Lectures and seminars HI Conversations with Farhad Abtahi
Join us when we welcome Farhad Abtahi, senior Research Infrastructure Specialist and Manager of the SMAILE Core Facility at CLINTEC, Karolinska Institutet, and a researcher and lecturer in biomedical engineering at KTH, to HI Conversations.
Topic
The MEDLEY Paradigm: Disagreement as Evidence in Medical AI
Most medical AI returns a single confident answer, carrying the biases of the one model and dataset behind it and hiding exactly the cases where it is wrong. The MEDLEY paradigm takes the opposite stance: it treats disagreement between diverse models not as noise to be averaged away but as evidence in its own right, a signal about where a case is genuinely uncertain that is preserved and acted on rather than collapsed into a verdict. This talk traces that one principle across four settings:
- MEDLEY keeps the full spread of differentials in diagnosis instead of one answer;
- ASR-MEDLEY uses divergence between independent speech systems as a reference-free signal to flag transcription spans for review;
- Triage-MEDLEY escalates to the most severe vote when reasoning models and rule engines disagree on emergency triage; and
- MEDLEY-BENCH measures how LLM models behave under epistemic and social pressure, giving a principled basis for selecting an ensemble whose members genuinely think differently.
Register
Register to this seminar, and read more about the following events on HI Conversations.