Global Al safety isn't one-size-fits-all, Microsoft researcher says

Kalika Bali, Senior Principal Researcher at Microsoft, discusses AI safety and bias in regional languages.

Transcript

00:00It's really difficult to talk about safety and bias in regional languages,

00:05especially because bias does not translate well, right? So if you think

00:12something is biased or something is harmful and it's expressed in English,

00:17just translating it into another language is not going to solve the

00:22problem because it would be, even if it was the same kind of bias, it would be

00:26very different. The other thing is that in a lot of societies because of, and this

00:32is maybe something that we haven't thought so much or I'm not aware of in

00:37the Western world, is that when there are inequities, right, even the notion of bias

00:44kind of varies quite a lot. So if I was to say that women should stay at home and

00:50only look after their family and children, I hope that everyone here will

00:56think that this is a biased sentence. But, you know, in India, if I was to just step

01:01out of my city, you know, a lot of women would think that's a perfectly normal thing.

01:07So having to work on a safety model under those circumstances, it's really difficult.