Post AdRoN5gGghrNXntX04 by tiotasram@kolektiva.social
(DIR) More posts by tiotasram@kolektiva.social
(DIR) Post #AdRLjh6tJFW7KUm0ky by Linux_in_a_Bit@linuxrocks.online
2024-01-02T15:00:51Z
0 likes, 0 repeats
Machine learning is great unless you use it to profit off of people or their work.Usually Good:Weather modelsMedical thingsVirtual assistantsSearch enginesSpam filtersTranslationetc.Usually Less Good:Generative "AI"Content recommendation systemsTargeted advertisingUser profilingetc.
(DIR) Post #AdRoN5gGghrNXntX04 by tiotasram@kolektiva.social
2024-01-02T18:06:09Z
0 likes, 0 repeats
@Linux_in_a_Bit feel like this is too simplistic: AI has several other typical harms, notably when it replaces more-reliable or less-biased systems. For example, I'd definitely hesitate to say that medical AI is "mostly good" when we'd expect such systems to typically do a lot of harm through bias.
(DIR) Post #AdRoXn647fRF5o3zl2 by Itty53@mstdn.social
2024-01-02T18:07:48Z
0 likes, 0 repeats
@Linux_in_a_Bit I think "medical things" is a bit too blanketed of a statement to be good. Famously, one "very accurate" model was found to be largely basing its decisions on just the presence of a ruler in the xray image it was looking at. Because human doctors put rulers _next to_ cancerous legions. Literally nothing good about that.There's also a need to account for biases. The data is trained on represents biased treatment, because it's human made data. That's really problematic too.
(DIR) Post #AdS8ha6dH8JMmVH528 by Linux_in_a_Bit@linuxrocks.online
2024-01-02T22:07:40Z
0 likes, 0 repeats
@tiotasram That is true, I'll update the post. Thank you :)