At the close of the twentieth century, human semantic memory was thought to be organized around dedicated, domain-specific modules operating independently of one another and of perception, action, and much of language. The intervening years have marked a transition toward a more holistic view in which semantics emerges from learned mappings among sensory, motor, and language representations, with recent work beginning to integrate semantic representation with theories of cognitive control and attention. This article reviews the reasons for this remarkable change in perspective, provides a snapshot of our current understanding of semantics in the mind and brain, and considers how this understanding may change in light of new innovations in artificial intelligence.