Neuro-Symbolic AI Reproducibility Audit Finds 6.5% Reproducible

Brandon Colelough, Vladimir Martirosyan, Ishan Tamrakar, William Regli, Aditya Kumar, Anh N. Nhu, Dhruv Dubey, Raj Ambavane, Haowei Deng· August 28, 2026 View original

Key takeaways

  • Only 6.5% of neuro-symbolic AI papers are reproducible from published artifacts.
  • Missing non-code artifacts and unusable code are major barriers to reproducibility.
  • A six-stage audit framework can quantify reproducibility deficits.
  • Mandatory, versioned artifact bundles are crucial for future research integrity.

Who benefits

AI ResearchAcademiaSoftware DevelopmentData ScienceScientific Publishing

Summary

A six-stage audit framework applied to the neuro-symbolic AI (NSAI) literature found that only 6.52% of eligible studies could be fully or partially reproduced from their published artifacts. The study highlights a significant reproducibility deficit, often due to missing non-code artifacts or unusable code repositories.

This paper introduces a comprehensive six-stage framework for auditing the reproducibility of scientific claims within computer science, specifically instantiating it for the neuro-symbolic AI (NSAI) subdomain. The audit began by identifying over 5,000 records, which were then filtered down to 1,304 eligible NSAI papers. A critical finding was that 849 of these papers lacked any verifiable public code artifact. Of the 455 studies with available artifacts, only 85, or 6.52% of the total eligible corpus, could be fully or partially reproduced. The primary obstacles to reproduction were missing non-code artifacts (e.g., datasets, configurations) in 321 cases, and missing or unusable code repositories in 42 cases. These results underscore a severe reproducibility crisis in NSAI research, even when authors claim "code available." The authors advocate for mandatory, versioned, and permanently archived artifact bundles for future NSAI publications.

Why it matters

For AI researchers and practitioners, this study exposes a critical reproducibility crisis in a key AI subdomain, emphasizing the need for more rigorous artifact sharing to ensure scientific integrity and accelerate progress.

How to implement this in your domain

  1. 1Adopt a policy requiring complete, versioned, and permanently archived artifact bundles for all internal research projects.
  2. 2Implement automated checks or peer review processes to verify the reproducibility of research findings before publication or internal dissemination.
  3. 3Invest in infrastructure for securely storing and managing research artifacts, including code, datasets, and experimental configurations.
  4. 4Educate researchers on best practices for creating reproducible research, including clear documentation and dependency management.
  5. 5Prioritize research that demonstrates high reproducibility, fostering a culture of transparency and rigor.

Original post by Brandon Colelough, Vladimir Martirosyan, Ishan Tamrakar, William Regli, Aditya Kumar, Anh N. Nhu, Dhruv Dubey, Raj Ambavane, Haowei Deng

"arXiv:2608.26236v1 Announce Type: new Abstract: We present a six-stage framework for auditing the reproducibility of scientific claims across a research literature within the computer science domain, and instantiate our framework for the neuro-symbolic AI (NSAI) subdomain. Instan…"

View on X

Originally posted by Brandon Colelough, Vladimir Martirosyan, Ishan Tamrakar, William Regli, Aditya Kumar, Anh N. Nhu, Dhruv Dubey, Raj Ambavane, Haowei Deng on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research

AI ResearchAI Engineering & DevTools

Emotional Preferences Regulate Goal Priorities in Reinforcement Learning Agents

This paper proposes a computational framework where higher-level goals autonomously generate state-dependent emotional preferences to regulate the priorities of competing lower-level objectives in reinforcement learning agents. It demonstrates how this emergent preference function exhibits contextual priority switching and improves performance over fixed-preference strategies in multi-objective exploration environments.

Shiqi Liu, Yihua Tan, Hu Fu, Guanyu QiAug 28, 2026
AI Engineering & DevToolsAI Research

New Framework Unifies Task Detection and Adaptation for Continual Learning

This paper proposes FiUni, a Fisher-guided unified framework for task-free continual learning in LLMs that combines batch-level task detection with parameter-efficient adaptation. FiUni uses Fisher information matrix (FIM) properties to dynamically determine whether to reuse, expand, or create new low-rank adaptation (LoRA) subspaces, effectively mitigating catastrophic forgetting without explicit task boundaries.

Dezheng Han, Anbang Zhang, Zhihao Zhu, Shuaishuai GuoAug 28, 2026
AI Engineering & DevToolsAI Research

Soft EMG Interface Enables Machine Learning-Powered Silent Speech Recognition

This paper introduces a soft, active electromyography (EMG) interface worn on the hand that enables word-level silent speech recognition (SSR) using machine learning. The device acquires stable EMG signals from a fingertip electrode near the lips, achieving 97.2% accuracy on a 30-word vocabulary and demonstrating real-time drone control in noisy environments.

Yuta Kurotaki, Shusuke Yamakoshi, Reitaro Yoshida, Yutaka Isoda, Tamami Takano, Yuji Isano, Yusuke Miyake, Kentaro Kuribayashi, Hiroki OtaAug 28, 2026