ℹ️ Authors: Kaushik Roy, Megha Chakraborty, Yuxin Zi, Manas Gaur, and Amit Sheth

📰 Preprint link: Neurosymbolic Customized and Compact CoPilots

Large Language Models (LLMs) are credible with open-domain interactions such as question answering, summarization, and explanation generation [1]. LLM reasoning is based on parametrized knowledge, and as a consequence, the models often produce absurdities and inconsistencies in outputs (eg, hallucinations and confirmation biases)[2]. In essence, they are fundamentally hard to control to prevent off-the-rails behaviors, are hard to fine-tune, customize for tailored needs, prompt effectively (due to the “tug-of-war” between external and parametric memory), and extremely resource-hungry due to the enormous size of their extensive parametric configurations [3, 4]. Thus, significant challenges arise when these models are required to perform in critical applications in domains such as healthcare and finance, that need better guarantees and in turn, need to support grounding, alignment, and instructibility. AI models for such critical applications should be customizable or tailored as appropriate for supporting user assistance in various tasks, compact to perform in real-world resource-constraint settings, and capable of controlled, robust, reliable, interpretable, and grounded reasoning (grounded in rules, guidelines, and protocols)[5]. This special session explores the development of compact, custom neurosymbolic AI models and their use through human-in-the-loop co-pilots for use in critical applications [6].

1. Wayne Xin Zhao, Kun Zhou, Junyi Li, Tianyi Tang, Xiaolei Wang, Yupeng Hou, Yingqian Min, Beichen Zhang, Junjie Zhang, Zican Dong, et al. A survey of large language models. arXiv preprint arXiv:2303.18223, 2023.
2. Vipula Rawte, Amit Sheth, and Amitava Das. A survey of hallucination in large foundation models. arXiv preprint arXiv:2309.05922, 2023.
3. Kevin Wu, Eric Wu, and James Zou. How faithful are rag models? quantifying the tug-of-war between rag and llms’ internal prior. arXiv preprint arXiv:2404.10198, 2024.
4. Zhuoran Jin, Pengfei Cao, Yubo Chen, Kang Liu, Xiaojian Jiang, Jiexin Xu, Qiuxia Li, and Jun Zhao. Tug-of-war between knowledge: Exploring and resolving knowledge conflicts in retrieval-augmented language models. arXiv preprint arXiv:2402.14409, 2024.
5. Amit Sheth, Manas Gaur, Kaushik Roy, Revathy Venkataraman, and Vedant Khandelwal. Process knowledge-infused ai: Toward user-level explainability, interpretability, and safety. IEEE Internet Computing, 2022.
6. Amit Sheth, Kaushik Roy, and Manas Gaur. Neurosymbolic artificial intelligence (why, what, and how). IEEE Intelligent Systems, 38(3):56–62, 2023.

💡 A Vision for a neurosymbolic compact and customized multi-agent copilot framework for facilitating discussions between healthcare professionals by navigating intricate information layers at school and district administration levels in a healthcare system (This example is from the MTSS system).

🔜 Comming soon ..