Hacker News Reader

Comment by TearsInTheRain | Hacker News Reader

> Even amidst of world wide protests people don’t hear our voices and try to learn from us, they assume they’re experts in everything. Let us lead her and you follow. Just listen.

She says talking down to an actual expert in ML who gave an opinion on ML

View on HN

Unclear what the usage of "actual expert" means in the context of an AI ethics question, when Timnit is an "actual expert" in AI ethics.

I can see how this would get frustrating fast if your opinion is held among many Google engineers.

"Ethics" does not really encompass the topic. The topic was that a net trained mostly on white people's faces upsamples a downsampled photo of a half-black man (Barack Obama) into a generic white face. LeCun's point is this is no surprise. If you trained it on leopards, it would upsample pixelated human faces into leopard faces instead. LeCun is an expert on how these nets work.

And nowhere did Timnit disagree that the cause he identified was a cause of bias in the system. Her contention is that there are others, and reducing the problem to just "get better datatsets" is a bad idea (and a way for researchers to abdicate responsibility).

To use an example that comes up: do we solve the ethical qualms of a system that predicts incarceration rate based on facial structure by getting a better dataset? Is it even ethical to attain such a "better" dataset, what does "better" even mean in the context?

Not sure why you are being downvoted. I see how reasoned your response is to what I said. It is a clear contrast to how the AI ethics researcher engaged in the Twitter thread. Two totally different hypothetical conversations:

  A: Training data is a cause of this bias
  B: But not the only cause
  A: True

vs.

  A: Training data is a cause of this bias
  B: I am so sick of this. You are not permitted to speak.
  A: I am deleting my Twitter account.

Any support for Dr. Gebru seems to be against the HN ethos at the moment, I have some theories as to why ;)

> You are not permitted to speak.

I really don't think this is a fair reading of her tweet. I don't see any way to read her actual tweet claiming Yann can't speak at all on the subject. I do see a plea to engage more with domain experts (which to reiterate: my understanding is that Yann has a history of ignoring on the subject)

Reading these threads is really disappointing. I can in may ways already empathize with Timnit -- its exhausting seeing people who are typically bright just completely miss the point.

But if you change to a different topic like privacy then these same people really get all of this nuance that they completely seem to miss here...

Reply on news.ycombinator.com