AI expert Professor Stephen Muggleton has argued that new developments in AI should take lessons from the basic research carried out in universities.
Stephen Muggleton, Professor of Machine Learning in Imperial’s Department of Computing, joined senior figures from government, industry and higher education last week to champion the innovative role played by universities in originating and shaping transformative technologies like artificial intelligence (AI).
Commenting on present day achievements in machine learning, Professor Muggleton told an audience at the Huawei Academic Salon: “What I see is a culmination of fundamental research going back to Turing.” Alan Turing, a founder of theoretical computer science, considered the possibility of machines that think in a seminal academic paper published in 1950.
Al computation could be done more efficiently if done locally and we could limit consumption of the world's energy resources Professor Stephen Muggleton
Professor Muggleton’s remarks echoed comments by senior figures in industry and government also speaking at the event. Ms Chen Lifang, Board Director and Senior Vice President of Huawei, the tech giant that collaborates in long-term research with 20 UK universities including Imperial and its Data Science Institute, told the audience: “AI is popular today but research started decades ago. Huawei benefits from and pioneers basic research.”
Mr Sam Gyimah MP, Minister for Higher Education at the Department for Business, Energy and Industrial Strategy, argued that partnerships between universities and industry are important for the economy. He said: “Developing the relationship between businesses and our world leading researchers, innovators and entrepreneurs is at the heart of our modern industrial strategy.”
Limitations of commercially-developed AI
Professor Muggleton challenged what he claims are ethical and engineering limitations of commercially-developed AI, and argued for a resumption of fundamental research that builds on insights generated by universities in past decades: “It’s key to understand where all this came from so we see which way things should go” he said. “In present day computation, the speed outstrips humans. Actually AIs can learn better than humans. Their difficulty is expressing what they learn.”
Professor Muggleton cited the example of DeepMind’s AlphaGo, which was the first AI to beat a champion at the ancient Chinese game of Go. Although it was an impressive performance, it did nothing to help experts in Go understand how to formulate a winning strategy, he said. “There was no explanation of how this happened.”
He also warned about the inefficiency of present day AI. “Based on current projections, within the coming decade one third of the world’s electricity will be used for data centres. This is horrific from an engineering and environmental point of view”, he said.
Professor Muggleton’s comments coincided with the launch this month of the Human Like Computing network, a collaboration between 20 UK university research groups and industrial partners, which he is jointly leading with £1.7m funding from the Engineering and Physical Sciences Research Council (EPSRC).
His work will adopt a conception of AI pioneered decades ago by the researcher Donald Michie, which stipulates that AI should generate logical procedures that humans can learn to use in their own reasoning. To achieve this, Professor Muggleton aims to develop AI that thinks more like a human.
He argued that this kind of AI is not only more interpretable, but also more efficient, since symbolic machine learning does not require data centres. “Humans typically learn from one example”, he told the Salon, adding later: “Al computation could be done more efficiently if done locally and we could limit consumption of the world's energy resources”.
AI research at Imperial
Professor Muggleton is a member of Imperial College London’s AI @ Imperial network, which brings together 70 experts from across the College’s engineering, science and business faculties to develop and deploy AI systems in a range of domains.
All photos: Huawei / Times Higher Education
Article text (excluding photos or graphics) © Imperial College London.
Photos and graphics subject to third party copyright used with permission or © Imperial College London.