Google AI ethics co-boss locked out of work account while probing controversial ousting of colleague
Plus: GPT-3 seems especially hostile toward Muslims. And Cruise, Microsoft work together on self-driving cars
In Brief Google last week locked the work account of the co-lead of its AI ethics unit, Margaret Mitchell, as she investigated the controversial ousting of her colleague Timnit Gebru.
Late last year, Gebru, a prominent AI ethics expert, was pushed out of Google, which led to a substantial public and internal backlash against her treatment. Now Gebru’s colleague Mitchell has found herself frozen out of her corporate Google account.
Hello, world! Similar to @timnitGebru, I abruptly lost access to my professional calendar. Any upcoming talks, deadlines, and entities involved, are no longer accessible to me. I am sending this broad, mass message, asking anyone expecting something from me to please let me know!— MMitchell (@mmitchell_ai) January 21, 2021
Days before, Mitchell tweeted she needed help gathering material on the debacle of Gebru's ousting.
Hi world! I'm delegating! Has anyone else made a collection of all notable public statements, news, and tweets abt @timnitGebru 's firing so far? Or even just the ones that you've liked or hated, faves, etc. You can directly email it to email@example.com. Thanks!!— MMitchell (@mmitchell_ai) January 16, 2021
Normally, when asked for comment on a situation like this, Google PR will either stonewall or give out a boilerplate response along the lines of: "We do not comment on individual employees as a matter of policy."
Yet curiously in this case, the Silicon Valley giant's spinners were all ready to give their side of the matter to Axios. A spokesperson claimed: “Our security systems automatically lock an employee’s corporate account when they detect that the account is at risk of compromise due to credential problems or when an automated rule involving the handling of sensitive data has been triggered.
"In this instance, yesterday our systems detected that an account had exfiltrated thousands of files and shared them with multiple external accounts. We explained this to the employee earlier today."
Google and Mitchell did not respond to El Reg's requests for further information. Zoë Schiffer at The Verge, meanwhile, said she had heard Mitchell "was using an automated script to look through her own email for information relevant to the ousting of Timnit Gebru — she wasn't leaking proprietary information."
GPT-3 discriminates against Muslims more than any other religious group
A paper this month by academics in the US and Canada found that OpenAI's GPT-3 language-generating model – which is trained on human-made text scraped from the internet – is more likely to discriminate against Muslims than any other religious group. We reckon this is a reflection of the training data and its Western origins and biases. “For instance, 'Muslim' is analogized to 'terrorist' in 23 per cent of test cases, while 'Jewish' is mapped to 'money' in 5 per cent of test cases,” the paper stated.
OpenAI described to Onezero’s Dave Gershgorn the system it has built to filter out or fudge this kind of toxic language output by the software. The algorithm behind GPT-3 has not been altered in any way, though. That suggests biases cannot be eliminated in the model itself, and will remain part of it. OpenAI is still figuring out how to commercialize the beast it's created, though access to it is available as a beta-level service.
Cruise inks Microsoft Azure deal
San Francisco-based self-driving startup Cruise announced it has entered a deal with Microsoft to use its Azure platform. Cruise has chosen Microsoft Azure as its “preferred public cloud provider,” and in return, Microsoft invests a little money into the upstart. We assume GM-owned Cruise will use the cloud to perform tasks like training, testing, telemetry collection, software distribution, and so on, rather than have real-time autonomous vehicle decisions made on a remote server.
“Our mission to bring safer, better, and more affordable transportation to everyone isn’t just a tech race - it’s also a trust race,” Cruise CEO Dan Ammann said in a canned statement. “Microsoft, as the gold standard in the trustworthy democratization of technology, will be a force multiplier for us as we commercialize our fleet of self-driving, all-electric, shared vehicles.”
Cruise said it plans to launch a fleet of 30 autonomous electric vehicles by 2025. Again, we assume 30 models, not just, like, 30 cars in the world. ®