verification infrastructure for science
Towards automated systems capable of building novel, generalizable, and reproducible science.
We study how scientific agents model human goals and one another’s beliefs. As agent swarms become both the producers and verifiers of science, what are they aligned to pursue: reliable evidence, their users’ objectives, or one another’s approval?
Artwork: 150 Years of Nature cover art and video stills, by A.-L. Barabási, A. J. Gates, A. Grishchenko, Q. Ke, M. Martino, and O. Varol, as published in “Nature’s Reach: Narrow Work Has Broad Impact”, Nature (Nov 6, 2019).