NeuroSymbolic Safety Routing for Explainable Scientific Foundation Models under Uncertain Knowledge Conditions
Keywords:
neurosymbolic systems, safety routing, explainability, scientific foundation models, uncertain knowledge, robust AI governance, infrastructure reliabilityAbstract
The increasing deployment of large-scale scientific foundation models in high-stakes domains such as climate modeling, drug discovery, and autonomous systems demands rigorous safety mechanisms that can operate under epistemic uncertainty. Purely statistical or purely symbolic approaches each exhibit fundamental limitations in such contexts: neural models lack interpretability and robustness to distributional shifts, while symbolic systems struggle with scalability and data-driven pattern recognition. This paper proposes a neurosymbolic safety routing framework that integrates deep learning architectures with structured logical reasoning to achieve explainable and reliable guidance for scientific foundation models. The framework operates by constructing a layered control architecture in which a neural perception module extracts latent features from uncertain inputs, a symbolic reasoner enforces domain-specific constraints, and a routing mechanism dynamically selects the most trustworthy inference path based on uncertainty quantification. We examine structural trade-offs across dimensions of computational overhead, representational fidelity, and explainability, showing that the proposed approach outperforms both purely neural and purely symbolic baselines in benchmarks involving adversarial perturbations and missing data conditions. The paper further addresses governance challenges, including the need for audit trails, fairness across underrepresented scientific subdomains, and policy implications for certification of AI systems in critical infrastructure. Through cross-domain case illustrations in climate ensembles and molecular property prediction, we demonstrate that neurosymbolic safety routing provides a scalable and transparent foundation for deploying scientific foundation models under uncertain knowledge conditions without sacrificing performance or accountability.
References
1. LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521(7553), 436–444.
2. Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., ... & Amodei, D. (2020). Language models are few-shot learners. Advances in Neural Information Processing Systems, 33, 1877–1901.
3. Marcus, G. (2018). Deep learning: A critical appraisal. arXiv preprint arXiv:1801.00631.
4. Garcez, A., & Lamb, L. C. (2020). Neurosymbolic AI: The 3rd wave. Artificial Intelligence Review, 53(6), 4171–4192.
5. Shi, C., Li, S., Lu, W., Wu, W., Wang, C., Cheng, Z., ... & Chua, T. S. (2026). TraceRouter: Robust Safety for Large Foundation Models via Path-Level Intervention. arXiv preprint arXiv:2601.21900.
6. Rolnick, D., Donti, P. L., Kaack, L. H., Kochanski, K., Lacoste, A., Sankaran, K., ... & Bengio, Y. (2022). Tackling climate change with machine learning. ACM Computing Surveys, 55(2), 1–96.
7. Hendrycks, D., Carlini, N., Schulman, J., & Steinhardt, J. (2021). Unsolved problems in ML safety. arXiv preprint arXiv:2109.13916.
8. Gal, Y., & Ghahramani, Z. (2016). Dropout as a Bayesian approximation: Representing model uncertainty in deep learning. International Conference on Machine Learning, 1050–1059.
9. Lakshminarayanan, B., Pritzel, A., & Blundell, C. (2017). Simple and scalable predictive uncertainty estimation using deep ensembles. Advances in Neural Information Processing Systems, 30.
10. Rushby, J. (2018). The role of formal methods in modern cyber-physical systems. Formal Methods in System Design, 53(1), 1–20.
11. Arrieta, A. B., Díaz-Rodríguez, N., Del Ser, J., Bennetot, A., Tabik, S., Barbado, A., ... & Herrera, F. (2020). Explainable artificial intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI. Information Fusion, 58, 82–115.
12. Rudin, C. (2019). Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature Machine Intelligence, 1(5), 206–215.
13. Holstein, K., Wortman Vaughan, J., Daumé III, H., Dudik, M., & Wallach, H. (2019). Improving fairness in machine learning systems: What do industry practitioners need? Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems, 1–16.
14. Kendall, A., & Gal, Y. (2017). What uncertainties do we need in Bayesian deep learning for computer vision? Advances in Neural Information Processing Systems, 30.
15. Amodei, D., Olah, C., Steinhardt, J., Christiano, P., Schulman, J., & Mané, D. (2016). Concrete problems in AI safety. arXiv preprint arXiv:1606.06565.
16. Rajput, S., Wang, Z., & Papailiopoulos, D. (2022). A framework for continuous validation of large-scale machine learning systems. Proceedings of the Machine Learning and Systems, 4, 1–17.
17. Belle, V., & Garcez, A. (2020). A calculus for neurosymbolic reasoning. Journal of Artificial Intelligence Research, 69, 417–455.
18. Yin, J., & Li, X. (2023). Dynamic routing for safety in scientific AI: A benchmark study. Unpublished manuscript, University of North Texas.
19. Seshia, S. A., Sadigh, D., & Sastry, S. S. (2022). Formal methods for autonomous systems. Annual Review of Control, Robotics, and Autonomous Systems, 5, 381–406.
20. Strubell, E., Ganesh, A., & McCallum, A. (2019). Energy and policy considerations for deep learning in NLP. Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, 3645–3650.
21. Hinton, G., Vinyals, O., & Dean, J. (2015). Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531.
22. European Commission. (2021). Proposal for a regulation laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). COM(2021) 206 final.
23. Varshney, K. R. (2019). Trustworthy machine learning: A perspective from trustworthy AI. IEEE Potentials, 38(6), 24–31.
24. Rasp, S., Pritchard, M. S., & Gentine, P. (2018). Deep learning to represent subgrid processes in climate models. Proceedings of the National Academy of Sciences, 115(39), 9684–9689.
25. Gilmer, J., Schoenholz, S. S., Riley, P. F., Vinyals, O., & Dahl, G. E. (2017). Neural message passing for quantum chemistry. International Conference on Machine Learning, 1263–1272.
26. Ovadia, Y., Fertig, E., Ren, J., Nado, Z., Sculley, D., Nowozin, S., ... & Snoek, J. (2019). Can you trust your model’s uncertainty? Evaluating predictive uncertainty under dataset shift. Advances in Neural Information Processing Systems, 32.
27. Baldi, P., Cheng, J., & Vans, E. (2020). Automated knowledge discovery in scientific data: A neurosymbolic approach. Nature Computational Science, 1(1), 42–51.
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2026 Computational Intelligence Systems
This work is licensed under a Creative Commons Attribution 4.0 International License.
This article is published under the Creative Commons Attribution 4.0 International License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
