By Eric Worrall
When the woke out woke the woke. Back in December, Google fired AI Ethics Unit co-leader Timnit Gebru, in relation to her paper “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?”. Google have now just fired their other ethics head, Margaret Mitchell, apparently for trying to gather evidence while investigating the ousting of Timnit.
I’m fired: Google AI in meltdown as ethics unit co-lead forced out just weeks after coworker ousted
Plus: IBM reportedly trying to sell Watson AI Health, and more
Katyanna Quach Mon 22 Feb 2021 // 12:21 UTC
Google has finished its probe into the controversial ousting of Timnit Gebru, co-leader of its Ethical AI unit. The ad giant promised to implement new procedures around “potentially sensitive employee exits,” though it did not make its findings public.
Gebru said she was fired for warning coworkers in an internal memo that, due to management apathy, it was a waste of energy trying to foster diversity, equality, and inclusion within the Silicon Valley goliath. Google claimed she effectively resigned.
Meanwhile, Margaret Mitchell, who also co-led the Ethical AI unit alongside Gebru, said on Friday she has been fired. Mitchell had been locked out of her corporate account for weeks.
3 ENVIRONMENTAL AND FINANCIAL COST
Strubell et al. recently benchmarked model training and development costs in terms of dollars and estimated 𝐶𝑂2 emissions . While the average human is responsible for an estimated 5t 𝐶𝑂2𝑒 per year,2 the authors trained a Transformer (big) model  with neural architecture search and estimated that the training procedure emitted 284t of 𝐶𝑂2. Training a single BERT base model (without hyperparameter tuning) on GPUs was estimated to require as much energy as a trans-American flight.
While some of this energy comes from renewable sources, or cloud compute companies’ use of carbon credit-offset sources, the authors note that the majority of cloud compute providers’ energy is not sourced from renewable sources and many energy sources in the world are not carbon neutral. In addition, renewable energy sources are still costly to the environment,3 and data centers with increasing computation requirements take away from other potential uses of green energy,4 underscoring the need for energy efficient model architectures and training paradigms.
Timnit has also criticised other issues with AIs, for example in 2018 she helped stop the rollout of an Amazon facial recognition system being used by police agencies, by demonstrating the flawed Amazon system was 34% less capable of correctly identifying black women, compared to its ability to correctly identify white men. The problem – the dataset used to train the AI mostly contained white faces.
What can I say – losing one ethics head could be an accident. Losing two in quick succession starts to look like carelessness, perhaps even raises suspicions that what Google really wants is a compliant ethics team which does whatever top management tells them to do.