To fulfil the promise of AI, it takes more than machine learning. We’re bringing together experts from all corners of computer science – human-computer interaction, network systems, interface design and many others – who are all needed to make effective AI systems.
Our position as one of the world’s leading research institutions for machine learning has given UCL Computer Science a significant advantage in AI. The historic influence of our academic experts in both research and policy (along with our extensive partnerships with giants such as Google DeepMind and Meta) has placed us firmly at the heart of the UK’s emerging AI ecosystem.
Every insight or approach from an established field used to accelerate AI gets repaid many times over. For example, our software engineering researchers are finding new ways to reduce AI’s carbon footprint, or eliminate inherited biases. In return, their work becomes easier through AI-driven developer tools, like automated program repair systems that hunt out and fix coding bugs.
This feedback loop of interdisciplinary innovation is enabling UCL Computer Science to realise and deliver the next generation of AI systems.
How is UCL Computer Science research contributing to the development of AI?
Explainable AI
Many of the misconceptions around the capabilities of foundation models stem from their ‘black box’ characteristics. It’s very difficult for anyone to assess the process by which the model has arrived at an answer.
Drawing on philosophy and mathematics as well as computer science, our research into programming principles and verification is laying out the theoretical basis on which explainable AI will be designed. Through the development of computing architectures based on logic thinking, we’re contributing towards the creation of future AI models which will be explainable, verifiable and safe.
Sustainable AI
One of the most important ways we can use AI technology is to support the global effort to mitigate the effects of the climate crisis.
Analysing the impact of human activity on planetary systems is a huge challenge, involving complex interactions (many of which we don’t fully understand) and massive datasets.
UCL’s climate models are used to set long-term energy policy and lay out transitional pathways at the highest levels of international cooperation. Our research into modelling extreme weather patterns is also providing new insights and short-term predictive capabilities for immediate disaster response and recovery.
The environmental impact of resource-intensive AI technologies is another important area of study. Ourresearchers are working to create more efficient transformers and diffusion models so that machine learning algorithms can use less energy and water.
Our machine learning research into stabilising nuclear fusion processes is providing another way for us to potentially reduce our reliance on fossil fuels by boosting other methods of energy generation.
Responsible AI
Our position as one of the few academic institutions in the UK with the technical capabilities to create new algorithms (and expertise with large-scale infrastructure through collaboration with industry partners like Google DeepMind and Meta) also carries a weight of responsibility.
The pace of progress in AI research, particularly in the private sector, means that it’s vitally important for us to try to ensure that the power of AI is harnessed for the benefit of our citizens, our communities and our economy, both here in the UK and abroad.
Our lead role in the AI Hub in Generative Models allows us to maximise our collective efforts and pool the research strengths of nine leading UK universities, alongside industry partners like Cisco and IBM. Professor John Shawe-Taylor's role as the UNESCO Chair in Artificial Intelligence provides additional research focus on developing frameworks for AI ethics to inform policy and education globally.
Embodied AI
Artificial intelligence is trained on the data we input into computer systems and networks. However, there are many other potential sources of data.
UCL Computer Science robotics researchers are aiming to bridge the gap between foundation models and the physical world. Developing robotics applications that can analyse scent, for example, or stretchable electronic skins that give robots a three-dimensional sense of touch using magnetic technology.
Meanwhile, our research in computer vision is equipping multimodal foundation models with the ability to analyse 2D images and make inferences about the 3D environments they represent.
Eventually, embodied AI systems will be able to train and learn through interactions and sensory inputs from the physical world, helping us to better align these systems with human needs and experience.
