No. 2
Bias (Statistical)
Statistical bias: a systematic error that causes results to deviate from the true value.
Nothing about prejudice.
Nothing about fairness.
Just wrong. Consistently.
Your algorithm doesn’t know which definition you meant. Your governance should.
Which bias is your data team actually measuring?
A model can be statistically unbiased and still produce discriminatory outcomes. That’s not a contradiction – it’s the gap most AI governance frameworks don’t account for. Statistical bias and social bias aren’t the same problem. Fixing one can make the other worse.
No. 1
Bias
Bias: an inclination or prejudice for or against one person or group.
Clear enough in English. In AI governance, this single word covers at least six different problems: statistical, cognitive, social, legal, measurement and selection bias. The people in the room are rarely arguing about the same one.
Your algorithm doesn’t know which definition you meant. Your governance should.
What definition is your organisation actually governing by?
Most AI governance frameworks I’ve reviewed use “bias” without qualifying which kind. That’s not a policy – it’s a placeholder.