Tonight in briefly got tangled in a rather pointless Twitter argument about #AIethics—pointless because the other party isn’t interested in discussing in good faith. One point of clarity has come out of that though, in the responses of a few of us and I want to pull it out here:
Ethical considerations are not a separate concern from whether the AI research is “sound”: sound AI research requires not only valid math but also sensible task design.
A lot of the ethically questionable things we’re seeing (predicting faces from voices, predicting “intellectual ability” from short text responses, etc) are cases where it doesn’t make sense to predict A from B, by any model.
Big neural nets are really good at pattern recognition and any given data set is likely to have enough artifacts in it that *for that dataset* your black box dataset can do better than baseline at predicting A from B.
But that doesn’t mean it’s learned anything interesting about the relationship between A and B or even that there is a relationship between the two.
Without careful task design, “AI” research is bad science. Depending on the specific task, it could be just bad science or it could be likely to cause harm, too. But these aren’t secondary considerations that eg shouldn’t be picked up by peer review. They’re fundamental.
I’m not the only one whose responses in that other thread (set of threads) included these points, but I’m tweeting on my phone on the train here and can’t quickly find the others again. I know @robokatherine is one!
s/dataset/model/
*sigh*. Tweeting past my bedtime on a train....
*sigh*. Tweeting past my bedtime on a train....
Here’s another person making the same point too:
