Notices by Denian (denian@chaos.social)

@MicroSFF Do wires count as string? I mean, once you have the whole "very long, thin stuff" thing down, I can't imagine variations being that hard to come up with... And surely a space ship needs at least one wire _somewhere_?

@MicroSFF Anybody who has ever tried to synthesize a violin (or any of its differently sized siblings) knows very well that there IS a difference, and most synthetic wood winds aren't that great, either...

@inthehands @brianary @skye @SallyStrange @blogdiva I'm not building a house in Minnesota, though. I'm building a tent, selling it in Florida or Texas or some other place near the equator, and somebody's bought it and went camping in Minnesota.

@inthehands @brianary @skye @SallyStrange @blogdiva But ok, say I'm building that house in Minnesota. I'm building it that way because it's the only type of house I ever built - the only type of house ANYBODY has ever built, in fact. There is no technology to build better houses. So why, exactly, are you expecting me to?

@inthehands @skye @SallyStrange @blogdiva Please tell me how you program any part of a neural network (or any other machine learning algorithm, for that matter) not only what humans are, but what skin is, that it has a color, and that it shouldn't change said color, without training it. Or rather, do the opposite - make it actively racist without training it.

@brianary @inthehands @skye @SallyStrange @blogdiva Basically neutrality is a bias on its own? I can accept that, but then where's the part that should be changed (if it was possible) - should it stop being neutral in order to avoid being biased?

@skye @SallyStrange @blogdiva I am TELLING you that it's IMPOSSIBLE to influence, no matter HOW biased you are, because it's too complex for anybody to be ABLE to manipulate it. I also showed you one place where you can learn about it, yourself, and there are lots of other places to do so. If you refuse to do so just so you can stay willfully ignorant and pretend more bias makes the impossible possible, I can't help you.

@blogdiva @meredithw @Wolven It's not a software issue, it's a training data issue. There's a difference. Nobody actively "programs" an AI to be racist, but people rarely, if ever, remember to actively escape the demographic distribution they see every day when they look for training data. And in some cases, it's actually impossible, because genetic outliers like albinos, for example, just ARE too rare to be well-represented in training image sets.