AI and Science: Opportunities and Pitfalls
The role of data in modern science is exploding due to the proliferation of high-throughput instruments and distributed sensors, as well as massive amounts of simulated data. Machine learning methods (both ‘classical’ and and ‘AI’) are already essential in interrogating and managing this datastream. It is hard to imagine that this role will become less important over time or even stagnate: Many opportunities and challenges make for an exciting future. At the same time, there is a question of whether AI methods can lead to sharper and deeper scientific insights — rather than merely mechanical improvements — from these large datasets. Here the story is much less clear and the lack of major breakthroughs is somewhat sobering. I will list some reasons why this is a hard problem.
Please log in to have access to the recording.