Much to the amusment of the wrongthink part of the internet, Amazon decided that automating it's HR functions still needed a little bit of work.
The team had been building computer programs since 2014 to review job applicants’ resumes with the aim of mechanizing the search for top talent, five people familiar with the effort told Reuters.
Automation has been key to Amazon’s e-commerce dominance, be it inside warehouses or driving pricing decisions. The company’s experimental hiring tool used artificial intelligence to give job candidates scores ranging from one to five stars - much like shoppers rate products on Amazon, some of the people said.
“Everyone wanted this holy grail,” one of the people said. “They literally wanted it to be an engine where I’m going to give you 100 resumes, it will spit out the top five, and we’ll hire those.”
But by 2015, the company realized its new system was not rating candidates for software developer jobs and other technical posts in a gender-neutral way.
That is because Amazon’s computer models were trained to vet applicants by observing patterns in resumes submitted to the company over a 10-year period. Most came from men, a reflection of male dominance across the tech industry.
In effect, Amazon’s system taught itself that male candidates were preferable. It penalized resumes that included the word “women’s,” as in “women’s chess club captain.” And it downgraded graduates of two all-women’s colleges, according to people familiar with the matter. They did not specify the names of the schools.
Amazon edited the programs to make them neutral to these particular terms. But that was no guarantee that the machines would not devise other ways of sorting candidates that could prove discriminatory, the people said.
The Seattle company ultimately disbanded the team by the start of last year because executives lost hope for the project, according to the people, who spoke on condition of anonymity. Amazon’s recruiters looked at the recommendations generated by the tool when searching for new hires, but never relied solely on those rankings, they said.
Instead, the technology favored candidates who described themselves using verbs more commonly found on male engineers’ resumes, such as “executed” and “captured,” one person said.
At roughly the same time, the Sun noted that robots and AI's can become racist and sexist on their own.
The study found that robots learn by copying each other, inevitably leading to group formation and shunning outsiders
You don't say?
"Our simulations show that prejudice is a powerful force of nature and through evolution, it can easily become incentivised in virtual populations, to the detriment of wider connectivity with others."
The professor explained that the study even showed how "prejudicial groups" accidentally led to outsiders forming their own rival groups, "resulting in a fractured population".
He added: "Such widespread prejudice is hard to reverse."
Anyone remember Tay?
Of course, those areticles all believe that this horrible natural thing can be corrected, and not by giving the AI more data to work with - something that one would expect when teaching neural networks, but "better" data.
In the computer world, the expression exists, "garbage in, garbage out". An algorithm or program will do exactly what you tell it, and if you feed it crap data, it will give you useless data back.
There is a second expression the programmers of these should be aware of. "Believe your instruments". This doesn't just apply to a steam plant. It applies to all scientific endeavors. It doesn't necessarily mean your instruments are good, but if, say, a pressure gage is reading low, then you better damn well know why it is reading lower than expected before dismissing it.
Similarly, if you design a learning AI to weed through resumes and feed it clues about what constitutes a good resume, and it starts weeding out the women in favor of men, maybe the AI isn't being sexist. Perhaps it is discriminating in the best possible sense of the phrase. Note how often the "fix" is to deliberately weigh the scales of the input data to weed out "masculine assumptions" and language.
In short - they wanted more women. Because they assumed the problem was men being sexist, they keyed a search for competence factors and handed it to an AI. It returned more men and tended to exclude women. So the problem is the AI, or in sexist assumptions of what constitutes a marker for competence, and not that perhaps they entered the competency weights correctly in the first place.
The thought that those markers they chose were accurate in the first place, and that the percentage of applciants who have those that are women is lower due to relatively different rates of obsession and interest apparently never crossed their mind. It's just an "insoluble problem."