7 comments

[ 2.7 ms ] story [ 26.8 ms ] thread
this has 'microsoft word is racist bc it let me type the n word' energy
yawn Tell me what isn't racist these days?
Training data based on societal norms before 1950 unquestionably would project that from straightforward semantic analysis. If Google ngram permitted the searches I am sure this would be trivial to demonstrate.

Since we're told the training sets include a lot of PD works and a lot of PD works predate the 1950s you'd want to believe the stats don't lie: the past is another country.

Prior experience in expert systems codifying med school admissions logic is, it's both sexist and racist. "Model what we do" looks at society stripped of emotion, somewhat.

Berk Breathed did a cartoon about "flesh" coloured bandages. It's not hard to find it, if you look for it.

What did we call those crayons again? Did we really do that? Yes.. we did. Yes, we did put that on jars of marmalade. Yes, we did make a principal actor in a major Hollywood film get her statue out back.

You'd have to define the model pretty hard to fencepost the underlying truth of what we are, as written culture.

The AI can reflect the bias of its creators and the training data given to it. It's also a product of the culture in which it was created.
(comment deleted)
At this point you can find an article on any topic where “X is racist”

The sun and moon, the climate, even our response to the climate. Everything is probably racist if you Google it hard enough.

"Did you train the model using real world data?"

"Yes. We want the model to be useful in real world applications."

"Then it is biased. The model is biased because data it was trained on was generated by people and people are biased. There is no such thing as an 'objective' model, just a model that is biased in a different way. The only unbiased model is an untrained model, but it is not very useful."