Pedro Domingos Is Racist and Sexist | Hacker News Reader

Pedro Domingos Is Racist and Sexist

medium.com

2 pointstartaglia4 comments

How very inconvenient that the evil party in this case has a name that does not place him within the primary WASP target group.

I will take issue with core-questions on one point, however. The use of pseudonyms has a long and glorious history in America, and in the present moment it serves to protect right-thinking persons more than it does advocates of Critical ideologies.

I wonder when all is said and done, what percentage of my professors will have been cancelled, and what percentage of them are the ones doing the cancelling.

> If an ML system is designed to maximize accuracy in predicting (say) criminality and there is a correlation between criminality and skin color, it’s quite likely that there will be more false positives of some skin colors than others. It’s not discrimination.

How is this wrong? Have you even looked at the statistics of crime by race, at all? If you're expecting a perfectly proportionate level of criminality representative of the host population, I've got news for you: it's not like that.

ML systems are simply "naive" in the sense that they don't know when they should hide the truth because it disagrees with officially promoted narratives. I predict a market for ML censorware that can tell you when your ML system has discovered yet another uncomfortable truth and help you to prevent it from escaping so that you don't end up with someone attacking you on Hacker News.

The OP is lambasting someone and assassinating their character simply for saying things that anyone with an open mind can research and confirm for themselves with a quick perusal of crime states. Disgusting. Cowardly, too, using a lame pseudonym: "Ann Imus", which is obviously a thinly veiled reference to "racial animus".

How about instead of leveling such insults, you instead familiarize yourself with reality, and then if you're still bothered, pursue a life where you can make a difference to these outcomes instead of hiding behind a keyboard insulting people who notice them?

I's complicated, and she's assuming a lot of high-ish level understanding of law, ML, and society, but the short answer is "Correlation doesn't imply causation, but correlation is what ML systems do."

So, it may, in fact, be true that if you aggregate over statistics in a society, you'll find correlation between skin color and crime recidivism, or likelihood to embezzle money, or some other factors. It's not "okay" to bake that signal into your system for risk assessment because it creates positive feedback loops on societal inequality that penalize people for qualities out of their control, and (at least in the US) we've broadly decided that's not cool (and, in fact, narrowly codified that into law---denying people a loan based on skin color is a crime, and denying people a loan if your ML system decides "Janelle" is probably not going to make good on repayment because the name correlates to dark skin and dark skin correlates to bad loan payment history is still a crime; you can't dodge the intent of the law by adding length to the lever).

80% of the time, I don't think I'd assume someone is being racist if they act like they don't know this, but the subject of the Medium post should, ostensibly, know this.

Reply on news.ycombinator.com