Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Neurosymbolic AI approach to Attribution in Large Language Models

Deepa Tilwani, Revathy Venkataramanan, Amit P. Sheth

TL;DR

NesyAI frameworks can enhance existing attribution models, offering more reliable, interpretable, and adaptable systems for LLMs by integrating neurosymbolic AI, which combines the strengths of neural networks, with structured symbolic reasoning.

Abstract

Attribution in large language models (LLMs) remains a significant challenge, particularly in ensuring the factual accuracy and reliability of the generated outputs. Current methods for citation or attribution, such as those employed by tools like Perplexity.ai and Bing Search-integrated LLMs, attempt to ground responses by providing real-time search results and citations. However, so far, these approaches suffer from issues such as hallucinations, biases, surface-level relevance matching, and the complexity of managing vast, unfiltered knowledge sources. While tools like Perplexity.ai dynamically integrate web-based information and citations, they often rely on inconsistent sources such as blog posts or unreliable sources, which limits their overall reliability. We present that these challenges can be mitigated by integrating Neurosymbolic AI (NesyAI), which combines the strengths of neural networks with structured symbolic reasoning. NesyAI offers transparent, interpretable, and dynamic reasoning processes, addressing the limitations of current attribution methods by incorporating structured symbolic knowledge with flexible, neural-based learning. This paper explores how NesyAI frameworks can enhance existing attribution models, offering more reliable, interpretable, and adaptable systems for LLMs.

Neurosymbolic AI approach to Attribution in Large Language Models

TL;DR

NesyAI frameworks can enhance existing attribution models, offering more reliable, interpretable, and adaptable systems for LLMs by integrating neurosymbolic AI, which combines the strengths of neural networks, with structured symbolic reasoning.

Abstract

Attribution in large language models (LLMs) remains a significant challenge, particularly in ensuring the factual accuracy and reliability of the generated outputs. Current methods for citation or attribution, such as those employed by tools like Perplexity.ai and Bing Search-integrated LLMs, attempt to ground responses by providing real-time search results and citations. However, so far, these approaches suffer from issues such as hallucinations, biases, surface-level relevance matching, and the complexity of managing vast, unfiltered knowledge sources. While tools like Perplexity.ai dynamically integrate web-based information and citations, they often rely on inconsistent sources such as blog posts or unreliable sources, which limits their overall reliability. We present that these challenges can be mitigated by integrating Neurosymbolic AI (NesyAI), which combines the strengths of neural networks with structured symbolic reasoning. NesyAI offers transparent, interpretable, and dynamic reasoning processes, addressing the limitations of current attribution methods by incorporating structured symbolic knowledge with flexible, neural-based learning. This paper explores how NesyAI frameworks can enhance existing attribution models, offering more reliable, interpretable, and adaptable systems for LLMs.
Paper Structure (11 sections, 2 equations, 3 figures, 1 table)

This paper contains 11 sections, 2 equations, 3 figures, 1 table.

Table of Contents

  1. Neurosymbolic AI as a Framework for Attribution
  2. How Neurosymbolic AI Solves the Attribution Problem?
  3. Structured Knowledge Representation
  4. Memory Structures and Temporal Reasoning
  5. Metacognitive Layers for Enhanced Attribution
  6. Explainability and Traceability Through Symbolic AI
  7. Dynamic Knowledge Update for Continual Learning
  8. Challenges and Limitations
  9. Current Progress and Future Directions
  10. Acknowledgments
  11. Authors

Figures (3)

  • Figure 1: A lawyer drafting a legal brief with the help of AI is misled into citing a fabricated case, "Doe v. State (2021)." The error is later discovered, damaging the lawyer's credibility and jeopardizing the case (for visualization purposes, not a real case)
  • Figure 2: State-of-the-art LLMs could not handle attribution reliably and accurately. This figure demonstrates how an AI agent, integrated with a knowledge graph, systematically retrieves relevant academic papers on a given topic, filters by time and venue, and ranks by citation count, ensuring the accuracy and reliability of the sources provided to the user.
  • Figure 3: An analysis of the recipe french fries in the context of diabetes with explainability and traceability. Explanations are provided in the form of reasoning. Given a recipe is deemed unsuitable, alternative suggestions are provided as a similar recipe or revising the existing recipe. Given a recipe is unsuitable, the alternative suggestions provided with explanations are called counterfactual reasoning. Explaining why an alternative recipe is similar to a given recipe refers to analogical reasoning.