Neuro-symbolic artificial intelligence: a survey

Summary

The article provides a comprehensive review of neuro-symbolic artificial intelligence over the past twenty years, aiming to achieve more human-like AI capabilities by combining symbolic reasoning with connectionist learning.

Why it matters

Neuro-symbolic AI aims to bridge the gap between deep learning's pattern recognition and symbolic AI's logical reasoning, addressing critical issues such as explainability, scalability, and ethics. This synthesis is crucial for advancing AI systems that can reason transparently and interact with complex real-world environments.

Highlights

• A comprehensive literature review covering books, articles, workshops, and doctoral theses from the past twenty years.
• The identification of four core aspects within neuro-symbolic AI: representation, learning, reasoning, and decision-making.
• The discussion of various applications such as question answering, robotics, computer vision, and healthcare, along with challenges like scalability and ethics.

Critical take

While the review offers a broad perspective, it falls short in providing empirical comparisons or concrete case studies that demonstrate the practical effectiveness and deployment challenges of contemporary neuro-symbolic models.