4 ms·

We have a prominent AI researcher being ostracized for a tweet (and presumably a more general attitude we don't see out of context) that marginalizes people of

We have a prominent AI researcher being ostracized for a tweet (and presumably a more general attitude we don't see out of context) that marginalizes people of color, reflecting a larger problem of systemic racism, particularly in academia. Neither his attitude nor the reaction to it seem productive in my mind. Online forums have always been prone to everyone shouting over everyone else until the loudest voice wins.

dlivingston 6y ago
Can you explain how his tweet “marginalizes people of color”? His position, from what I understand, is that the ML algorithm which reconstructed a photo of Obama as white was biased solely because the training data set consisted of mostly white people. If it had been a more diverse data set then the algorithm would have behaved more appropriately.
- flatline 6y ago
  The complaint is that he was being dismissive of racism in the field by saying it's just the training data. Technically correct, but missing the larger picture - who is generating this training data? Doesn't his comment subtly attempt dismiss the issue entirely? Deployed ML systems are notorious for these kinds of biases. He was offloading the criticism to engineers building these data sets, as though he could not have any possible role to play as a researcher. He was thus shot down for a seemingly innocuous comment that is set against a very large backdrop of racism that tends to get ignored. Haven't you read White Fragility three times over by now?? /s On the one hand this guy was pushed off twitter for a seemingly innocuous comment - one that some people just happened to take as insensitive. The type of comment that, in fact, increasingly large numbers of people seek out to take offense at - so it seems like an overreaction. On the other hand this is twitter, exactly where these types of narratives play out, he was unable to defend his position competently or even acknowledge the criticisms against him, gave up, and walked off.
  - sillysaurusx 6y ago
    I think I see the disconnect then: ML researchers are often the people who build the datasets. Datasets aren't something that grow in the wild, nor is it usually a task for ML engineers. So the original tweet wasn't really offloading the problem; it was self-criticism (though it's understandable it didn't sound like it).

dlivingston 6y ago

Can you explain how his tweet “marginalizes people of color”? His position, from what I understand, is that the ML algorithm which reconstructed a photo of Obama as white was biased solely because the training data set consisted of mostly white people. If it had been a more diverse data set then the algorithm would have behaved more appropriately.

flatline 6y ago
The complaint is that he was being dismissive of racism in the field by saying it's just the training data. Technically correct, but missing the larger picture - who is generating this training data? Doesn't his comment subtly attempt dismiss the issue entirely? Deployed ML systems are notorious for these kinds of biases. He was offloading the criticism to engineers building these data sets, as though he could not have any possible role to play as a researcher. He was thus shot down for a seemingly innocuous comment that is set against a very large backdrop of racism that tends to get ignored. Haven't you read White Fragility three times over by now?? /s On the one hand this guy was pushed off twitter for a seemingly innocuous comment - one that some people just happened to take as insensitive. The type of comment that, in fact, increasingly large numbers of people seek out to take offense at - so it seems like an overreaction. On the other hand this is twitter, exactly where these types of narratives play out, he was unable to defend his position competently or even acknowledge the criticisms against him, gave up, and walked off.
- sillysaurusx 6y ago
  I think I see the disconnect then: ML researchers are often the people who build the datasets. Datasets aren't something that grow in the wild, nor is it usually a task for ML engineers. So the original tweet wasn't really offloading the problem; it was self-criticism (though it's understandable it didn't sound like it).