A system has bias if it unfairly benfits or is more positive to one gorup of people compared with another. Bias can arise for many reasons (see chap:bias) including when the training data in some way embodies human discrimination; this may happen even if the algorithm appears to be being fair. Bias is diffidult to avoid entirely as it depends on what we consider fair characetrstics, for example, job recruitment might quite reasonably use exam results of relevant subjects, so is in some ways biased against those who, for no fault of their own , have not had as good an education. What is deemed acceptable and avodiing unacceptable bias is in part a personal and corporate ethical decsion, but is also a legal one as many nations have anti-discrimination laws. Note the term bias also has a very precise statsitsical meaning relating to the whether or not the long-term output of a statistical measure of some phenomenon is equal to the true value.
Used on pages 11, 97, 98, 161, 328, 336, 487, 488, 493, 495, 497, 498, 499, 500, 501, 502, 508, 512, 514, 515, 517, 529, 530, 567, 568, 569
Also known as biasing