Post #3385654

I mean, I think he is entirely too credulous of people who claim to be doing things better with AI and discounts a lot of the possible costs of AI systems that malfunction silently and produce plausible bullshit. But I think that those elements complicate his point more than they fully contradict it. Like, consider his last example. Lawyers looking for possible cases for something like the innocence project have to start somewhere, and I can fully believe that the kind of statistical analysis marketing itself as AI is going to be able to pick out viable leads better than doing it randomly or alphabetically or whatever, and that might save the lawyers time and let them help more people than they otherwise would have. But by replacing a naive algorithm with an opaque one you’re essentially baking in any underlying biases in the current system. The people who aren’t going to get seen now are still probably not going to get seen unless they’re right on the margins somehow, and that “somehow” is almost certainly going to be racism, sexism, etc. But by moving that bias from the immediate decision and placing it in the AI model it becomes that much harder to unpack, identify, and address. Like, I fully agree that much if not most of the harm being done by AI right now is more tied to the business and economic structures that it’s embedded in rather than the technology itself. There are very good reasons why so many crypto/metaverse/nft grifters moved straight to AI, and when they try and move on to quantum or web67 or whatever else comes next they will keep right on hurting the world in the same ways unless something about those structures changes. But that doesnt necessary mean we shouldn’t also focus on the harms and limitations that are inherent to the way these things function rather than how they’re used.