OAKLAND, CA –
Google told Alphabet Inc on Monday that it had recently fired a senior engineering manager after colleagues whose remarkable research on artificial intelligence software he tried to discredit accused him of harassment.
The controversy, which stems from efforts to automate chip design, threatens to undermine Google’s reputation in academia. It could also disrupt the flow of millions of dollars in government subsidies for AI and chip research.
Google’s research unit has been facing scrutiny since the end of 2020, after workers openly criticized staff handling complaints and publishing practices.
The new episode came after the June issue of the scientific journal Nature published a Methodology for Graphing Rapid Chip Design, led by Google scientists Azalea Mirhoseini and Anna Goldie. They found that AI could complete a key step in the chip design process, known as floor planning, faster and better than an unspecified human expert, a subjective reference point.
But other Google colleagues in an article anonymously published online in March, “Stronger Bases for Assessing Deep Reinforcement Training in Chip Deployment,” found that two alternative approaches based on core software outperformed AI. One beat it in a well-known test, and the other in Google’s own section.
Google declined to comment on the leaked draft, but two workers confirmed its authenticity.
The company said it refused to publish Stronger Baselines because it did not meet its standards, and shortly afterwards fired Satratjit Chatterjee, the driving force behind the work. He declined to say why he fired him.
“It’s unfortunate that Google has taken this turn,” said Lori Burgess, Chatterjee’s lawyer. “His goal has always been to be transparent about science, and he has called on Google to address this for two years.”
A Google Goldie researcher told the New York Times, which first reported the dismissal on Monday, that Chatterjee had been harassing her and Mirhosseini for years by spreading misinformation about them.
Burgess denied the allegations, adding that Chatterjee had not distributed Stronger Baselines.
Patrick Madden, an associate professor of chip design at Binghamton University who has read both articles, said he had never seen an article before in Nature that lacked a good comparison point.
“It’s like a reference problem: everyone gets the same puzzle pieces, and you can compare how close you are to fixing everything,” he said. “If they had to give results according to some standard standard and they were stellar, I would praise them.”
Google said the human comparison was more appropriate and that software licensing issues had prevented it from mentioning tests.
Research by large institutions such as Google in well-known journals can have a huge impact on whether such projects are funded in the industry. A Google researcher said the leaked article unfairly opened the door to questions about the authenticity of any work published by the company.
After “Stronger Baselines” appeared online, Zubin Gahramani, vice president of Google Research, wrote on Twitter last month that “Google supports this work, published in Nature on ML for Chip Design, which is replicated independently, with open code and used in Google production. “
Nature, citing an official holiday in the UK, did not comment immediately. Madden said he hoped Nature would review the publication, noting that reviewers’ notes indicated that at least one had asked for benchmarks.
“Somehow this never happened,” he said.
(Report by Paresh Dave. Edited by Gary Doyle)
Add Comment