Yes, it would be extremely bad if the statistical weight of the total corpus of training data caused a system using an LLM to make decisions about extending credit to offer worse terms (say) to women.
> sing an LLM to make decisions about extending credit to offer worse terms (say) to women.
In general, or if it isn't the correct answer?
Like: young men pay more for car insurance than young women (today). This is based on statistical models. Should they be outlawed? I think that is a very interesting question (but they aren't, today).
If the LLM was in charge, would it be wrong for it to charge young men more? Should we train that "bias" out? Or should we only train out biases that are wrong? And would that be different than how we train them today?
I don't know the answer. But I think it is less obvious than some people seem to think.
young men pay more for car insurance than young women (today). This is based on statistical models. Should they be outlawed?
EU has outlawed them. their argument is that differentiation is only valid if the difference is the actual cause and not merely statistical correlation.
Ironically, in the US it is ok to charge men more for car insurance, since they cost more in aggregate. It is illegal to charge women more for health insurance even though they cost more in aggregate.
It would obviously be very bad if those decisions were being made based on the statistical weight of the training corpus of a general large language model.
That just shows how biased you yourself are. Every human is. It is FAR more likely that the algorithm would give better credit terms to women and worse terms to men, as it is already the case with insurance. Yet you assume the opposite because of your personal biases.
At least LLMs offer a way to be tuned against that. Not that their creators would be interested in that, since the LLM's bias is exactly the mainstream opinion that they like very much.
I wasn’t assuming anything. I was asking whether the problem was bias — which we already see in some things that are highly regulated — or just wrong bias.
I’m trying to understand what people think we should correct for.
In general, or if it isn't the correct answer?
Like: young men pay more for car insurance than young women (today). This is based on statistical models. Should they be outlawed? I think that is a very interesting question (but they aren't, today).
If the LLM was in charge, would it be wrong for it to charge young men more? Should we train that "bias" out? Or should we only train out biases that are wrong? And would that be different than how we train them today?
I don't know the answer. But I think it is less obvious than some people seem to think.