Pedro Domingos

Would the New England Journal of Medicine accept the Tuskegee Syphilis Study today?

The experiment would be illegal, so the problem wouldn't arise.

Due to regulatory processes that have been put in place partly due to experiments such as those. No such regulatory processes yet exist for machine learning research.

Who says they're needed? And who decides?

Same argument can be made for medical ethics. Many disciplines, even other engineering disciplines, have clearly regulated ethical standards.

How is our field an exception given well-known applications that affect people's lives and freedoms like predictive policing, person re-identification that can be used in surveillance states, etc?

We already have computer science-wide ethical standards, set by ACM and others. It's not clear we need special ones for AI. Also, standards for research and standards for deployed applications are very different things.