Re: AI/ML: overhyped, frequently mal-used
The worst part of it is, there are ways to keep your AI/ML under control, but you need to design it with that in mind from the very beginning. And even then, you'll need to constantly keep looking into it to see what it's doing, and make sure it's not come up with some crazy solution that doesn't make sense but fits the data.
I'm personally against using this opaque type of AI for these tasks. AI is discriminatory by nature, it's what it's designed to do, and it will just go with what you give it.
Here is a Reuters piece about an AI Amazon used in their hiring process: https://www.reuters.com/article/us-amazon-com-jobs-automation-insight-idUSKCN1MK08G
Turns out that if you have a sexist hiring process for (at least) a decade, you turn that into a dataset, and feed it into an AI, the AI will be sexist too! Who would've thought. Assuming Amazon had good intentions here (which may be a bit of a stretch), it just shows how difficult it is to rid your dataset of any accidental discrimination. Confounding factors can be a massive pain too, especially with self-training AIs.