Toward a Foundation Model of Human Behavior

How might deep learning offer a modern answer to old questions about understanding the human mind?

We explore three fields, human-computer interaction, simulation, and recommendation, that predict, simulate, and model humans for a wide range of tasks, and argue that they are converging on the same model: a foundation model of human behavior. Two of the three ingredients that carried deep learning elsewhere, a generic objective and scaling, are already in place. The third, a corpus of what people do that spans the silos holding it, is not.