Comment on AI hiring tools may be filtering out the best job applicants
lvxferre@mander.xyz 9 months agoThere are two additional issues, related to each other:
- opacity - most of the time you have no clue on what prompted the model to output one or another "decision"
- responsibility - no matter how good or bad it is, software is not a moral agent, thus it should not be put in charge of decisions concerning human beings
Based on that I think that a better approach would be to use the AI model to create a filter, that can be analysed and tweaked by human beings, and then use that filter to select candidates. They won’t do this though - because it screws with their “I did nothing!!! the ai did it!!” excuse to be unfair.
frunch@lemmy.world 9 months ago
Damn, that’s an angle i hadn’t been considering–the “AI did it, not me!” accountability loophole. Delta Airlines was just attempting to pull that on a customer that was given wrong info by a customer service bot. They only managed to get Delta to make good on their offer for bereavement rates when they were taken to court. Thanks AI!
pdxfed@lemmy.world 9 months ago
Two categories of issues at play here:
Companies will miss good talent when AI doesn’t prioritize the way they would otherwise have wanted, doesn’t understand candidate data, or AI hasn’t been trained how to prioritize on the areas a candidate has. With how quickly job markets can change it’s realistic a piece of software or website could rise in popularity and fall over the period of a few years and it might take that long to update and correctly test the damn AI models to recognize and prioritize. All of this should hurt the company and it’s their fault and limit incentives to use AI or black box AI at least as was said above.
Accountability - In the US, it’s illegal now if you have an employment practice (hiring, promotions, firing, etc.) that while it can’t be proven directly or evidence doesn’t exist for a specific case to win in court (prima facie) it can be shown on aggregate to have discriminatory outcomes for protected classes(race, sex, ethnicity, religion, etc.). It’s often impossible to find a smoking gun of “we don’t hire Protected Class X”, but if it can be shown that your employment practices lead to a protected class having much worse outcomes in a company or group, something can be shown to have disparate impact which is illegal and must be remedied.
I fully expect many shittily-trained, poorly or not tested “tools” to be sold and implemented by companies who will eventually be sued for disparate impact. There will be a frenzy of related suits between companies and the AI tool companies.
Creative destruction indeed.