Dr-Claude

Dr Claude

Inspiration

Bias and misdiagnosis are omnipresent, with each of us likely to be misdiagnosed in our lifetime. Large Language Models (LLMs) alone are powerful system 1 thinkers, but require access to evidence-backed data to give transparent and reliable output. Symbolic systems are powerful system 2 thinkers, but are hard to navigate and can have holes in their coverage.

What it does

We combine the LLMs and Symbolic systems for tackling the diagnosis problem in a realistic patient-doctor encounter setting. Claude is supplemented with a Monte Carlo Tree Search engine that searches a Disease-Symptom database to provide Claude with evidence backed suggestions for the questions to ask from the patient. Claude can swiftly handle the patient interaction while remaining grounded to a transparent diagnosis based on the given, steerable database.