Bringing to Light: The Challenges of Representing and Reasoning Common Sense Knowledge in AI Systems

Date

2024-12

Author

Kabadere, Zeynep

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

186
views

341
downloads

Cite This

This thesis, firstly, investigates the challenges of imitating common sense reasoning in artificial intelligence (AI) by focusing on three core issues: representing common sense knowledge, identifying tacit knowledge, and addressing the frame problem. In the first chapter, the study examines these challenges through the lens of knowledge representation, reasoning, and learning processes, highlighting their significance in enhancing AI's ability to handle everyday reasoning tasks. In the second chapter, the thesis presents a comprehensive evaluation of two Large Language Model (LLM)-based AI systems, ChatGPT 4.o and Claude Sonnet 3.5, to assess their capacity to simulate common sense reasoning. This evaluation is structured around six primary benchmarks: context-based information integration, future planning and adaptation ability, comprehensive causality and linked information management, operational execution competence, background knowledge integration and application, and accuracy and relevance management. These benchmarks are further refined into 27 detailed sub-benchmarks designed to address the challenges identified in the first chapter comprehensively. By analyzing the experimental results, the thesis identifies the strengths and limitations of both models in imitating common sense reasoning. The findings contribute to the broader understanding of AI's capabilities and limitations in replicating common sense reasoning, providing insights into areas requiring further development. Ultimately, this study bridges philosophical inquiry and empirical evaluation to offer a robust framework for advancing the design of contextually aware and reasoning-capable AI systems.

Subject Keywords

common sense reasoning, artificial intelligence, large language models, benchmarking analysis, knowledge representation

URI

https://hdl.handle.net/11511/112703

Collections

Graduate School of Social Sciences, Thesis