It also elides “AI safety” (Toner’s thing) and “AI ethics” (Gebru’s thing). They’re two different things. Jammed together here because both are women (FFS).

“AI safety” is the sci-fi, paperclip maximisation, fantasies about the potential future of AI.

“AI ethics” is the real actual harms done in the here and now, by embedding existing biases into decision-making, and consuming enormous amounts of resource.

Meredith Whittaker sums up the difference nicely in this interview:

So in 2020-21 when Timnit Gebru and Margaret Mitchell from Google’s AI ethics unit were ousted after warning about the inequalities perpetuated by AI, did you feel, “Oh, here we go again”?

Timnit and her team were doing work that was showing the environmental and social harm potential of these large language models – which are the fuel of the AI hype at this moment. What you saw there was a very clear case of how much Google would tolerate in terms of people critiquing these systems. It didn’t matter that the issues that she and her co-authors pointed out were extraordinarily valid and real. It was that Google was like: “Hey, we don’t want to metabolise this right now.”

Is it interesting to you how their warnings were received compared with the fears of existential risk expressed by ex-Google “godfather of AI” Geoffrey Hinton recently?

If you were to heed Timnit’s warnings you would have to significantly change the business and the structure of these companies. If you heed Geoff’s warnings, you sit around a table at Davos and feel scared.

Geoff’s warnings are much more convenient, because they project everything into the far future so they leave the status quo untouched. And if the status quo is untouched you’re going to see these companies and their systems further entrench their dominance such that it becomes impossible to regulate. This is not an inconvenient narrative at all.