🧠 Causal Reasoning: Understanding cause-and-effect relationships in simulation results PHASE 1 #3
Description
Idea
If AI agents are to support the NHS make decisions, we want agents to be able to understand the physical world and the consequences and risks of decisions. Using a human validated digital (offline) model is a first step towards this idea. Effectively we want an agent, via a LLM, to be able to explore the different experiments and synthesis the the results.
Proposal
To be able to reason about cause and effect we want the agent to be able to run experiments, and reflect and learn from outcomes. A first step could be have an analysis agent that is provided with results in a base case versus a simulated "what-if" experiment. It could report results and trade-offs to a human user.
Extension: sensitivity analysis could be use to assess risk in the decision.
@AliHarp any thoughts?