BIRD: A Trustworthy Bayesian Inference Framework for Large Language Models

2024-04-18

arXiv.org

AI Summary

Current large language models (LLMs) struggle to produce reliable probabilistic estimates under partial observability, limiting their trustworthy deployment in high-stakes decision-making and planning tasks. To address this, we propose a novel hybrid reasoning framework that synergistically integrates LLM-based causal abduction with Bayesian networks. First, prompt engineering elicits the LLM’s ability to automatically extract causal factors and infer conditional dependencies among variables. Second, these extracted relationships are formalized into a structured Bayesian network, enabling rigorous probabilistic inference and posterior calibration. This work pioneers the end-to-end coupling of LLM-driven inductive causal reasoning with Bayesian deductive inference for principled probability calibration. Experimental results demonstrate that our framework improves probabilistic estimation accuracy by 30% over LLM-only baselines, while substantially enhancing both decision reliability and interpretability.

Technology Category

• Reasoning under Uncertainty: Probabilistic Inference
• Machine Learning: Calibration & Uncertainty Quantification
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Predictive models often need to work with incomplete information in real-world tasks. Consequently, they must provide reliable probability or confidence estimation, especially in large-scale decision-making and planning tasks. Current large language models (LLMs) are insufficient for accurate estimations, but they can generate relevant factors that may affect the probabilities, produce coarse-grained probabilities when the information is more complete, and help determine which factors are relevant to specific downstream contexts. In this paper, we make use of these capabilities of LLMs to provide a significantly more accurate probabilistic estimation. We propose BIRD, a novel probabilistic inference framework that aligns a Bayesian network with LLM abductions and then estimates more accurate probabilities in a deduction step. We show BIRD provides reliable probability estimations that are 30% better than those provided directly by LLM baselines. These estimates further contribute to better and more trustworthy decision making.

Problem

• Improving probability estimation in incomplete information scenarios
• Enhancing trustworthiness of large language models for decision-making
• Aligning Bayesian networks with LLM outputs for accurate deductions

Innovation

• Bayesian network aligns with LLM abductions
• Deduction step enhances probability accuracy
• 30% better than baseline LLM estimations