LG AI Research: AAAI 2026 Demo Paper Presentations and Technical Insights

The paradigm of artificial intelligence is evolving from Passive AI, which supports simple text generation, to Agentic AI, which sets its own goals and uses tools. As agent autonomy increases, the problem of opacity in decision-making paths arises, and to address this, establishing a foundation for trusting and controlling AI judgments through Ontology (logical scaffolding) is key.

At this year's AAAI 2026, Structural Reasoning research that combines knowledge graphs (KG) with LLMs to improve reasoning performance while optimizing resource usage drew attention. In particular, practical approaches were introduced for mitigating LLM Hallucination problems and reducing operational costs.

Key research examples are as follows:

• PathMind: Through the 'Retrieve-Prioritize-Reason' paradigm, it reduces the use of unnecessary information and selects important reasoning paths, achieving SOTA performance while cutting API costs.
• DoM (Debate over Mixed Knowledge): To address situations where knowledge graphs are incomplete, this introduces a Multi-agent Debate approach. The KG agent, RAG agent, and Judge agent collaborate to produce more stable and reliable reasoning results.