Causally-Enhanced Reinforcement Policy Optimization
This addresses the issue of reward hacking and unfaithful reasoning in LLMs for AI safety and reliability, though it is an incremental improvement as it builds on existing policy optimization methods.
The paper tackles the problem of large language models (LLMs) producing superficially correct answers with unfaithful reasoning, degrading under causal perturbations, by introducing Causally-Enhanced Policy Optimization (CE-PO), a reward-shaping framework that improves accuracy by 5.49% on average and enhances robustness to causal stress tests.
Large language models (LLMs) trained with reinforcement objectives often achieve superficially correct answers via shortcut strategies, pairing correct outputs with spurious or unfaithful reasoning and degrading under small causal perturbations. We introduce Causally-Enhanced Policy Optimization (CE-PO), a drop-in reward-shaping framework that augments policy optimization with a differentiable proxy for causal coherence along the generation pathway from prompt (Z) to rationale (X) to answer (Y). CE-PO estimates model-internal influence with Jacobian-based sensitivities, counterfactually hardens these signals to suppress nuisance cues, and fuses the resulting coherence score with task-accuracy feedback via a Minkowski (power-mean) combiner, exposing a single tunable between accuracy and coherence trade-off. The unified reward integrates with PPO/GRPO without architectural changes. Across reasoning benchmarks and causal stress tests, CE-PO reduces reward hacking and unfaithful chain-of-thought while improving robustness to correlation-causation flips and light counterfactual edits, all at near-parity accuracy. Experimental results across 4 datasets show that CE-PO improves accuracy over baselines by 5.49% on average (up to 9.58%), while improving robustness to correlation-causation flips and light counterfactual edits.