He said the basic research unit was not affordable and needed to be downsized. He was quite concerned about IBM’s software business, mainframe business, and midrange products. As I look back at my notes, it is clear he understood most, if not all, of the business issues we tackled over the ensuing years. What’s striking from my notes is the absence of any mention of culture, teamwork, customers, or leadership—the elements that turned

out to be the toughest challenges at IBM…

I went home with a deepening sense of fear. Could I pull this off? Who was going to help me?

Within two days, Bender had sent Gebru an outline. They later came up with a title, adding a cheeky emoji for emphasis: “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?”

Gebru assembled a research team for the paper within Google, including her colead Mitchell. In response to the encouraging words in Dean’s annual review, she flagged the paper as an example of the work she was pursuing. “Definitely not my area of expertise,” Dean said, “but would definitely learn from reading it.”

In total, it presented four key warnings: First, large language models were growing so vast that they were generating an enormous environmental footprint, as found in Strubell’s paper.

Second, the demand for data was growing so vast that companies were scraping whatever they could find on the internet, inadvertently capturing more toxic and abusive language as well as subtler racist and sexist references.

Third, because such vast datasets were difficult to audit and scrutinize, it was extremely challenging to verify what was actually in them, making it harder to eradicate toxicity or more broadly ensure that they reflected evolving social norms and values. Finally, the model outputs were getting so good that people could easily mistake its statistically calculated outputs as language with real meaning and intent. This would make people prone not only to believing the text to be factual information but also to consider the model a competent adviser, a trustworthy confidant, and perhaps even something sentient.

On the Thursday a week before Thanksgiving, after Gebru had submitted the paper to the conference, she received a calendar invite without explanation to meet Megan Kacholia, Google Research’s VP of engineering, over a video call less than three hours later. The meeting lasted only thirty minutes, and Kacholia cut to the point: Gebru needed to retract the paper.

The request was a dramatic aberration from the way Google and the rest of the industry handled research. Like many labs at other companies, Google Brain had until then largely conducted itself as an academic operation and given researchers wide latitude to pursue the questions they wanted to. At times, the company reviewed papers to ensure they didn’t expose sensitive IP or customer data. But researchers like Gebru had never known the company to block or retract a paper simply for shedding light on inconvenient truths. That Google was even willing to pull this move, some researchers would later reflect, was not only because of the new competitive pressure from OpenAI but also because of the work OpenAI had done to legitimize withholding research after GPT-2.

A few hours later, I published a story for MIT Technology Review with the first detailed account of the paper’s contents. The signatories on the open letter would quickly double, reaching nearly 7,000 people from academia, civil society, and industry, including almost 2,700 Google employees. On December 9, as protests continued, Google CEO Sundar Pichai issued an apology. “We need to accept responsibility for the fact that a prominent Black, female leader with immense talent left Google unhappily,” he wrote. “Dr. Gebru is an expert in an important area of AI Ethics that we must continue to make progress on—progress that depends on our ability to ask ourselves challenging questions.”