Can Large Language Models Learn Independent Causal Mechanisms?
This work addresses the robustness issue in LLMs for AI applications, but it appears incremental as it builds on existing causal concepts without a major breakthrough.
The authors tackled the problem of poor generalization in Large Language Models (LLMs) under distribution shifts by proposing a new architecture with sparsely interacting modules based on Independent Causal Mechanisms (ICMs). They showed that this approach improves out-of-distribution performance on abstract and causal reasoning tasks, though specific numerical gains are not provided.
Despite impressive performance on language modelling and complex reasoning tasks, Large Language Models (LLMs) fall short on the same tasks in uncommon settings or with distribution shifts, exhibiting a lack of generalisation ability. By contrast, systems such as causal models, that learn abstract variables and causal relationships, can demonstrate increased robustness against changes in the distribution. One reason for this success is the existence and use of Independent Causal Mechanisms (ICMs) representing high-level concepts that only sparsely interact. In this work, we apply two concepts from causality to learn ICMs within LLMs. We develop a new LLM architecture composed of multiple sparsely interacting language modelling modules. We show that such causal constraints can improve out-of-distribution performance on abstract and causal reasoning tasks. We also investigate the level of independence and domain specialisation and show that LLMs rely on pre-trained partially domain-invariant mechanisms resilient to fine-tuning.