Pedro Domingos

Your ethics are that if the machines decide we shouldn’t have ethics, then we shouldn’t have ethics. I think a reasonable stance is somewhere in the middle.

Biased algorithms regarding people rely on biased data from biased people which creates biased hyperparameters and a whole lot of hurt. So to keep the reliability of the algorithm should we remove data points such as etnicity and age and see how the algorithm works without them?

because they sure can't stay as they are. College admissions for example are heavily biased regarding ivy league schools and now we know many students scammed their way into them. Should we scan applicants for a job without knowing their college?

If the law says you can't discriminate based on race, that law ought to cover any machine-learning algorithm that tunes parameters in a way that leads to racially-biased results. Developing AI systems that are robust to *not* violating local anti-discrimination laws seems good?

By the same definition of 'biased' policing will not be anti-discriminatory until they arrest men and women at the same rates. Can you imagine how ridiculous that would be in practice?