in conversation with AIM
summary
Explains that global and local feature attribution lets models say which inputs drove a decision — for images: beak or wing versus background color; for hiring: names or education buckets that correlate with gender — and that revealing those features enables debugging bias in models by drilling into specific values and retraining or removing spurious predictors. Argues measurable fixes: audit feature importances, add controls or reweighting, and support return-to-work upskilling so talent loss doesn’t perpetuate skewed outcomes.
AI-written summary · about
chapters (5)
- 0:00Tell Us What Your Talk Was About At Rising
- 1:35How Do You Think The Biases Can Be Avoided At The Gender Level?
- 3:05There Is A Need For More Representation For Women In The Industry. How Do You Think This Can Be
- 5:06What Are Your Thoughts On Upskilling?
- 6:04How Was Your Experience At Rising?
related talks
54mHow diversity drives excellence in our data driven tech world
30mFireside chat: Breaking the wall of unethical AI
56mGlobal Bias in AI: When Western Data Shapes the World
10mWe want to believe, but we don't
10mWhy diversity matters in Machine Learning communities
05mInterview
39mBecoming Conscious of Our Unconscious Bias in Design
34mAI and Gender: Building Fair AI Systems for the Future