,
13 tweets,
3 min read
Read on Twitter
Google's new AI ethics board is a PR disaster, which is funny because as far as I can tell, PR is the only thing it was supposed to be in the first place: vox.com/future-perfect…
Googlers are demanding the removal of Kay Coles James, Heritage Fnd president who thinks trans rights is a scheme to stop women's rights in its tracks by 'redefining men as women', and as I look through her history I am mostly confused why she was chosen in the first place.
She doesn't have any background in tech. She doesn't have any prior writing or research related to AI. No background in the other topics that come up - surveillance, data privacy, international security. She's mostly a culture warrior, so no wonder now we're having a culture war.
This group of eight people who will meet 4 times total and have no power of any kind isn't where you'd start if you were trying to make progress on AI ethics. It's the kind of thing you do if you want to build relationships in DC and get some headlines about your AI ethics board.
While I was writing this piece my editor asked me 'so what *would* Google do if they were actually trying to make progress on AI ethics questions?' That's hard to answer. But I found it instructive to compare them with OpenAI.
OpenAI is the largest AI org with routine public statements that I think takes seriously the ethical implications of what they're doing. They have an internal safety team that includes their top researchers. They release long interesting statements on policy pretty regularly.
OpenAI's press releases say weird-sounding stuff like "we are willing to merge with a value-aligned organization (even if it means reduced or zero payouts to investors) to avoid a competitive race which would make it hard to prioritize safety."
'weird-sounding' is a compliment here - if you're thinking seriously about being responsible in a emerging field, some of your resulting decisions will be really different from industry norms. If you're doing everything the same as everyone else, you haven't thought about it much
When OpenAI came up with GPT-2 a month ago, they thought about its implications for spam, fake accounts, plagiarism, etc. and decided to delay releasing the full model to have a conversation about how to manage risks/benefits.
A bunch of people criticized this - felt they'd gotten the risk-benefit calculus wrong. Responsibility with advanced technology is hard, and I don't agree with everything about how OpenAI approaches it. But all of the above? that's what it looks like if you're actually trying.
You visibly publicly restrict projects to manage their impact on the world. You've thought about the policy implications of your work enough to have explicitly spelled-out policies about what you'd do if your work seems to be fueling bad dynamics.
You have public statements specific, detailed and serious enough that there's anything there to disagree with in the first place! Your ethics team is part of your team, your external ethics advisors are experts, there are visibly decisions that are a result of these concerns.
If Google were trying to make progress on AI ethics, it'd look like that. Instead, I feel like they set this up to cultivate some ties and Be The Kind of Organization That Has An Ethics Board, and it blew up in their face. Go figure.
