Using AI at Work: AI in the Workplace & Generative AI for Business Leaders 94: Using AI vs Human Intelligence: When Should Leaders Trust Machines with Vasant Dhar
46 snips
Mar 9, 2026 
Vasant Dhar, NYU Stern professor and veteran AI researcher in finance and healthcare, outlines a practical trust framework for deciding when to rely on machines. He discusses predictability versus cost of error, why many pilots fail, how leaders should pick first AI projects, and why domain knowledge and data readiness matter. The conversation stresses thinking slowly while learning to think with machines.
AI Snips
Chapters
Books
Transcript
Episode notes
Trust Depends On Error Frequency And Cost
- Trust in AI depends on two axes: how often it's wrong and the consequence of errors.
- Vasant Dhar's trust map places predictability on a 0–1 axis and cost of error vertically, defining an automation frontier for when machines should be trusted.
Examples Of Automation Frontier Crossings
- Some domains have crossed the automation frontier because cost of error is low and predictability is sufficient.
- Examples: sports refereeing, urban taxis, and high-frequency trading where humans now trust machines.
Prioritize High Probability High Payoff Projects
- Do prioritize AI projects where data quality is sufficient, probability of technical success is high, and payoff is meaningful.
- Dhar recommends ranking projects by success probability and opportunity size, then multiply to prioritize.
