Professor Alessio Lomuscio has been awarded one of ten highly prestigious Chairs in Emerging Technology from the Royal Academy of Engineering.
The award (£1.3m over ten years) will enable Alessio to develop verification techniques for autonomous systems and AI.
From self-driving cars to robotic factory workers, society is now primed for the mass introduction of a wide range of systems based on artificial intelligence, while machine learning techniques are already used to underpin the deployment of many applications in banking, healthcare and beyond.
The potential benefits of the technology are clear, but fears remain about perceived dangers and risks, and the consequences of malfunctions are growing as the tasks delegated to machines become more powerful.
While progress has been made in the verification of traditionally programmed, standalone autonomous systems, the recent wide adoption of machine learning techniques hinders the applicability of these results to current and future systems, many of which will be entirely impossible to verify.
As verification is vital for enabling safe deployment and widespread acceptance, a regulatory policy is required to certify autonomous systems and AI systems more generally. Unless solved, these challenges will become a barrier for adoption of the technology.
The Chair in Emerging Technologies will focus on this goal, by developing underlying mathematical methods, as well as building toolkits, that can demonstrate the safety and reliability of forthcoming AI systems - including autonomous and robotic systems - built from machine learning techniques.
Speaking of the award, Professor Lomuscio said: "I am thrilled that I will be able to dedicate considerable amounts of time to contribute to these technical challenges.
"I feel this is the question of the decade and next in Artificial Intelligence and Computer Science as a whole. The underlying problems are challenging; but this is what excites us in science."
Looking ahead to his research programme, Professor Lomuscio added: “I will get started by investigating how we can give guarantees about the behaviours of simple neural systems based on Feed Forward and Convolutional Neural Networks based on a variety of activation functions. From there I will move to more complex systems governed by large and recurrent networks.
"These are challenging questions; we will need to develop the underlying methods and evaluate their potential effectiveness."
Professor Nick Jennings (Vice-Provost for Research and Enterprise) said: "Professor Lomuscio has made a long and sustained contribution to the various areas of AI with particular emphasis on verification and validation of AI systems. His scientific contribution in this area has been deep and significant.
"This research will radically change the present landscape and will enable the precise analysis of existing and forthcoming machine learning methods so that they can be safely deployed in numerous applications.
"The programme of research will also form the basis for certification efforts at government level, thereby enabling the UK to be at the forefront of AI technology."
Article text (excluding photos or graphics) © Imperial College London.
Photos and graphics subject to third party copyright used with permission or © Imperial College London.