How to develop useful and unbiased AI
summary
Describes concrete failures and fixes: Rotterdam used an Accenture model that cut welfare support for young mothers; Alexa under‑recognizes non‑male native English voices; an e‑shop’s recommender locks on one click for washing machines. Recommends pragmatic steps: use a data product canvas, score ideas with a business scorecard, prepare data (add diverse and synthetic samples), run multiple models and debias with pre/post‑processing. Claims measurable gains — up to 40% efficiency, 25% downtime reduction and 50% customer satisfaction in cited examples.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 1:14Algorithm exercise
- 3:02Biased AI in welfare
- 4:46Consumer AI biases
- 7:00Introduction
- 7:43Defining AI
- 9:49AI for business
- 13:05Steps to useful AI
- 16:19Prioritizing AI ideas
- 17:24Preparing AI data
- 19:28How bias enters AI
- 20:40Sources of bias
- 22:46Mitigating AI bias
- 24:32Bias tools
- 24:53AI success tips
related talks
49mAn AI with an Agenda
31mAvoid mistakes building AI products
26m(Un)ethical Artificial Intelligence: How to Keep the AI Fair for Everyone
56mAn AI with an Agenda: How Our Cognitive Biases Leak Into Machine Learning
47mFight AI Bias With… Bias
23mDelivering business success with AI
27mHow can we use AI to bring new value to the business
54mEthical Machine Learning