Bias and Discrimination
How data bias, model bias, and human bias interact to produce skewed or discriminatory AI outputs, from predictive policing to image generation.
Image: Beckett LeClair / Better Images of AI / Binarisation / CC-BY 4.0
Articles
Teaching AI Ethics: Bias and Discrimination
Algorithmic bias occurs when the data used to train AI systems reflects the biases and prejudices of society, resulting in discriminatory outputs. Large Language Models like GPT are almost unavoidably biased.
Teaching AI Ethics 2025: Bias
This instalment revisits the theme of bias in generative AI, explaining how data bias, model bias, and human bias interact to produce skewed or discriminatory outputs.
Case Studies
Predictive Policing in the US
Predictive policing algorithms used by US law enforcement disproportionately target poor, Black, and Latinx communities. The ACLU and 16 organisations issued a formal statement of concern about racial bias in these systems.
Teaching Ideas
Can AI-generated media influence democracy? Explore how synthetic content shapes political discourse.
Who is responsible for biased political information created by AI? Examine accountability for AI-generated misinformation.
How should democratic societies regulate biased technologies? Consider policy frameworks for algorithmic fairness.
How does training data shape AI bias? Examine how dataset composition influences model outputs.
How do we build fair algorithms? Explore technical approaches to reducing bias in AI systems.
Can AI make ethical decisions? Debate whether machines can be programmed to act morally.
Where does an AI model get its information from? Investigate the sources and curation of training data.
Is AI output biased? Analyse examples of bias in AI-generated text across genres.
Should AI systems be taught about cultural sensitivity? Explore how health technologies can respect diverse cultural contexts.
How can we design health technologies that reduce bias rather than amplify it? Investigate equitable approaches to health AI.
Can an AI write accurate history? Examine how AI-generated historical narratives may distort the past.
What happens when AI is trained on biased historical sources? Investigate how skewed archives shape AI understanding.
Who decides which voices are remembered in training datasets? Explore the politics of inclusion and exclusion in data curation.
How can we tell if an AI-made source is trustworthy? Develop critical evaluation frameworks for synthetic content.
What impact does AI have on our understanding of truth in civic life? Examine how AI shapes public discourse and knowledge.
How can biased data lead to unfair algorithms? Use statistical analysis to identify and measure algorithmic bias.
Can math help detect or prevent bias in AI? Explore mathematical fairness metrics and their limitations.
Should we prompt AI to represent diversity? Debate the ethics of guiding AI image generation toward inclusive outputs.
Why do some stereotypes persist despite prompt guidance? Investigate how training data encodes and reproduces cultural biases.
Can machines be objective? Explore whether algorithmic decision-making can escape human bias.
How do biases in programming shape our perception of truth? Examine how technical choices embed particular worldviews.