Post AYDuveOCrrT22sIOfY by ddrake@mathstodon.xyz
 (DIR) More posts by ddrake@mathstodon.xyz
 (DIR) Post #AYDuveOCrrT22sIOfY by ddrake@mathstodon.xyz
       2023-07-30T11:57:38Z
       
       0 likes, 1 repeats
       
       "Algorithms are to bias what centrifuges are to radioactive ore: a way to turn minute amounts of bias into pluripotent, indestructible toxic waste."Brilliant quote from @pluralistic in https://pluralistic.net/2023/07/26/dictators-dilemma/. Training a machine learning algorithm is a very conservative act: it starts from the training data -- the past -- and presumes that's precisely what you want in the future. So it's conservative in several senses: conserving past observations/data; making the future just like the past, and so on.If that training data is good, and conserving past behavior is really and truly what you want, this is great!But, wait: what does "good" mean? And who decides what we really and truly "want"?Required reading, in my not-at-all humble opinion, is "Data Feminism": https://data-feminism.mitpress.mit.edu. My biggest criticism of their book is the title, which can mislead you into thinking it's somehow specific to gender-related issues, but it addresses bias of all kinds, and how we, as a society, use algorithms.
       
 (DIR) Post #AYDuvgx1LOVHzQbEDA by ddrake@mathstodon.xyz
       2023-07-30T12:59:21Z
       
       0 likes, 0 repeats
       
       @pluralistic Also, required reading: Cathy O'Neil's "Weapons of Math Destruction": https://en.wikipedia.org/wiki/Weapons_of_Math_DestructionThat came out in 2016, but it's (1) still super, super relevant; and (2) would be great to see an update reflecting the current hype with AI and large language models.