Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From Data to Commonsense Reasoning: The Use of Large Language Models for Explainable AI

Stefanie Krause, Frieder Stolzenburg

TL;DR

This work investigates whether large language models (LLMs) can perform commonsense reasoning at near-human levels and provide human-understandable explanations in QA tasks. It evaluates three models—GPT-3.5, Gemma, and Llama 3—across $11$ diverse QA benchmarks (sampling $30$ examples per dataset) and supplements the findings with a questionnaire to assess explanation quality. Results show Llama 3 achieving a mean accuracy of $90\%$ across datasets and surpassing human performance by $21\%$ on average, while GPT-3.5 and Gemma hover around the $72\%$ range; explanations from GPT-3.5 are generally rated good to excellent, with a positive link between task comprehensibility and explanation quality. The study demonstrates the potential of LLMs for zero-shot commonsense reasoning and explainable AI, while highlighting limitations (e.g., context gaps, medical knowledge, semantic relations) and the need for careful evaluation of explanations and user concerns in real-world deployments.

Abstract

Commonsense reasoning is a difficult task for a computer, but a critical skill for an artificial intelligence (AI). It can enhance the explainability of AI models by enabling them to provide intuitive and human-like explanations for their decisions. This is necessary in many areas especially in question answering (QA), which is one of the most important tasks of natural language processing (NLP). Over time, a multitude of methods have emerged for solving commonsense reasoning problems such as knowledge-based approaches using formal logic or linguistic analysis. In this paper, we investigate the effectiveness of large language models (LLMs) on different QA tasks with a focus on their abilities in reasoning and explainability. We study three LLMs: GPT-3.5, Gemma and Llama 3. We further evaluate the LLM results by means of a questionnaire. We demonstrate the ability of LLMs to reason with commonsense as the models outperform humans on different datasets. While GPT-3.5's accuracy ranges from 56% to 93% on various QA benchmarks, Llama 3 achieved a mean accuracy of 90% on all eleven datasets. Thereby Llama 3 is outperforming humans on all datasets with an average 21% higher accuracy over ten datasets. Furthermore, we can appraise that, in the sense of explainable artificial intelligence (XAI), GPT-3.5 provides good explanations for its decisions. Our questionnaire revealed that 66% of participants rated GPT-3.5's explanations as either "good" or "excellent". Taken together, these findings enrich our understanding of current LLMs and pave the way for future investigations of reasoning and explainability.

From Data to Commonsense Reasoning: The Use of Large Language Models for Explainable AI

TL;DR

This work investigates whether large language models (LLMs) can perform commonsense reasoning at near-human levels and provide human-understandable explanations in QA tasks. It evaluates three models—GPT-3.5, Gemma, and Llama 3—across diverse QA benchmarks (sampling examples per dataset) and supplements the findings with a questionnaire to assess explanation quality. Results show Llama 3 achieving a mean accuracy of across datasets and surpassing human performance by on average, while GPT-3.5 and Gemma hover around the range; explanations from GPT-3.5 are generally rated good to excellent, with a positive link between task comprehensibility and explanation quality. The study demonstrates the potential of LLMs for zero-shot commonsense reasoning and explainable AI, while highlighting limitations (e.g., context gaps, medical knowledge, semantic relations) and the need for careful evaluation of explanations and user concerns in real-world deployments.

Abstract

Commonsense reasoning is a difficult task for a computer, but a critical skill for an artificial intelligence (AI). It can enhance the explainability of AI models by enabling them to provide intuitive and human-like explanations for their decisions. This is necessary in many areas especially in question answering (QA), which is one of the most important tasks of natural language processing (NLP). Over time, a multitude of methods have emerged for solving commonsense reasoning problems such as knowledge-based approaches using formal logic or linguistic analysis. In this paper, we investigate the effectiveness of large language models (LLMs) on different QA tasks with a focus on their abilities in reasoning and explainability. We study three LLMs: GPT-3.5, Gemma and Llama 3. We further evaluate the LLM results by means of a questionnaire. We demonstrate the ability of LLMs to reason with commonsense as the models outperform humans on different datasets. While GPT-3.5's accuracy ranges from 56% to 93% on various QA benchmarks, Llama 3 achieved a mean accuracy of 90% on all eleven datasets. Thereby Llama 3 is outperforming humans on all datasets with an average 21% higher accuracy over ten datasets. Furthermore, we can appraise that, in the sense of explainable artificial intelligence (XAI), GPT-3.5 provides good explanations for its decisions. Our questionnaire revealed that 66% of participants rated GPT-3.5's explanations as either "good" or "excellent". Taken together, these findings enrich our understanding of current LLMs and pave the way for future investigations of reasoning and explainability.
Paper Structure (19 sections, 6 figures, 3 tables)

This paper contains 19 sections, 6 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Commonsense Reasoning Approaches
  4. LLMs
  5. Combining LLMs and Reasoning
  6. Evaluating LLMs on QA Tasks
  7. Benchmark Datasets
  8. LLM Results on Different Datasets
  9. Analysis of LLM Results
  10. Questionnaire
  11. Design of the Questionnaire
  12. Questionnaire Participants
  13. Questionnaire Responses
  14. Discussion and Future Directions
  15. Main Findings
  16. ...and 4 more sections

Figures (6)

  • Figure 1: Example for a invalid response from GPT-3.5 due to insufficient context information (COPA example 612).
  • Figure 2: Example of one QA task with the common structure (ARC example). Only after participants answer the first two questions the next two questions with the possible response and explanation are shown.
  • Figure 3: Comparison of accuracy of human questionnaire participants (organge) and three different LLMs: GPT-3.5, Gemma and Llama 3 (blue) on ten different datasets.
  • Figure 4: Participants’ rating of GPT-3.5's explanation quality on a 5 Likert scale from "very poor" to "excellent".
  • Figure 5: Participants believe how many explanations from zero to twenty are generated by an AI. Actually all 20 explanations were generated by GPT-3.5.
  • ...and 1 more figures