4 ms·
Large language models, which are increasingly used in AI applications, display undesirable stereotypes such as persistent associations between Muslims and viole
by wrycoder 5y ago
Large language models, which are increasingly used in AI applications, display undesirable stereotypes such as persistent associations between Muslims and violence. New approaches are needed to systematically reduce the harmful bias of language models in deployment.
IOW, we need to bias the models until they conform to socially acceptable standards.
- twofornone 5y agoI do wonder how far the activist researchers are willing to contort the science to avoid acknowledging that many stereotypes are in fact manifestations of statistically valid priors. Culture predicts behavior, and ethnicity predicts culture, because of the nature of human reproduction and cultural diffusion (or lack thereof). Note that this is not an excuse to prejudge individuals, but in this case the statistics learned by our models are not necessarily lying. These activists are putting the cart before the horse, presuming that any ML model which shows socially unacceptable bias is obviously and unarguably wrong. That's just bad, backwards science.