Better Behavioral Prediction, More Faithful Model Ablations? Evidence from Sequential Choice

2026-09-28

AI Summary

This study addresses the limitation that models with high predictive accuracy in sequential choice tasks often yield ablation results that fail to faithfully reflect underlying cognitive generative mechanisms. Using a synthetic multi-armed bandit task, this work compares the predictive and ablation responses of neural and cognitive models—including GRUs, Transformers, fine-tuned LLaMA, and reinforcement learning agents—thereby disentangling predictive sufficiency from ablation fidelity for the first time. The findings reveal the independence of these two properties, demonstrating that highly predictive models may exhibit substantially weaker responses to input removal than known ground-truth generators, with inconsistent model rankings across tasks. This research underscores the necessity of independently validating ablation responses when inferring behavioral generative mechanisms.

Abstract

Using predictive models to explain cognition requires more than accurate behavioral predictions. Input ablations offer an appealing route: remove information from a model and interpret the resulting performance change as evidence of its importance for behavior. Yet this inference assumes that the model's dependence on information reflects the dependence of the process generating the behavior. We test it in two synthetic sequential bandit tasks with known generating policies, where past choices can remain informative when feedback is unavailable to a predictor. We compare GRUs and Transformers trained from scratch, a fine-tuned LLaMA model, and cognitive models across systematically varied reward contributions. Our analyses distinguish prediction after training without reward observations from the response of a fixed predictor to donor-reward replacement. Three findings emerge. First, in the restless task, neural models trained without rewards predict held-out choices better than four simple training-fitted behavioral baselines. Second, under matched donor replacement, accurate predictors can respond much less than the known generator. Third, at some reward weights, neural networks predict better than a pooled reinforcement-learning model but have less faithful changes in choice probabilities; the model ordering differs between the two tasks. These independent-test results separate information sufficient for prediction from response fidelity under a specified ablation in sequential choice. They motivate validating model-ablation responses independently of predictive performance before using them to infer how the observed behavior was generated.

Problem

• behavioral prediction
• model ablation
• faithfulness
• sequential choice
• cognitive modeling

Innovation

• model ablation
• sequential choice
• predictive fidelity
• restless bandit
• cognitive modeling