI acts.

The first panel asked what is left for scientists once machines can come up with ideas and check the results. One candidate is intention: deciding which problem to solve. Patrik Reizinger noted that checking isn't solved either: a model once gave Patrik a valid Lean proof of a statement Patrik never meant to prove. Ksenia Briling reminded us that the race is partly our choice: nobody forces us to build machines that push the field toward ten million NeurIPS submissions. And what counts as understanding when a model predicts what no one can explain? Iyad Rahwan raised the prospect of knowledge that machines can discover but cannot pass on to us. Yoshua Bengio was more hopeful: chemistry is an explainable approximation of physics, biology of chemistry, and AI can help us keep digging.

The second panel turned to AI that acts: labs that run their own experiments, and whether to let them rip. Annalena Kofler warned that a lab with every surprise designed out stops discovering. Vincent Stimper drew a hard line on risks, e.g. in areas such as bio safety, which have to be avoided at all costs, and there are plenty of low risk applications, such as in silico screening, where AI can make a big difference instead. Alexander Hammer pointed out that chemistry already has the discipline AI needs: no experiment without a risk assessment first.

The problem isn't only the machines. Roman engineers, Iyad told us, had to sleep under the bridges they built. Machines let us skip that: in Iyad's research, the share of people who cheat can rise from 5% to as much as 85% when they can hand the task to a machine. Yoshua's answer is to train AI to be honest rather than to chase reward, and, when the choice looks like six months or never, to consider five years.

Near the end, Annalena made the point that stayed with us: you could have asked an agent what the panel thinks, but you came to hear people argue it out in person. That is why we keep running these evenings.

Thank you to Yoshua, Iyad, Annalena, Ksenia, Vincent, Alexander and Patrik for making it such an interesting evening, and to Sif, Arantxa and others at Merantix AI Campus for the teamwork in pulling this event together.

Photo from AI4Science #5 at Merantix AI Campus in Berlin

Photo from AI4Science #5 at Merantix AI Campus in Berlin

Photo from AI4Science #5 at Merantix AI Campus in Berlin

Photo from AI4Science #5 at Merantix AI Campus in Berlin

First posted on LinkedIn, November 10, 2026.