Dan Roberts, PhD

Dan Roberts is an AI Fellow at Sequoia and a research affiliate at the Center for Theoretical Physics at MIT and a principal researcher at Salesforce. He studies how tools and perspective from theoretical physics can be applied to artificial intelligence.

Dan arrived at Salesforce via acquisition of Diffeo, a big data startup that builds document writing tools that recommend novel content to users. Along with Hertz Fellows John Frank and Max Kleiman-Weiner, he co-founded Diffeo and served as chief technology officer. He was previously a postdoc at the Institute for Advanced Study in Princeton, New Jersey.

As a Hertz Fellow, Dan completed his PhD in high energy theoretical physics at MIT, where he focused on trying to come to terms with the nonsensical nature of quantum gravity via the holographic gauge/gravity duality conveniently known as holography (not the one with the lasers, which is inconveniently also known as holography). Dan’s interests are in CHAOS, Holography, And “Other Science.” Specifically, he’s interested in the relationship between black holes, chaos, and computation. (And, as it turns out, trees. A lot of this reduces to understanding trees.)

Dan graduated from Duke University, with majors in physics and electrical & computer engineering. As a Goldwater Scholar, Duke University Faculty Scholar, and Marshall Scholar, Dan conducted research under David R. Smith in the emerging field of transformation optics and metamaterials, in which coordinate transformations are used to create unconventional electromagnetic devices, such as invisibility cloaks. Following graduation, Dan used his Marshall Scholarship to study physics in England. In his first year at the University of Cambridge, Dan read for Part III Maths, focusing on the physics of high energy theory. In his second year, Dan acquired a darker habit, trading in his light blue gown of Cambridge for the rivalrous darker blue of Oxford (which is, through a complex chain of intercollegiate color dissemination involving Yale, a direct chromatic ancestor of the Duke blue of his undergrad days). While there, he picked up some more “maths” and worked on applying category theory to topological quantum computing.

Orthogonally, Dan has strong interest in machine learning and artificial intelligence. He has co-organized the Knowledge Base Acceleration (KBA) track at the NIST sponsored Text REtrieval Conference (TREC), and sometimes uses reinforcement learning to train an AI to compete in poker competitions.