Better Think Thrice: Learning to Reason Causally with Double Counterfactual Consistency
This addresses the challenge of evaluating and enhancing causal reasoning in LLMs for AI applications, though it is incremental as it builds on existing work on counterfactual tasks.
The paper tackles the problem of large language models' brittleness in counterfactual questions by introducing double counterfactual consistency (DCC), a lightweight inference-time method that measures and guides causal reasoning without labeled data, showing it improves performance on reasoning tasks across multiple model families.
Despite their strong performance on reasoning benchmarks, large language models (LLMs) have proven brittle when presented with counterfactual questions, suggesting weaknesses in their causal reasoning ability. While recent work has demonstrated that labeled counterfactual tasks can be useful benchmarks of LLMs' causal reasoning, producing such data at the scale required to cover the vast potential space of counterfactuals is limited. In this work, we introduce double counterfactual consistency (DCC), a lightweight inference-time method for measuring and guiding the ability of LLMs to reason causally. Without requiring labeled counterfactual data, DCC verifies a model's ability to execute two important elements of causal reasoning: causal intervention and counterfactual prediction. Using DCC, we evaluate the causal reasoning abilities of various leading LLMs across a range of reasoning tasks and interventions. Moreover, we demonstrate the effectiveness of DCC as a training-free test-time rejection sampling criterion and show that it can directly improve performance on reasoning tasks across multiple model families.