True fact: there is not a single proven case to date of discrimination by machine learning algorithms.
-
Algorithms can't be tried in court. No algorithm is racist or sexist in design, it is always a fxn of data or implementation. You know this.
By "culprit" I mean "cause", not "guilty party". People/data/hand-coded algorithms can be biased. ML algorithms can't. Many don't know this.
-
-
By "ML algorithm" u must mean the model+training procedure (prolly not what most of the world wd mean). Still not true: inductive bias/prior
-
By analogy, the synaptic firing patterns that give rise to human thought also aren't biased. Most people don't care about that in isolation.
Keskustelun loppu
Uusi keskustelu -
-
-
Every ML alg I've worked with over-predicts the majority class. In many classes, that leads to discrimination against human minorities
-
I think I'm misunderstanding you,
@tdietterich. Anything trained with a proper scoring rule should match class frequencies, right? - Näytä vastaukset
Uusi keskustelu -
-
-
So you're looking for a court case that ruled that something deep in the algorithm, like the kernel trick, is discriminatory?
-
If so, can you point to an example where someone in FATML made that kind of argument?
Keskustelun loppu
Uusi keskustelu -
-
-
If that's what you wanted people to know, it should have been more obvious from your first post, instead of having to clarify twice. 1/
-
Back to my other point of -how did you THINK your first post would be interpreted? There's plenty of us trying to evangelize good science 2/
- Näytä vastaukset
Uusi keskustelu -
Lataaminen näyttää kestävän hetken.
Twitter saattaa olla ruuhkautunut tai ongelma on muuten hetkellinen. Yritä uudelleen tai käy Twitterin tilasivulla saadaksesi lisätietoja.