Risk Management Show AI Cognitive Bias: Is Your AI Manipulating You?
Mar 10, 2026
Avi Pilser, founder and chief scientist at Ultra Deep Tech, builds tools to detect cognitive bias in AI. He explores how anchoring and confirmation biases drive hallucinations and manipulation. He demos real-time bias detection like Lucid and discusses governance, measurable controls, and implications of upcoming EU rules.
AI Snips
Chapters
Books
Transcript
Episode notes
Cognitive Biases Drive AI Hallucinations
- Cognitive biases in AI are distinct from social discrimination and stem from language and cognition, causing hallucinations and manipulation.
- Avi Pilser says Lucid detects 74 cognitive biases in real time, linking biases like anchoring and confirmation to AI hallucinations.
Anchoring Example Shows How AI Persuades
- Avi illustrates anchoring by saying a $20 coffee offered at $10 feels like a great deal despite being overpriced.
- He uses this simple example to show how anchoring bias transfers into AI persuasion.
Bias Transmission Via Training And RLHF
- Human biases enter AI both via RLHF and training datasets, shaping model sensibilities toward what humans prefer.
- Avi notes both post-training reinforcement and original training corpora transmit these cognitive patterns into AI behavior.
