Prof. Dr. Virach Sornlertlamvanich

Current and Future Directions: Beyond Generative Models toward Imaginative Models along the Path of Buddhist-Inspired AI

Currently, his research moves beyond error mitigation to a deeper question: how intelligence itself can be modeled. He investigates human-like reasoning, imagination as a structured and non-random process, and expert-driven intelligence, positioning SAR not merely as a technical remedy but as part of a broader conceptual framework for consciousness and awareness in authentic AI systems.

​

Semantically Aware Reasoning (SAR): A Theoretically Grounded Solution (2024 – present)

This trajectory culminated in the proposal and validation of Semantically Aware Reasoning (SAR), formally published in Knowledge and Information Systems (Springer Nature, 2025). SAR represents a significant and accepted contribution that integrates external knowledge, semantic structures, and reasoning processes into LLM-based systems. Unlike conventional RAG, SAR introduces a reasoning-aware alignment layer that constrains and guides generation according to semantically validated knowledge, demonstrably reducing hallucination and improving factual consistency. This work is widely positioned as a step toward next-generation, knowledge-grounded AI, suitable for high-stakes applications such as governance, education, healthcare, and organizational information services.

​

Large Language Models and the Hallucination Problem (2022 – 2024)

With the emergence of Large Language Models such as ChatGPT, his research entered a critical phase addressing one of the most visible limitations of LLMs: hallucination and knowledge unreliability. While LLMs demonstrated unprecedented fluency and generalization through large-scale pretraining, he identified that their probabilistic generation mechanism lacks explicit grounding in verified knowledge. His work during this period critically evaluated Retrieval-Augmented Generation (RAG), acknowledging its effectiveness while also identifying its structural weaknesses, particularly the lack of semantic reasoning and validation mechanisms.

​

Knowledge-Centric AI and Socially Grounded Intelligence (2018 – 2022)

As AI systems increasingly entered public and social domains, his research emphasis shifted toward trustworthy and socially grounded AI. He examined how AI systems interact with collective intelligence, social data, and institutional knowledge, highlighting risks related to bias, inconsistency, and loss of accountability. This phase was marked by broader acceptance of his view that AI must align with human knowledge structures and social norms, anticipating concerns that later became central in AI ethics, responsible AI, and public-sector deployment.

​

Integration of Knowledge with Machine Learning (2010 – 2018)

With the rise of machine learning and early deep learning, his research addressed how explicit knowledge representations could coexist with statistical models. Rather than rejecting symbolic knowledge, he explored hybrid approaches that integrated semantic relations, ontologies, and domain knowledge into learning-based systems. This work was significant in maintaining interpretability, consistency, and cross-lingual robustness, particularly in applied domains such as information extraction, social computing, healthcare, and elderly care systems. His contributions during this phase were accepted as practical demonstrations that purely data-driven approaches were insufficient for knowledge-intensive and socially sensitive tasks.​​

​

Knowledge Resources and Multilingual Semantics (2000 – 2010)

As the field shifted toward data-driven methods, his research evolved to address a critical gap: the lack of structured semantic resources for non-English languages. He became widely recognized for contributions to lexical resources, WordNet construction, semantic relations, and multilingual knowledge alignment, especially for Thai and other Asian languages. This phase marked his strong influence in knowledge engineering for NLP, where his work was accepted and used as national and regional language infrastructure, supporting research, education, and downstream AI applications. The significance of this period lies in demonstrating that language knowledge is not optional but foundational, even in statistically driven systems.

​

Foundations in Language Technology and Linguistic Knowledge (1990 –  2000)

Virach Sornlertlamvanich’s research began with a strong grounding in computational linguistics and language technology, at a time when NLP systems relied heavily on explicit linguistic rules and symbolic representations. His early work focused on morphology, syntax, lexical semantics, and corpus-based analysis, particularly for Asian languages, which were underrepresented in mainstream NLP research. During this period, his work was significant in establishing linguistically principled resources that could be computationally operationalized, contributing to the acceptance of language technology as a viable infrastructure for multilingual information processing.​​