Raw LLM Responses

Algorithmic bias is actually really tricky to deal with. It can be mathematically proven that three notions of fairness (that would be quite reasonable to expect a fair algorithm to respect) are actually incompatible with one another. Without being overtly technical, this is the essential result from the [paper](https://arxiv.org/pdf/1609.05807.pdf): > To take one simple example, suppose we want to determine the risk that a person is a carrier for a disease X, and suppose that a higher fraction of women than men are carriers. Then our results imply that in any test designed to estimate the probability that someone is a carrier of X, at least one of the following undesirable properties must hold: (a) the test’s probability estimates are systematically skewed upward or downward for at least one gender; or (b) the test assigns a higher average risk estimate to healthy people (non-carriers) in one gender than the other; or (c) the test assigns a higher average risk estimate to carriers of the disease in one gender than the other. The point is that this trade-off among (a), (b), and (c) is not a fact about medicine; it is simply a fact about risk estimates when the base rates differ between two groups. This issue was first brought to mainstream attention by this 2016 [ProPublica article](https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing), where the risk would criminal reoffending, and we would replace "women vs men" with "blacks vs whites." Analogously, this would also apply directly to **any** decision process used to hire employees, regardless of it being done by humans or ML.