Emily Bender

You Are Not a Parrot, Elizabeth Weil, New York Magazine, March 2023.

A profile of computational linguist Emily M. Bender that starts with an academic paper about a hypothetical octopus.

We go around assuming that speakers mean what they say and expect to live with the implications of their words. Philosopher Daniel Dennett calls this “the intentional stance.” Bender’s point is that we’ve built machines that mindlessly generate text, but we haven’t figured out how to stop imagining a mind behind it. That gap, between what the machine does and what we project onto it, is where the danger lives.

Bender challenges the use of the term “artificial intelligence” itself. “Intelligent” according to what definition? She’s fond of an alternative proposed by a former Italian parliamentarian: Systematic Approaches to Learning Algorithms and Machine Inferences. SALAMI. Then people would be asking, “Is this SALAMI intelligent? Does this SALAMI deserve human rights?” (The absurdity is the point.)

At a 2019 conference, Bender raised her hand and asked “What language are you working with?” for every paper that didn’t specify, even though everyone knew it was English. In linguistics, this is called a “face-threatening question,” meaning you’re being rude and risking lowering the status of both yourself and the person you’re addressing. “Always name your language” is now known as the Bender Rule. Carried inside the form of language is an intricate web of values, and defaulting to English without saying so erases that.

When people claim “well, humans are just stochastic parrots too,” they are so eager to believe language models are intelligent that they’ll devalue themselves as a point of reference to match what the model can do.

The meaning debate: Bender and Christopher Manning disagree on how meaning is created. The standard position (Bender’s): meaning requires referents, actual things and ideas in the world, like coconuts and heartbreak. This refers to that. Manning sees this as outdated and proposes “distributional semantics,” where the meaning of a word is simply a description of the contexts in which it appears.