<aside> 💬

"Only one simple question: Who does this answer to?"

</aside>

Thursday October 29, 2026 · 2:05 PM · Talk

Key ideas from the talk

• He said 230 million people a week ask OpenAI for health advice. People already trust AI by default.
• Medical authority has moved from the lone doctor, to the team, to evidence, and now to an AI in the patient's pocket. Nobody is accountable for that answer.
• Research with Northeastern University in Vancouver and the University of Victoria uses an LLM tribunal. Three models judge each health answer on precision, accuracy, privacy and ethics.
• His platform keeps data at home, has a crisis level that brings in a doctor, and denies insurers access.
• He cited a US case where an insurer used AI to cancel a cancer patient's coverage.
• In the Q&A, Simon Haworth argued for a moral duty to share data. Dean said sharing needs consent, de-identification and research-only use.

Quotes

• "AI is smart, but where is accountability?"
• "When it comes to healthcare, it's all about n equals one, which means you as an individual and your health."
• "If AI and a doctor disagree, who is right?"
• "There's a shift of authority, where AI is taking on the authority for evidence in medical."
• "There's no doctor to blame. There's no team with evidence to blame. You took that advice. Well, it's your own fault."
• "If it can't explain it, it shouldn't speak."
• "Privacy consent must be asked for people."

About the speaker

Dean Shev · CTO and co-founder, VHT.ai