It’s less of a bias of the programmer and moreso a bias of data, particularly when a factor like gender or ethnicity correlates with something without direct causation, such as crime rates correlating with ethnicity largely because of immigrants being poorer on average, and economic standing being a major correlating factor. If your dataset doesn’t include that, any AI will just see “oh, people in group x are way more likely to commit crimes”. This can be prevented but it’s generally more of a risk of overlooking something than intentional data manipulation (not that that isn’t possible).
It’s less of a bias of the programmer and moreso a bias of data, particularly when a factor like gender or ethnicity correlates with something without direct causation, such as crime rates correlating with ethnicity largely because of immigrants being poorer on average, and economic standing being a major correlating factor. If your dataset doesn’t include that, any AI will just see “oh, people in group x are way more likely to commit crimes”. This can be prevented but it’s generally more of a risk of overlooking something than intentional data manipulation (not that that isn’t possible).