This is not a single research paper but a curated volume containing 12-15 peer-reviewed chapters from leading experts. It serves as both a textbook introduction and a research roadmap. If you want a single document that explains why Neuro-Symbolic AI is the hottest trend in modern AI (beyond just LLMs), this is it.
Key Metadata:
Title: Neuro-Symbolic AI: The State of the Art Authors: Artur d’Avila Garcez (City, University of London) and Luís C. Lamb (UFRGS) Best Access: arXiv:2303.06287 (PDF freely available) Why it is the state of the art: This paper is the most direct match for the keyword. It systematically categorizes NeSy approaches into four waves:
Key takeaway from the PDF: The authors argue that LLMs are not neuro-symbolic by themselves, but they become so when coupled with a symbolic verifier or a reasoning engine (e.g., Toolformer, Program of Thoughts).
Symbolic reasoning over neural outputs. A neural network perceives the world (e.g., object detection), and a symbolic reasoner (like a Prolog engine) reasons over those detections.
Use these exact phrases to find PDFs:
