Google Has a Striking History of Bias Against Black Girls

Machine learning systems are good at finding the unconscious biases of large populations. Any change to force notions of political correctness on top of that is very complicated, and makes the algorithm work worse for its intended purpose. Literally, failing to accept the discrimination inherent in the data loses you money directly or indirectly.

This is well-known, and not Google's fault. Getting offended at Google for a reality they didn't create and can't control seems silly to me. If you don't like the result, work to change the incentives. Or work to change people's biases.

This kind of problem is not new. For example insurance companies have long known that where you live affects how much they are likely to pay out. If they base insurance rates on the data, the result is that your zip code becomes a bigger determinant of insurance rates than your driving record. Which results in very large effective racial discrimination. There are laws limiting that, for example prop 109 in California. However it is an eternal struggle because, in fact, competing insurance companies have a good motivation to have the cost of insurance reflect their best estimate of the expected cost of insuring you, and their cost really is a lot higher if you're in a "black neighborhood".

> Given that women are half the population, and African Americans and Hispanics each make up about 12%, we can state categorically that Google's hiring doesn't match the general US population.

But if you account for demographics within the industry do the numbers remain biased?

For example: men account for less than 10% of nurses so if you were to go to a hospital with 50% male nurses they would be way overrepresented. Statistically speaking of course, I'm in no way trying to give the impression we need to do something about the lack of diversity within the nursing profession.

--edit--

...or maybe we need to do something about the lack of diversity in the nursing profession? Don't actually know how this ball rolls?

I do not think the word "racism" is a good fit for this. There is no prejudice, or a belief that another race is superior.

This is not singling out black women. Any search for "blonde teens" during that time would yield porn-related results too. A search engine just tries to give the most relevant search. Likely, a large part of the population was in search of sex when searching for these type of queries. The "bias" is not with Google, the "bias" is in society, and Google merely reflects this.

I can also understand why "three black teenagers" would give mugshots over "three white teenagers". The term "thee black teenagers" just appears more in crime news stories. I think this is a legit societal problem, and changing perception does not change harsh reality.

I really am on the fence about the efforts to change this. For instance, the majority of people have a negative view of the "black lives matter" movement, and this is not reflected in the search engines right now. I still think people searching for "black girls" are looking for porn, not "black girls who code"-type stuff, but there is no porn in sight. So you cripple the search engines accuracy in favor of political correctness.

I feel a search for "professional" or "CEO" should reflect reality, no more, no less. I don't think a search engine index is the place to start change in society (I do think the number of female CEO's is too low, but this needs fundamental change, not a lick of paint).

I also think "cultural appropriation" is antithetical to more inclusion. If white people are restricted from wearing afro hair, dreadlocks, or hoop ear rings, these styles will always remain associated with people of color. The reverse is also true: If no black man dons a business suit, there will be no black men when searching for "professional business man", because there will be no data to support that.

I do not think anyone needs protection from search technology. I do not think it affects people's lives to a big degree, if a neural network makes a mistake and labels their partner as a gorilla. These are honest mistakes, due to not having enough training data, that nobody deliberately makes.

Probably not fair to compare against the entire population, you would need to compare against the working age population. Additionally you might want to take into account local populations for where you're hiring into.

The problem statement is given the sentiment space of reactions to similar clusters of search queries, minimize negative sentiment across the most clusters given resource constraints (time, cost, CPU).

This is a hard problem. There are many search queries, there are many clusters of similar search queries that address partially overlapping (topic,audience segment) pairs. Deriving sentiment scores from these segments is difficult without any feedback mechanisms. Choosing which segments to focus on is difficult or counterintuitive and depends on the topic and audience and characteristics of the audience. Once a topic and audience segment is identified for remediation, the task of effecting change is again topic and audience specific (resistant to automation).

This is a hard problem. Neither a brute force "human review squad" approach nor an automated "deep learning" approach will provide 100% coverage through all time. Does that mean that they shouldn't try? No, of course not, and they almost certainly try every day. However, it's inevitable that they will be subject to sniping articles no matter what they do.