AI Dictionary
Bias
Definition
Systematic error introduced into sampling or testing by selecting or encouraging one outcome or answer over others.
Deep Dive
In the context of AI and data science, bias refers to a systematic error or predisposition that skews results in a particular direction, often leading to unfair or inaccurate outcomes. This can manifest as statistical bias, where a model consistently over- or under-predicts due to underlying assumptions or an imbalanced representation within the data, or as algorithmic bias, where societal prejudices present in the training data are inadvertently learned, amplified, and propagated by the AI system itself.
Examples & Use Cases
- 1A facial recognition system exhibiting higher error rates for individuals with darker skin tones due to underrepresentation in its training data
- 2A hiring algorithm systematically favoring male candidates for technical roles based on historical, gender-biased resume data
- 3A loan approval system disproportionately denying credit to applicants from certain socioeconomic backgrounds
Related Terms
Fairness in AIAlgorithmic TransparencyData Quality