Avoid mistakes building AI products
summary
Claims that 85% of AI projects fail and ten case studies show why: dataset bias makes a husky be identified by snow background, a Tesla crash traces to background pixels, Microsoft Tay becomes racist, IBM Watson worsens treatment, and an autonomous Uber kills a pedestrian. Practical fixes include a five question due diligence, POC under three weeks, choosing simpler tools like use regex not NLP when appropriate, and planning hardware, security and retraining.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 2:49What is machine learning
- 6:24Why AI projects fail
- 9:15Deep learning risks
- 12:10Avoiding AI hype
- 13:15AI failure patterns
- 16:22Data strategy for AI
- 17:25POC process setup
- 20:41Security and adversarial attacks
- 23:51AI project lessons
- 26:45Transforming for AI
- 28:33Summary checklist
- 29:36Q&A
related talks
52mSome pitfalls in AI
37mSecrets of Successful AI Projects
29mWhen AI goes wrong and how to fix it fast
49mAn AI with an Agenda
26m(Un)ethical Artificial Intelligence: How to Keep the AI Fair for Everyone
26mHow to develop useful and unbiased AI
41mWhen AI Failures Reflect Our Own Social Biases
42mIs your AI plotting to kill you? Why AI and Machine Learning is dangerous