AI bias: word2vec data encodes racism & sexism https://arxiv.org/abs/1606.06121 need more diversity http://www.bloomberg.com/news/articles/2016-06-23/artificial-intelligence-has-a-sea-of-dudes-problem …pic.twitter.com/odbhY3CwCp
This is not bias, it's an accurate reflection of reality. Why blame ML algorithms for telling the truth? Or tweak them not to?
-
-
we should tweak them not to, as this paper is describing, I feel?
-
No, ML algorithms should tell the truth, and we should decide what to do with it, not tamper with the evidence.
- Näytä vastaukset
Uusi keskustelu -
-
-
this is the entire point of decision-theoretic classification: you weight predictions to avoid high-cost mistakes.
-
applying a raw probability model without considering utilities of outcomes is asking to make expensive mistakes.
Keskustelun loppu
Uusi keskustelu -
-
-
Or a reflection of a biased train set & removing might be a legal requirement. ML can helphttp://arxiv.org/abs/1511.05897
Kiitos. Käytämme tätä aikajanasi parantamiseen. KumoaKumoa
-
-
-
Not sure if it's worse that you think black_male:assault white male:entitled_to is reality or that a news corpus is reality.
Kiitos. Käytämme tätä aikajanasi parantamiseen. KumoaKumoa
-
-
-
Forgive me
@pmddomingos, but this notion of "truth" seems simplistic. Is it "true" that professor - associate = man - woman?@jackclarkSF -
It is true only in the sense that these are the embeddings that best predict skipgrams, not true meanings of words
Keskustelun loppu
Uusi keskustelu -
Lataaminen näyttää kestävän hetken.
Twitter saattaa olla ruuhkautunut tai ongelma on muuten hetkellinen. Yritä uudelleen tai käy Twitterin tilasivulla saadaksesi lisätietoja.