Lani's Share: How AI reflects human bias … and what to do about it

It's been quite unfortunate because we've been aware of bias in algorithms since even before AI became an issue. I remember back in the 2010s when it was a decent deal that google translate was assigning feminine and masculine pronouns to certain jobs even when the initial phrase was gender neutral. Now AI has allowed that bias to be supercharged. I personally pray that AI will scale down in the next couple years to a more reasonable scale, making it easier to monitor and correct these biases.

Loved this talk. Not only because of the topic, but also because the guests are such great communicators, clear, practical, thoughtful, and very accessible.
A few stand-outs for me:
• The reminder that models rely on patterns and stereotypes. It reinforced the idea that AI is probably best used as a collaborator rather than a pure generator.
• Tessa’s answer to why we should care really stayed with me. First, because the information might not be accurate, and ideally we want our beliefs and conversations to be grounded in truth. Second, because the outputs might not be aligned with our own values. Had never thought of this second point!!
• The penny really dropped for me when one of the guests said that we tend to think of AI outputs as objective reality. If people start conforming to those outputs, stereotypes risk feeding back into public opinion.
• I was also very interested in the idea of value-aligned AI, especially the challenge of optimising for social representativeness. It highlights the tension between short-term goals (like engagement) and longer-term societal goals.
• Lastly, I enjoyed rethinking regulation in light of this point: regulation might harm innovation, but the absence of regulation might harm innovation too.
Lots to think about.