Ray Solomonoff (1926-2009) was an American mathematician and one of the small group present for the whole of the 1956 Dartmouth Summer Research Project on Artificial Intelligence, the meeting that named the field. Where most early AI work focused on symbolic reasoning, Solomonoff set out to formalize learning and prediction. He circulated a report on “An Inductive Inference Machine” at Dartmouth and went on to invent what he called algorithmic probability.
His theory, set out in papers around 1960 and 1964, asks how a learner should predict the next item in a sequence given only the data so far. Solomonoff’s answer weighted every possible explanation by the length of the shortest program that could produce the data, favouring simpler explanations - a formal, computational version of Occam’s razor. This work, developed in parallel with Kolmogorov, helped found the field now known as algorithmic information theory or Kolmogorov complexity, as recounted in the obituary by his collaborator Paul Vitanyi.
Solomonoff never became a household name, and his theory was uncomputable in its pure form, so it could not be run directly. But it gave machine learning a rigorous account of what it means to generalize from data, and it remains a touchstone for theorists of induction, prediction, and the limits of what learning systems can do. He kept working on algorithmic probability until his death in 2009.