Bias
A systematic skew in what a system produces, inherited from the data it learned on.
Bias in an AI system is a pattern in its outputs that reflects something about its training data rather than about the world. Ask for a picture of a doctor and it will produce one that matches the demographics of the images it saw, not the ones its users are.
The sources are ordinary: the data was collected unevenly, the labels encode the annotators' assumptions, and the objective rewarded the most common pattern rather than the correct one. None of these require anyone to be prejudiced.
Bias is not fully removable and pretending otherwise is its own failure. What is achievable is measuring it on your own use case with your own data, documenting what you found, and deciding which skews you can accept for a given decision. A ranking system for loan applications and a system that summarises meeting notes do not deserve the same standard.
Related terms
Dataset
A collection of examples used to train, evaluate, or ground a system.
Data Labeling
Attaching the right answer to each example so a model can learn from it.
Diffusion Model
The architecture behind most image and video generators.
Data Governance
The rules and habits that decide what data exists, who may use it, and how long it lives.
Evaluation
Measuring whether the system does the job, on a fixed set of cases, repeatedly.
Want this working inside your team?
It starts with one conversation about your workflow.