Raw LLM Responses

This seems like low hanging fruit that we’ve heard before. While these ppl are right to point this issue out and the various ways in which it might distort institutes and social practices within our society, I look a bit askance at criticism that merely rests on the fact that these algorithms are trained off biased data that perform some task—like a facial recognition task—with an increased false positives rate with respect to some population related to those tasks (us black ppl, for example). The reason is the question is not whether these systems are totally accurate or even totally free from bias. It’s whether they do better on these metrics than ordinary humans (in this case, cops). It is not unheard of to be falsely identified as the suspect of a crime and certaintly not unheard of to be subjected to such false identification to a higher degree if you’re black. So we have to judge this tech, with respect to bias, on whether it motivates or exacerbates the problems we already experience with policing. The privacy angle is more interesting. It we could have perfect recognitional systems, would we want them?—because of all the personal data we would have to part with, among other things. I wonder on the convenience handcuffs/opt-in stuff she talks about at the end, if you cannot start addressing that through a document descruction/expoliation policy and then fold in fourth amendment analysis on how the information can be used before destruction, coupled with exclusionary rule stuff regarding any prosecution that tries to make use of that data without the appropriate exception… just an idea. Anyways, interesting stuff, but not as much for the reasons on which the speaker focuses absent deeper comparative analysis with current human practices (and the problems that accompany them)