Andrii IURLOV - stock.adobe.comSamsungs on-device AI: A Computer Weekly Downtime Upload podcastByCliff Saran, Managing EditorListen to this podcastWe speak to the co-founder of Oxford Semantic Technologies, which has developed the AI for the new Samsung Galaxy S25PodcastA few weeks ago, Ian Horrocks took to the stage at an event which featured DeepMinds co-founder, Demis Hassabis, to help Samsung launch its latest phone, the Galaxy S25.He says: It was pretty amazing for me as the sort of humble academic, finding myself in that setting. But it was of course great as well and a culmination of many, many years of research.Horrocks is co-founder of Oxford Semantic Technologies, a startup, spun out of Oxford University, which has developed a knowledge graph AI system called RDFox, that powers the artificial intelligence (AI ) in the latest Samsung Galaxy S25.Samsung acquired Oxford Semantic Technologies in July 2024 and the Galaxy S25 is the first new smartphone from the manufacturer to include AI post acquisition.Explaining the work with Samsung, Horrocks says: One of the reasons why Samsung was so excited about our knowledge graph system is the fact that it can actually run on the phone. You can build it with a relatively small footprint and relatively small compute requirement.Among the benefits of using on-device AI is, as Horrocks points out is: You do need to move potentially sensitive personal data off into the cloud. You can do everything on your own device, so you're in control. The AI on the phone cant use what it can't see and isn't sharing your sensitive personal data.Computer Weekly spoke to Horrocks the week the world of AI was disrupted by Chinas DeepSeek. While there are many questions over how it was developed, he says: I still think it's very interesting how it challenges the orthodox view that generative AI is just all about compute power for training and inference.Horrocks field of expertise is knowledge representation and reasoning. During the time he has worked in this area, Horrocks says: I soon learned there that the impact of algorithms and optimisations can vastly outstrip the impact of better hardware.Among the challenges of knowledge-based AI is what Horrocks calls the scalability of logical reasoning. I was working with a brilliant colleague at Oxford called Boris Motik, who had this idea that to address this problem using a combination of modern computer architecture with some very clever, novel data structures and algorithms.Horrocks says Motiks approach did not focus just on the computational power available in new hardware. He also had some brilliant algorithmic ideas and demonstrated that this could produce a really scalable graph knowledge-based and reasoning system.After joining forces with Motik, the pair faced the challenge of moving from academic research to the next level. He says: We really needed to start a company that could provide a more robust implementations with all of the kind of bells and whistles thatreal industry users require so we joined forces with another colleague, Bernardo Cuenca Grau, and with some more entrepreneurial people who knew how to set up and run a company.For Horrocks, DeepSeek effectively shows the world that there are other ways to achieve results with AI, which do not need to follow-up the same path as the current batch of large language models. I was always a bit sceptical about the idea that we basically know how to do AI. The problem has been sort solved and all we need to do is crank up the hardware and invest billions in equipment, he adds.