By Paresh Dave
OAKLAND, Calif. (Reuters) – Alphabet Inc.’s Google said on Monday it recently fired a senior engineer after colleagues whose pioneering research into artificial intelligence software he had tried to discredit accused him of harassing behavior.
The dispute, which stems from efforts to automate chip design, threatens to undermine Google’s research’s reputation in the academic community. It could also disrupt the flow of millions of dollars in government grants for AI and chip research.
Google’s research unit has come under scrutiny since late 2020 after employees openly criticized its handling of staff grievances and disclosure practices.
The new installment comes after the science journal Nature published “A Graph Placement Method for Rapid Chip Design” in June, led by Google scientists Azalia Mirhoseini and Anna Goldie. They discovered that AI can complete an important step in the chip design process, known as floorplanning, faster and better than an unspecified human expert, a subjective reference point.
But fellow Googlers found in a paper anonymously posted online in March — “Stronger Baselines for Evaluating Deep Reinforcement Learning in Chip Placement” — that two alternative approaches, based on foundational software, outperform AI. One beat it in a well-known test and the other in a proprietary Google rubric.
Google declined to comment on the leaked draft, but two employees confirmed its authenticity.
The company said it refuses to release Stronger Baselines because it doesn’t meet its standards and soon after fired Satrajit Chatterjee, a leading driver of the work. It declined to say why it fired him.
“It’s unfortunate that Google has taken this turn,” said Laurie Burgess, an attorney for Chatterjee. “It was always his goal to create transparency about science, and over the course of two years he pushed Google to look into it.”
Google researcher Goldie told The New York Times, which first reported Monday about the firing, that Chatterjee had been harassing her and Mirhoseini for years by spreading misinformation about them.
Burgess denied the allegations, adding that Chatterjee Stronger Baselines was not leaked.
Patrick Madden, an associate professor specializing in chip design at Binghamton University who read both articles, said he had never seen an article prior to the one in Nature that lacked a good point of comparison.
“It’s like a reference problem: everyone gets the same puzzle pieces and you can compare how close you get to getting everything right,” he said. “If they gave results on a standard benchmark and they were outstanding, I would commend them.”
Google said the comparison to a human was more relevant and that software licensing issues prevented it from mentioning tests.
Studies by large institutions such as Google in well-known specialist journals can have an outsized influence on whether similar projects are funded in the industry. A Google researcher said the leaked paper unfairly opened the door to questions about the credibility of any work published by the company.
After “Stronger Baselines” surfaced online, Zoubin Ghahramani, vice president at Google Research, wrote on Twitter last month that “Google stands by this work published in Nature on ML for Chip Design, which is independently replicated, open-source and in production used by Google.”
Nature did not immediately comment, citing a UK public holiday. Madden said he hopes Nature would revisit the publication, noting that the peer reviewer notes show at least one asked for results on benchmarks.
“Somehow that never happened,” he said.
(Reporting by Paresh Dave. Editing by Gerry Doyle)