Algorithmic Bias in Depth
Examine the sources, mechanisms, and real-world impacts of algorithmic bias in AI systems.
Algorithmic bias is not an abstract technical problem — it affects hiring decisions, loan approvals, medical diagnoses, criminal sentencing, and more. Understanding its sources and mechanisms is essential for responsible AI use.
Sources of Algorithmic Bias
- Historical bias: Training data reflects past discrimination
- Representation bias: Some groups are underrepresented in training data
- Measurement bias: Proxies used to measure outcomes are themselves biased
- Aggregation bias: Models trained on aggregate data perform poorly for specific subgroups
- Deployment bias: Systems used in contexts different from those they were designed for
COMPAS, an algorithm used in US courts to predict recidivism (likelihood of reoffending), was found to be biased against Black defendants — falsely flagging them as high risk at twice the rate of white defendants. The algorithm used factors that correlated with race without explicitly using race.
Addressing Algorithmic Bias
- Diverse and representative training data
- Regular auditing for disparate impact across demographic groups
- Transparency about how algorithms make decisions
- Human oversight for high-stakes decisions
- Mechanisms for appeal and correction when algorithms cause harm
When evaluating any AI system, ask: Who is most likely to be harmed if this system makes errors? Are the people most likely to be harmed the same people who have the least power to challenge the system? These questions reveal the equity dimensions of AI deployment.
Ready to test your knowledge?
3 questions · Grade 8 level