An automated hiring tool failed because its training data, based on a male-dominated workforce, taught it to penalize female candidates.
The Gender Shades project revealed that commercial facial recognition systems have significantly higher error rates for dark-skinned women, leading to misidentification.
Algorithms used to predict student exam scores during the pandemic unfairly downgraded high-achieving students from historically underperforming schools.
Algorithmic bias leads to real-world discrimination, and government regulations similar to GDPR may be necessary to ensure ethical AI development.
The first steps to mitigate bias are acknowledging its existence, performing exploratory data analysis, and incorporating fairness into the design phase.
The European Union's guidelines for trustworthy AI provide a seven-point framework covering human oversight, transparency, fairness, and accountability.
Diverse teams with varied backgrounds and perspectives are crucial for challenging assumptions, fostering innovation, and identifying potential biases early.
A summary of how biased algorithms cause discrimination and the four key strategies to prevent it: analyzing data, designing for fairness, using ethical frameworks, and building diverse teams.
58:06Anna Fritsch-Weninger
28:05Toju Duke
47:35Laurie Voss, Rey Bango, Hannah Foxwell, Rizel Scarlett & Thomas Steiner
23:50Aditi Godbole
30:49Marin Niehues
26:12Ayoub Alouane
27:26Cassie Kozyrkov
35:01Angie Jones, Kent C Dobbs, Liran Tal & Chris Heilmann
Jobs that call for the skills explored in this talk.
Avelios Medical