FLARE: Faithful Logic-Aided Reasoning and Exploration
Erik Arakelyan, Pasquale Minervini, Pat Verga, Patrick Lewis, Isabelle Augenstein
2024-10-17
Summary
This paper introduces FLARE, a new approach to improve how large language models (LLMs) reason and answer questions by ensuring that their reasoning process is clear and reliable.
What's the problem?
Current methods for question answering using LLMs often rely on techniques like Chain-of-Thought (CoT), which helps the model think step-by-step. However, these methods can produce answers that don't accurately reflect the model's actual reasoning process. This lack of clarity can lead to problems, especially when the model needs to provide trustworthy answers. Additionally, some existing methods that aim to improve accuracy require specialized training and struggle with complex or ambiguous tasks.
What's the solution?
To tackle these issues, the authors propose FLARE (Faithful Logic-Aided Reasoning and Exploration), which combines LLMs with a structured approach to reasoning. FLARE uses logic programming to break down questions into simpler parts and simulates the reasoning process step-by-step without relying on external tools. This allows the model to show how it arrived at its answer, making its reasoning more transparent and trustworthy. The authors tested FLARE on various reasoning tasks and found it performed exceptionally well, achieving state-of-the-art results on most benchmarks.
Why it matters?
This research is important because it enhances the reliability of AI systems in critical areas like healthcare and education, where accurate information is crucial. By improving how LLMs reason and ensuring their explanations are faithful to their actual thought processes, FLARE can help build trust in AI technologies and make them more effective for real-world applications.
Abstract
Modern Question Answering (QA) and Reasoning approaches based on Large Language Models (LLMs) commonly use prompting techniques, such as Chain-of-Thought (CoT), assuming the resulting generation will have a more granular exploration and reasoning over the question space and scope. However, such methods struggle with generating outputs that are faithful to the intermediate chain of reasoning produced by the model. On the other end of the spectrum, neuro-symbolic methods such as Faithful CoT (F-CoT) propose to combine LLMs with external symbolic solvers. While such approaches boast a high degree of faithfulness, they usually require a model trained for code generation and struggle with tasks that are ambiguous or hard to formalise strictly. We introduce Faithful Logic-Aided Reasoning and Exploration (\ours), a novel interpretable approach for traversing the problem space using task decompositions. We use the LLM to plan a solution, soft-formalise the query into facts and predicates using a logic programming code and simulate that code execution using an exhaustive multi-hop search over the defined space. Our method allows us to compute the faithfulness of the reasoning process w.r.t. the generated code and analyse the steps of the multi-hop search without relying on external solvers. Our methods achieve SOTA results on 7 out of 9 diverse reasoning benchmarks. We also show that model faithfulness positively correlates with overall performance and further demonstrate that {\ours} allows pinpointing the decisive factors sufficient for and leading to the correct answer with optimal reasoning during the multi-hop search.