Machine Learning and Knowledge Engineering for Knowledge-Grounded Semantic Agents (MAKE 2026)

MAKE 2026 aims to bring together a diverse group of researchers and practitioners to explore integrating machine learning and knowledge engineering in the development of knowledge-grounded semantic agents.

As artificial intelligence enters a new decade, research is returning to one of its most fundamental paradigms—agents and multi-agent systems—to build systems that can reason, plan, and act autonomously across digital and physical environments. Recent advances in large language models, multimodal perception, and generative AI have accelerated the rise of agents capable of tool use, communication, and complex task execution.

Yet many of these capabilities rely on approximate forms of reasoning and planning that remain insufficient for dependable decision-making. Integrating machine learning with explicit semantic knowledge structures—such as ontologies, knowledge graphs, and logical reasoning frameworks—offers a path toward agents that are verifiable, explainable, and aligned with human objectives.

This symposium will provide a forum for fostering collaboration between academia and industry, addressing the challenges of building semantic agents. The goal is to semantically advance agent architectures—including LLM, VLM, VLA, LxM, and multi-agent systems—to become knowledge-grounded, trustworthy, robust, interpretable, and capable of human-aligned reasoning and decision-making across diverse applications.

MAKE 2026 • AAAI Spring Symposium Series

MAKE 2026 is part of the AAAI Spring Symposium Series, an annual set of meetings run in parallel at a common site.

The AAAI Spring Symposium Series «[…] is designed to bring colleagues together in an intimate forum while at the same time providing a significant gathering point for the AI community.

The two-and-one-half-day format of the series allows participants to devote considerably more time to feedback and discussion than typical one-day workshops. It is an ideal venue for bringing together new communities in emerging fields.» (aaai.org)

April 7-April 9, 2026 @ Hyatt Regency, San Francisco Airport | Burlingame, CA, USA

Topics

MAKE 2026 serves as a platform to shape the next generation of hybrid AI by bridging the gap between machine learning and knowledge engineering. It emphasizes grounding AI agents in explicit semantics and structured knowledge to achieve reliable, explainable, and human-aligned intelligence. Relevant topics include, but are not limited to:

• Machine Learning, Deep Learning, and Neural Networks
• Knowledge Engineering, Representation, and Reasoning
• Trustworthy, Commonsense, and Explainable AI
• Hybrid AI, Neurosymbolic, and Metacognitive AI
• Human-Centered AI, Dialogue Systems, and Conversational AI
• Generative AI and Large Language Models (LLMs)
• Semantic, Hybrid, and LLM-Based Agent Architectures
• Multi-Agent Collaboration, Communication, and Coordination
• Hybrid (Human–Artificial) Intelligence and Human-in-the-Loop AI

Format & Keynotes

The 2.5-day event will follow the traditional AAAI Spring Symposium Series schedule, with a diverse mixture of keynotes and morning and afternoon sessions.

• Peter Clark is Senior Research Director and a founding member at the Allen Institute for Artificial Intelligence (Ai2), where he also served as Interim CEO during 2022–2023. His research focuses on automated scientific discovery, knowledge representation and reasoning, and natural language understanding. He co-leads Ai2’s Asta Project, a large-scale initiative developing agentic frameworks for both assisted and automated scientific discovery. He has published over 250 papers, including five Best Paper awards, and is an ACL Fellow, Boeing Associate Technical Fellow, and Senior Member of AAAI. Furthermore, he was a founding member of the MAKE organizing committee in 2019.
  • Title: Building Discovery Machines: Systematic Search, Structured Knowledge, and the Outer Loop of Long-Horizon Research Agents
  • Abstract: LLMs can now read papers, write code, analyze datasets, and generate reports, thus performing substantial portions of the scientific workflow. Yet executing a single experiment is quite different from conducting long-horizon research. In this talk I’ll present our work on Asta, an AI research assistant that supports and partially automates scientific discovery, and aspires to scale to long-term projects. In particular, I will argue that successful long-horizon research agents require more, not less, explicit knowledge representation and systematic exploration. I’ll distinguish between the inner loop (single experiment) and the outer loop of research, where the latter involves generating, revising, and testing (via the inner loop) hypotheses, and requires representing both the evolving knowledge state and research state. I’ll illustrate and discuss two examples of the “outer loop”. First, Asta’s AutoDiscovery agent performs systematic search over a structured hypothesis space, using the notion of “LLM surprise” to find insights. Second, exploring more open-ended iterative search with Asta, we find agents can quickly meander or compound errors unless they maintain organized representations of the state of knowledge (what is currently known) and research (what is currently done). This suggests that building discovery machines is not just about building bigger models, but finding the right architecture that blends language, code, and structured knowledge representations together in an effective way.
    

Submission

Organizing Committee

Andreas Martin (primary contact, 📧), FHNW, Olten, Switzerland

Pedro A. Colon-Hernandez, Apple, Mountain View, CA, USA

Maaike de Boer, TNO, The Hague, Netherlands

Hans-Georg Fill, University of Fribourg, Switzerland

Aurona Gerber, Stellenbosch University, South Africa

Leilani H. Gilpin, University of California, Santa Cruz, CA, USA

Pascal Hitzler, Kansas State University, Manhattan, KS, USA

Jane Yung-jen Hsu, Chang Gung University, Taoyuan, Taiwan

Yen-Ling Kuo, University of Virginia, Charlottesville, VA, USA

Emanuele Laurenzi, FHNW, Olten, Switzerland

Alessandro Oltramari, Carnegie Bosch Institute, Pittsburgh, PA, USA

Thomas Schmid, Martin Luther University Halle-Wittenberg, Germany

Paulo Shakarian, Syracuse University, NY, USA

Reinhard Stolle, Fraunhofer IKS, München, Germany

Frank van Harmelen, Vrije Universiteit Amsterdam, Netherlands