Künstliche Intelligenz ist nur so “smart” wie die Daten, mit denen sie trainiert wurde. Aber: Diese Daten spiegeln häufig gesellschaftliche Vorurteile.
A short educational resource explaining how bias can arise in AI systems through training data, algorithms and their use. It introduces examples of racist, sexist and class-based bias, explains debiasing approaches, and provides suggestions for addressing AI bias in education.
Why it matters for ENACT-AI
The resource directly supports understanding how AI systems can reproduce social biases and discrimination. It promotes critical evaluation of AI outputs, understanding of training data, verification of information, consideration of diverse perspectives, and reflection on stereotypes.
Learning outcomes
Learners understand what AI bias is and how it can arise, recognize the role of training data and social inequalities in AI outputs, identify potential discriminatory patterns, understand basic debiasing approaches, critically question AI-generated information, and reflect on their own assumptions and stereotypes.
Suggested activity
Ask learners to test an AI system with prompts involving different social groups, professions or characteristics. Compare the outputs, identify possible stereotypes or unequal representations, and discuss where the bias might come from. Learners then research the role of training data and develop suggestions for reducing bias, such as using diverse datasets, checking outputs for fairness, and including perspectives of affected groups.