Artificial intelligence, a new engine for scientific research

Will the next Nobel Prize be an artificial intelligence (AI), a machine that, after swallowing all the knowledge in the world, will have found a new cancer drug, or a physical theory beyond quantum physics, or demonstrated an unsolved math conjecture? We are not there yet, but AI is invading laboratories at high speed to improve instruments, speed up calculations, point to fruitful hypotheses, etc.

At the beginning of October, a Chinese team digitally improved the resolution of optical microscopy images in biology by ten times, making the shots sharper. One more example of the effervescence of the last few months. In May, a Facebook team had mathematical theorems demonstrated by an AI. The following month, competitor Google presented software that solved a third of 200 undergraduate level scientific problems in mathematics, physics, economics, biology… The same month, a Franco-German team entrusted an AI with the task of controlling a quantum object with well dosed microwaves to preserve its properties as long as possible (theoretical success which will be tried experimentally).

And also in our file: Article reserved for our subscribers Artificial intelligence: the many faces of deep learning

And, this summer, the best-known scientific AI, AlphaFold, released in 2021 by DeepMind, a Google subsidiary, was talked about again. The algorithm, which predicts the three-dimensional shape of proteins from their chemical formula, added a million configurations to the reference database, which contained “only” 200,000 experimentally determined structures.

Ever-increasing amounts of data

Every day, the list of AI applications grows in all areas of science. “It’s going all over the place! In 2016, a database on AI applications in particle physics had about ten articles; now it is growing by ten a month.notes David Rousseau, from the Irène-Joliot-Curie Two Infinite Physics Laboratory, in Orsay, co-author, in 2022, of the book Artificial Intelligence for High Energy Physics (World Scientific, untranslated). AI didn’t find the Higgs boson in 2012, but it will surely help make the following discoveries. Because, driven by the ever-increasing quantities of data to be processed, it will be used everywhere to sort out collisions, to simulate to compare theory and experience, and even to control trajectories within the particle accelerator.

You have 84.4% of this article left to read. The following is for subscribers only.

Leave a Comment