Societal Impact
Bias & Fairness
An AI can copy the unfairness in its data, and call it objective.
1
The mirror problem
An AI trained on the past learns the past, including its prejudices, and its output can make that bias look neutral.
2
'The algorithm decided' is never an excuse
Someone chose the data, the goal, and to deploy it, and those people are accountable for the result. A good first question: who is helped, and who is harmed?
3
Fairness is a choice
There are real, conflicting definitions of 'fair.' Engineers and communities must choose openly, and that choice is made by people.
Take it further
Full lessons on this topic
Each classroom module below is a full lesson: script, activity, slides, and an AI tutor.
AI tutor is turned off for this class
Use the CS182 lectures, notes, and assignments below to keep learning.