Emily Bender maps four ways AI research dehumanizes people

Emily M. Bender, a professor of linguistics at the University of Washington, uses a CogSci 2022 keynote to lay out several ways she says AI research and product development treat people as less than human. The keynote was delivered in Toronto on July 29, 2022, and later published as a journal article in Current Directions in Psychological Science in 2024.

Her framework names specific mechanisms. There is the 'computational metaphor,' which she describes as framing brains as computers and comparing child language acquisition to machine learning. There is 'digital physiognomy,' computer-vision systems used to classify people by appearance, predicting traits like criminality or sexual orientation from physical characteristics. There is 'ghost work,' the hidden human labor in data labeling and content moderation that gets treated as interchangeable machine components. And there is the reinforcement of what she calls a white racial frame in design choices, workforce demographics, and English-centric datasets.

Bender argues humans are 'thoroughly relational,' shaped by historical, cultural, and social contexts. Machines that extract patterns from language form alone, she says, cannot reach that. Her broader claim is that cognitive scientists and the humanities have a responsibility to challenge inaccurate capability claims that treat artificial intelligence as a settled category rather than a marketing metaphor.