Algorithmic Bias
AI & Emerging Tech
When an automated system treats groups of people unfairly because of what it learned from.
In practice
Algorithmic bias is systematic unfairness in a model's outputs (hiring tools that downgrade women, recognition that fails on darker skin, credit models that echo old discrimination) usually inherited from biased training data or from what the designers chose to measure. It is a design problem as much as a data one, because someone decided what the system would optimise.
Questions
What is Algorithmic Bias?
Algorithmic bias is when a system's decisions are consistently skewed against particular people, because the data or the objective it was built on carried that skew.
Why do you use Algorithmic Bias?
Because these systems scale a decision to millions of people at once. A bias that would be one person's prejudice becomes institutional the moment it is automated.
How do you use/apply Algorithmic Bias?
Audit outputs across groups before shipping, test with the people most likely to be harmed, question what the system is optimising for, and keep a human route to appeal.
When do you use Algorithmic Bias?
Before any model that affects access (to jobs, money, housing, health) goes near real people. And again after, continuously.
Who uses Algorithmic Bias?
Every designer and team shipping a model that decides something about a person.
