Surrogate modeling for interpreting black-box LLMs in medical predictions
This work addresses interpretability and bias issues in LLMs for medical applications, offering a tool to flag inaccuracies and support safer deployment, though it is incremental as it applies existing surrogate modeling concepts to a new domain.
The authors tackled the problem of interpreting black-box LLMs in medical predictions by developing a surrogate modeling framework, which quantitatively revealed that LLMs encode associations contradicting medical knowledge and persistent racial biases, with proof-of-concept experiments demonstrating its effectiveness in assessing input-output perceptions.
Large language models (LLMs), trained on vast datasets, encode extensive real-world knowledge within their parameters, yet their black-box nature obscures the mechanisms and extent of this encoding. Surrogate modeling, which uses simplified models to approximate complex systems, can offer a path toward better interpretability of black-box models. We propose a surrogate modeling framework that quantitatively explains LLM-encoded knowledge. For a specific hypothesis derived from domain knowledge, this framework approximates the latent LLM knowledge space using observable elements (input-output pairs) through extensive prompting across a comprehensive range of simulated scenarios. Through proof-of-concept experiments in medical predictions, we demonstrate our framework's effectiveness in revealing the extent to which LLMs "perceive" each input variable in relation to the output. Particularly, given concerns that LLMs may perpetuate inaccuracies and societal biases embedded in their training data, our experiments using this framework quantitatively revealed both associations that contradict established medical knowledge and the persistence of scientifically refuted racial assumptions within LLM-encoded knowledge. By disclosing these issues, our framework can act as a red-flag indicator to support the safe and reliable application of these models.