Human bias and usefulness of AI solutions

Human bias can reduce the usefulness of AI when biased assumptions are built into the problem definition, data collection, labelling or evaluation of the solution. If the training data over-represents some groups or behaviours and under-represents others, the AI may learn patterns that work for some users but give inaccurate or unfair results for others.

Bias can also affect which outputs are treated as correct. For example, people may label data using stereotypes or may ignore edge cases that do not match their own experience. The resulting AI solution may appear accurate in testing but fail in real situations, make unfair decisions, or lose user trust. A strong answer should connect these effects to the example shown in the video.