Samsung’s push for accurate AI: The role of knowledge graphs

An Interview with OST co-founder Ian Horrocks on the future of AI development

Ian Horrocks, co-founder of Oxford Semantic Technologies (OST) and a professor at the University of Oxford, delivers a keynote speech at the Samsung AI Forum 2024, held at Samsung Electronics’ Seoul R&D Campus on Nov. 5. 2024 /Samsung Electronics

Since the release of OpenAI’s ChatGPT, generative artificial intelligence (AI) has been integrated into various devices, including smartphones and home appliances. The most widely used technologies for developing generative AI are large language models (LLMs) and machine learning.

While these advancements have greatly improved user convenience, they still have limitations, such as generating hallucinations or inaccurate answers. Oxford Semantic Technologies (OST), a British AI startup acquired by Samsung Electronics in July, is working to address these issues with its “knowledge graph” technology.

Ian Horrocks, 66, a professor at the University of Oxford and co-founder of OST, recently visited South Korea and explained about the technology. “Generative AI provides probabilistic estimates, whereas knowledge graphs deliver accurate answers based on logic,” Horrocks said in an interview.

OST was co-founded in 2017 by Horrocks and his Oxford colleagues, Boris Motik and Bernardo Cuenca Grau. Horrocks earned a Ph.D. in computer science from the University of Manchester and has been a professor at Oxford since 2007. Known as the creator of the Web Ontology Language (OWL), a standard for inference used in web technologies, Horrocks has focused on knowledge graphs since his doctoral studies.

Both LLMs and knowledge graphs share the goal of establishing logical connections between terms, but Horrocks emphasized the distinction. “Although we’ve long conducted research on the foundational technology, implementing it at the university level was challenging,” he said. “We wanted to commercialize our research outcomes and apply them directly to society, which led us to start OST.”

The core of OST’s technology lies in knowledge graphs, which organize and process data in a manner similar to human memory and recall. “The strength of knowledge graphs is their precision,” Horrocks noted. “LLMs process vast amounts of data but can’t guarantee accuracy. Moreover, they lack transparency, as we can’t see how they arrive at their answers.”

LLMs are often described as “black boxes” because their reasoning processes are not easily understandable. Knowledge graphs, on the other hand, connect verified pieces of knowledge and use rules to deduce precise answers, providing a clear and traceable reasoning path.

“The development of AI so far has focused on improving the probabilistic accuracy of LLMs, but their limitations are clear,” Horrocks said. “AI will likely evolve toward a coexistence of LLMs and knowledge graphs, as probabilistic reasoning and logical reasoning complement each other.”

[Yoo Ji-han]

- Copyrights ⓒ 조선일보 & chosun.com, 무단 전재 및 재배포 금지 -