Pedro Domingos

It's alarming that NeurIPS papers are being rejected based on "ethics reviews". How do we guard against ideological biases in such reviews? Since when are scientific conferences in the business of policing the perceived ethics of technical papers?

there has been substantial discussion about this for years now, for instance, see https://acm-fca.org/2018/03/29/negativeimpacts/ … … reality is review has always been about more than some pure notion of "science", communities should set standards for their scholarship

This is a controversial proposal, as you know. The road to hell is paved with good intentions.

That proposal only argued we should mention negative impacts along w/ the positive. That it was so controversial was revealing. Relevance to your post is that somehow we entrust reviewers to judge positive things; once it’s about negative things, suddenly we’re on a road to hell.

No, it's more than that. It said that all papers should be required to consider broader impacts. But, as I said in a previous tweet, it's absurd to (e.g.) require all papers about speeding up gradient descent to consider the broader impacts of faster gradient descent.

Maybe I'm missing some nuance, but if a paper is __only__ about faster gradiant descent, then broader implications aren't necessary, but if the faster gradiant descent is embedded in an applications paper why would it be a neg for ethical review to happen on the applications?