As neural language models begin to change aspects of everyday life, they understandably attract criticism. This position paper was commissioned for a roundtable at Princeton University, dedicated to one of the most influential critiques: “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?” by Emily M. Bender, Timnit Gebru, Angelina McMillan-Major, and Margaret Mitchell.

My paper agrees that neural language models pose a variety of dangers, starting with and not limited to the list in “Stochastic Parrots.” But to understand those dangers, I think we need to look beyond the premise that these models mimic “language understanding” on an individual level. That may have been what linguists and computer scientists intended them to do. But the models’ actual potential (for both good and ill) is more interesting, and will be easier to grasp if we approach them as models of culture. Science-fictional scenarios about robots that become autonomous (or remain mere “parrots”) are less useful here than humanistic cultural theory.