The go-ahead was given to test an elegant idea by Anna and Azalia. The pair wondered if artificial intelligence software could help improve the hardware used in machine learning.
The project attracted interest from the company's chipmaking team, as well as support from Jeff Dean, who is the head of the company's artificial intelligence. It focused on a step in chip design when engineers must decide how to physically arrange blocks of circuits on a chunk of Silicon, a complex, months-long puzzle that helps determine a chip's performance. In June of 2021, the authors of a paper in the journal Nature claimed that a technique called reinforcement learning could be used to perform that step in a few hours.
The results were covered in the media. The professor at UC San Diego predicted in a commentary that the technique would be quickly adopted by chipmakers. Reinforcement learning can be used to improve chip designs.
According to five current and former employees of the company, Mirhoseini and Goldie spent years fighting off claims that their results were wrong or even faked.
The employees claim that Satrajit Chatterjee used the cover of scientific debate to undermine the women. They were not authorized to speak about company matters. Some employees said that he received a written warning for his behavior toward the women, but he continued to criticize their results.
The conflict came to a head in March of this year, after Chatterjee sought permission from research managers to publish a public rebuttal of Mirhoseini and Goldie's Nature study. A committee of senior executives formed to review that paper denied his request. In the same month, he was fired.
On May 2, Goldie posted a document on an internal discussion list about the committee rejecting the paper and accusing him of a series of attacks on the coleads.
The document was posted in a thread where people were reacting to a New York Times article that first reported the firing of Chatterjee, and then complaints from his attorney that researchers were attacking him to shut down a scientific discussion. Most people who joined the thread expressed their support for the two women and their work.
Laurie M. Burgess denied that her client had acted in a way that was inappropriate and that he had evidence that was suppressed by the internet giant. She did not respond to an email that asked detailed questions about the behavior of Chatterjee towards the Mirhoseini.