> Only time I hear about those teams is when they get fired. It is a risk mitigation measure. The company is accused of bad things, and the higher ups want to

> Only time I hear about those teams is when they get fired. It is a risk mitigation measure. The company is accused of bad things, and the higher ups want to be able to say that they are doing something about it. They create a team who will investigate such things and make recommendations. For example before elon times twitter had a problem. They had a feature which picked the "most important" part of an image so they can display the images automatically on mobile. The problem they had is that someone found an image where there were two politicians side by side and the algorithm was focusing only on the white one, ignoring the black one. It was quite literally "marginalising" the black politician. Not a good look. Not the end of the world. The walls were not crumbling yet. The advertisers haven't whitdrawn their budgets yet but you know, you don't want to be known as "that racist social network". In a situation like this one possible solution is to start a small team, give them a bit of budget and task them figuring out if the machine is racist and what can be done about it. Preferably you fill the team with well-spoken academic types to give the effort more credence. (without actually, you know, cutting into the profit) This is obviously the cynical view. The less cynical view is that you want a team who prevents issues like that from getting into production. Same way you have legal to tell you what is and isn't legal, you have AI ethics to tell you what is and isn't consistent with the ethics of your organisation. Someone who doesn't do the work itself, but helps other teams by reviewing their proposals for potential issues. For example in the "picking the important part of an image" if they were in the room during the design phase they could have asked why someone thinks this task should be done by the machine? Couldn't we ask a user what they feel is the right cropping of their image for mobile? And only if it must be done at all would they ask if the developers have tested their solution on representative test datasets.