From Priors to Predictions: Explaining and Visualizing Human Reasoning in a Graph Neural Network Framework

2025-12-19

AI Summary

Human abstract reasoning from minimal samples relies on informal inductive biases whose formal characterization remains elusive. Method: We propose the first explicit, differentiable graph-structured prior encoding human inductive biases, instantiated within a graph neural network (GNN)-based prior-driven reasoning framework. The framework learns task-specific structural priors via graph topology search and employs key subgraph visualization and computational graph attribution to uncover individual problem-solving differences and error origins. Contribution/Results: Systematic ablation studies and evaluations on the Abstraction and Reasoning Corpus (ARC) demonstrate that our model not only replicates human behavioral patterns but also localizes human-like errors attributable to flawed priors. It achieves substantial improvements in out-of-distribution generalization, interpretability, and human–model alignment—bridging cognitive modeling and deep learning through structured, learnable inductive bias.

Technology Category

• Reasoning under Uncertainty: Graphical Models
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Humans excel at solving novel reasoning problems from minimal exposure, guided by inductive biases, assumptions about which entities and relationships matter. Yet the computational form of these biases and their neural implementation remain poorly understood. We introduce a framework that combines Graph Theory and Graph Neural Networks (GNNs) to formalize inductive biases as explicit, manipulable priors over structure and abstraction. Using a human behavioral dataset adapted from the Abstraction and Reasoning Corpus (ARC), we show that differences in graph-based priors can explain individual differences in human solutions. Our method includes an optimization pipeline that searches over graph configurations, varying edge connectivity and node abstraction, and a visualization approach that identifies the computational graph, the subset of nodes and edges most critical to a model's prediction. Systematic ablation reveals how generalization depends on specific prior structures and internal processing, exposing why human like errors emerge from incorrect or incomplete priors. This work provides a principled, interpretable framework for modeling the representational assumptions and computational dynamics underlying generalization, offering new insights into human reasoning and a foundation for more human aligned AI systems.

Problem

• Formalizing inductive biases as explicit priors in graph neural networks
• Explaining individual differences in human reasoning using graph-based priors
• Visualizing critical computational graph structures for interpretable generalization analysis

Innovation

• Formalizes inductive biases as explicit graph-based priors
• Optimizes graph configurations for connectivity and abstraction
• Visualizes critical computational graphs to interpret predictions