Causal Reasoning in AI Models

The field of artificial intelligence is moving towards improving causal reasoning capabilities in large language models and graph neural networks. Researchers are exploring new methods to enhance the performance of these models in various tasks, such as causal mechanism identification, video-based long-form causal reasoning, and actual causality reasoning. The development of new benchmarks, such as VCRBench and AC-Bench, is facilitating the evaluation of these models and highlighting their limitations. Furthermore, studies are investigating the biases of language models in causal reasoning, including the disjunctive bias, and proposing methods to mitigate these biases. Noteworthy papers in this area include:

  • AC-Reason, which proposes a semi-formal reasoning framework for actual causality reasoning with large language models, achieving state-of-the-art performance on the AC-Bench benchmark.
  • Language Agents Mirror Human Causal Reasoning Biases, which examines the disjunctive bias in language models and proposes a test-time sampling method to reduce this bias.

Sources

LLM Enhancers for GNNs: An Analysis from the Perspective of Causal Mechanism Identification

VCRBench: Exploring Long-form Causal Reasoning Capabilities of Large Video Language Models

AC-Reason: Towards Theory-Guided Actual Causality Reasoning with Large Language Models

Language Agents Mirror Human Causal Reasoning Biases. How Can We Help Them Think Like Scientists?

Built with on top of