AI systems really do absorb and amplify biases from their training data. Hiring tools that discriminated against women. Healthcare algorithms that deprioriti...
AI systems really do absorb and amplify biases from their training data. [Hiring tools that discriminated against women](https://www.reuters.com/article/us-amazon-com-jobs-automation-insight-idUSKCN1MK08G). [Healthcare algorithms that deprioritized Black patients](https://www.science.org/doi/10.1126/science.aax2342). [Resume screeners that never — not once — preferred Black male-associated names](https://voxdev.org/topic/labour-markets-migration/racial-and-gender-biases-llm-based-evaluations). Case after case. The word *inherent* is where it gets complicated. The bias originates with us — our history, our data, our decisions — not with the technology itself. Which means it can be measured, studied, and reduced. Not easily. But it's not baked into the math the way "inherent" implies.
Verdict: The concern is real and well-documented. "Inherent" overstates it. AI bias has been demonstrated in [hiring](https://www.reuters.com/article/us-amazon-com-jobs-automation-insight-idUSKCN1MK08G), [healthcare](https://www.science.org/doi/10.1126/science.aax2342), [criminal justice](https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing), and [resume screening](https://voxdev.org/topic/labour-markets-migration/racial-and-gender-biases-llm-based-evaluations) — case after case. Anyone downplaying this hasn't looked at the evidence. But "inherent" implies it can't be fixed, and that's misleading. Bias can be measured, tested for, and reduced — even if it's rarely eliminated completely. The practical stance: assume it's present until proven otherwise. Test systematically. Demand transparency about what tradeoffs were made. The technology reflects us.. what we do about that reflection is a choice.