Different sources of bias in AI

Update on 19th of April 2026, found on theregister.com

Initially found on google.com after search for "What are the different sources of bias in AI models?"

What are the different sources of bias in AI models?

• Training data bias

  • Historical data --> historical bias
  • Sampling --> representation bias (stereotyping, recognition, denigration, underrepresentation, ex-Nomination)
  • Training Data --> labeling / annotation bias
  • Feature selection and data curation --> unconscious Bias
  • Measurement --> measurement bias
  • Distillation bias --> In model distillation the student model cam inherit the bias from teacher model even the training data during distillation was unbiased, see more on
    https://www.theregister.com/2026/04/15/llms_inherit_bad_traits/
    https://www.nature.com/articles/s41586-026-10319-8

• Training bias --> learning bias

• Algorithmic bias

  • Model definition --> aggregation bias
  • Model Evaluation / selection --> evaluation bias
  • Model deployment --> deployment bias

• Human bias and cognitive bias

  • Human in the loop --> automation bias
  • Output skevness --> allocation bias