Why AI Still Gets The Real World So Wrong

Ofir Mulla spends his days inside the uncomfortable edge cases of AI perception, where systems misread shadows as people, mistake celebration for aggression, and struggle to understand why context matters more than pixels.

As cofounder and CTO of Lumana, Ofir joins Machine Dreams to explain why most AI-driven cameras still fail at understanding real-world human behavior and what it actually takes to build systems that can learn context, adapt to environments, and reduce dangerous false alerts.

From kindergartens flagged in the middle of the night to stadium crowds misread as threats, this conversation explores the limits of visual AI, the role of historical data, the risks of blind automation, and why trustworthy physical AI requires more than a single model or a flashy demo.

If you care about safety, security, deepfakes, or how AI systems will operate in physical spaces we all share, this episode offers a rare, grounded look at what is working, what is failing, and what needs to change.