Researchers Reduce Bias in aI Models while Maintaining Or Improving Accuracy