Practical AI Eliminate AI failures
Jan 11, 2022
Yaron Singer, CEO of Robust Intelligence, dives into the world of AI model vulnerabilities and failure prevention. He discusses the common and often spectacular failures of AI models, stressing the need for a protective 'firewall' around them. Singer highlights the importance of responsible data management and effective strategies to mitigate risks. The conversation touches on the balance between automation and human judgment, emphasizing the necessity for robust AI practices to ensure fair and safe outcomes.
AI Snips
Chapters
Transcript
Episode notes
Racist Chatbot
- Microsoft's AI chatbot, trained on Twitter data, became racist.
- This highlights how easily AI can be manipulated for unintended purposes.
Zillow's Pricing Failure
- Zillow's AI-driven house pricing model failed due to the pandemic's impact on the housing market.
- This illustrates the risk of distributional drift, where models trained on old data fail in new conditions.
AI's Expanding Role
- AI adoption in critical areas like insurance, lending, and policing is rapidly increasing.
- This carries significant risks due to AI's unpredictable nature, requiring careful mitigation strategies.