The problem of AI bias, once ignored, then a minority concern, is now well into the stage of being popularised and more widely recognised. It’s a debate to which Sandra Wachter has herself been an thoughtful contributor. The fact that AI can replicate and reinforce human biases is of course critically important, but it risks obscuring the fact that the seed of the bias is unautomated human behaviour. So AI which is no more biased than humans should be seen as a pretty minimal target, rather than the ceiling of aspiration.
This post is a manifesto for doing better, for rejecting the idea that the new ways need only be no worse than the old. It’s not about specific solutions, but it is an important framing of the question.