I’m sick of this framing. Tired of it. Many people have tried to explain, many scholars. Listen to us. You can’t just reduce harms caused by ML to dataset bias.https://twitter.com/ylecun/status/1274782757907030016 …
Show this thread
-
If I had wanted to "reduce harms caused by ML to dataset bias", I would have said "ML systems are biased *only* when data is biased". But I'm absolutely *not* making that reduction. 1/N
I'm making the point that in the *particular* *case* of *this* *specific* *work*, the bias clearly comes from the data. 2/N
-
-
There are many causes for *societal* bias in ML systems (not talking about the more general inductive bias here). 1. the data, how it's collected and formatted. 2. the features, how they are designed 3. the architecture of the model 4. the objective function 5. how it's deployed
-
(that was 3/N). When you use raw inputs with no hand-crafted features, as is common in modern DL system, #2 becomes a considerably less important source of designer-caused bias. E.g. Modern image reco systems work directly from pixels, and generative models produce raw pixels 4/N
- Show replies
New conversation -
-
-
And I have made the point that the issue is one of information theory that transcends your issue of data set bias: even if I had a data set containing an image of every human who ever lived, I could never produce an accurate representation of reality from a down sampled image.
-
Define reality?
- Show replies
New conversation -
-
-
I concur, in this regard.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Who was the audience of this statement? Is it AI ethics experts, who are already familiar with this? Or lay-people, who could easily misconstrue your earlier statement as implying that the only source of bias comes from the data?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Why do you think this is clear? Have you built metrics that measure societal bias of (1) datasets (2) predictors in an 'unbiased' way and shown that this *particular* algorithm amplifies / reduces the existing bias in an information theoretically optimal way?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
I’m making the point that there is a systemic flaw in all these systems such that even if you had a data set made of the images of every human who ever lived that any ML instance would produce results that were at best amusing and at worst dangerous.
-
So GAI is a fantasy
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.