Words In Review: AI or 'stochastic parrots'?
You've probably heard chatbots like ChatGPT described as "artificial intelligence." Emily Bender, director of the University of Washington Computational Linguistics Laboratory, wants you to call it a “text synthesis machine” or “stochastic parrot.”
"Stochastic" means "randomly determined; having a random probability distribution or pattern that may be analyzed statistically but may not be predicted precisely." When you “parrot” something, you're repeating it without understanding. Calling chatbots "stochastic parrots" counteracts your mind's tendency to think these text responses come from a fellow mind.
Bender says that when you call it “artificial intelligence,” even though you’re saying it’s artificial, “it sounds much more reliable, sophisticated, independent, maybe even accountable than, ‘this algorithm or this automated process or this pile of linear algebra.’”
Bender compares ChatGPT to the old “Magic 8 Ball” toy from the 1970s. When you asked it a question and shook it up, it showed answers like “yes,” “no," and “ask me later.”
“So, you learned to only ask it yes-no questions,” Bender said. “If you said, ‘What should I have for lunch?’ and it said, ‘Yes,’ that wouldn't work. So you are bringing the frame, you are bringing the context, so that what comes back out makes sense.”
That doesn’t make it an intelligent machine.
“What's behind it is just a pile of equations. There's no emergent intelligence. There are no feelings in there.”
And be careful when the machine uses first-person, like, “I understand that …” There’s no “I” there and it does not understand. Bender is asking programmers to switch these tools to the third-person and describe themselves accurately.
“I think words are important here,” Bender said. “If we make a habit of talking about ‘large language models,’ ‘text synthesis machines,’ and ‘stochastic parrots’ instead of ‘artificial intelligence’ or ‘Sydney,' that can help deflate the AI hype.”
EDITOR'S NOTE: This article has been edited to more accurately define the term "stochastic."