Artificial intelligence technologies
Artificial Intelligence (AI) technologies aim to reproduce or surpass abilities (in computational systems) that would require 'intelligence' if humans were to perform them. These include: learning and adaptation; sensory understanding and interaction; reasoning and planning; search and optimisation; autonomy; and creativity. The applications of AI systems, including but not limited to machine learning, are diverse, ranging from understanding healthcare data to autonomous and adaptive robotic systems, to smart supply chains, video game design and content creation. This research area primarily covers fundamental advances in AI technologies, while applications of such technologies are captured within other subject domains.
We aim to maximize opportunities arising from the current increased global interest in AI and its widespread applications, as well as the Government’s Industrial Strategy, namely via the National Productivity Investment Fund (NPIF) . This strategy recognizes that AI and the data economy was named as one of four Grand Challenges in the Industrial Strategy, as well as AI research's importance to data science in general and Robotics and Autonomous Systems (RAS) specifically It aims to show where activity can be focused to allow the UK to grow in international expertise in both fundamental theory and more applied research.
By the end of the current Delivery Plan, we aim to have:
- A balanced portfolio of AI research and training that encompasses both sub-symbolic and symbolic AI with the aim of encouraging links between these. Sub-symbolic AI includes learning (adaptation) from data and the science and processing of data, using techniques such as (deep) neural networks and reinforcement learning. The recent growth and popularity of AI has largely been in this field, including work at the Alan Turing Institute as the national institute for data science and AI, which plays and important role in this space. Symbolic AI which includes search, reasoning, planning and knowledge representation has seen less attention but EPSRC is looking to encourage research in this space particularly, including areas such as explainable, and trustworthy AI.
- A strong AI community which actively engages in the challenges of Responsible Research and Innovation (RRI) and public acceptability of artificial intelligence, ensuring that research outcomes are socially beneficial, ethical, trusted, and deployable in real world situations.
- A supply of people with high-level skills across the breadth of AI technologies, reflecting growing demand, and who can contribute expertise across a wide range of domains (e.g. the future of healthcare delivery)
- Researchers combining development of new methodology and applications (e.g. by working alongside research enablers such as research engineers, translational researchers and industry collaborators with application expertise)
A strong portfolio that contains AI-enabled RAS technologies co-created with other disciplines (e.g. robotics, human-computer interaction, computer vision and the humanities and social sciences). This should take into account how these intelligent systems interact and collaborate with humans, and consider their validation and verification, especially in application areas where the dependability, safety or security of implementations is a concern. AI researchers will play a key role in furthering EPSRC's Future Intelligent Technologies and Data Enabling Decision Making cross-ICT priorities and are well-placed to contribute to the other cross-ICT priorities. In order to maximise the impact of these contributions, they should ensure effective communication with researchers in other contributing areas such as natural language processing, visualisation and HCI.
We recognise the need for researchers to work with large-scale data and we encourage them to develop collaborations with users to facilitate this. We also encourage them to explore alternative routes to access sufficient computational resources (e.g. use of commercial clouds). However, UK academia should not try to imitate industry, and should focus on AI opportunities not yet identified by industry or not yet commercially viable, particularly those leading to beneficial societal impacts.
AI is likely to become an increasingly dominant feature of our world and understanding the future of AI and its impact on future society are critically important research areas. Explainability of AI decisions is key, as is the use of AI to simplify data in order to facilitate human understanding and decision-making (Evidence Source 13).
The recent growth in this research area, namely on the data intensive sub-symbolic side of the AI technologies portfolio has implications for the ICT hardware related research areas. We recognise that research in ICT hardware should keep up with the advances made in AI technologies and that better links between across these communities need to be encouraged.Highlights:
The importance of AI to UK industry was recognised in the UK’s Industrial Strategy White Paper: “Building a Britain fit for the future” which identified AI and the data economy as one of the four Grand Challenges for current and future UK industrial leadership (Evidence Source 10). The strength of this sector in the UK was also acknowledged in the government commissioned report “Growing the AI Industry in the UK” which stated that the UK has AI companies that are seen as some of the world’s most innovative, in an ecosystem that includes large corporate users of AI, providers large and small, business customers for AI services, and research experts (Evidence Source 11). This report made a number of key recommendations for actions and interventions required to sustain and develop the UK’s AI sector, which were responded to in the 2018 AI Sector Deal (Evidence Source 12).
A key recommendation of the report “Growing the AI Industry in the UK” was the need for a major step-change in UK development of high-level skills for AI. Amongst other things, the review recommended 200 more PhD places per annum in AI at leading UK universities, attracting candidates from diverse backgrounds and from around the world. The recent UKRI investment in AI studentships through centres for doctoral training will start to tackle this recommendation.
The UK is a strong international contributor in this area, evidenced by the existence of world-leading UK research groups. In 2017 we invested £42 million in the Alan Turing Institute as part of a joint data science venture with five university partners, in 2018 the ATI evolved its role to national institute for data science and AI and eight further universities are joining as partners.
AI and RAS technologies have an increasing impact on UK policy and the economy, improving business competitiveness, providing effective solutions to societal problems and empowering people to live more fulfilled lives and make more informed choices. (Evidence source 1,5). A wide range of influential commentators have spoken of the societal risks and opportunities associated with AI research. Research investment is needed not only on development of tools and techniques but also on understanding risks and maximising opportunities.
The importance of - and UK international competitiveness in - machine learning (ML) is evidenced by the significant industrial investment being made in UK ML, including Google's acquisition of start-up DeepMind in 2014. ML has become an enabler of many technologies (including but not restricted to language technologies) and is expected to play an increasingly important role in data analytics.
Many UK universities are actively investing in data science/ML and are attracting international experts. Nevertheless, we need to ensure the UK has the trained people to cater for the serious upward trend in demand for ML skills. Recruitment at PhD level is healthy (with keen demand from students and employers), but recruitment/retention in academia beyond this is a problem. There is the threat of key capacity being lost to industry, with universities unable to compete (e.g. in terms of salary, provision of computational resource and access to large scale data). (Evidence Source 3, 8, 9)
The applications of AI systems are very diverse. The UK is in a particularly strong position internationally to develop AI technologies in healthcare, partly due to having National Health Service (NHS) data stakeholders interacting with universities and partly because the UK has one of the world's best-established AI communities, able to offer diversity of research. This makes AI an area of national importance for the future of healthcare delivery. (Evidence Source 3)
This area is linked with many research areas and Themes across EPSRC. Areas of highest current relevance are: Biological Informatics, Human-Computer Interaction, ICT Networks, Image and Vision Computing, Information Systems (especially the direct link with this Research Area and knowledge representation and reasoning), Natural Language Processing, Operational Research, Pervasive and Ubiquitous Computing, Robotics, Software Engineering, Statistics and Applied Probability, and Verification and Correctness. Some of these areas are users of AI technologies; some are providing important technical drivers.
AI will most strongly contribute to Connected, Healthy and Resilient Nation Outcomes over a shorter timeframe. Contributions to some of the Ambitions within the Productive Nation are expected to be at a lower level and/or to take place over a longer timeframe. Particularly relevant Ambitions include:
C1: Enable a competitive, data-driven economy
AI is expected to contribute to advances in data science that will deliver the smart tools and analytical techniques required to generate actionable information from large and diverse datasets.
C2: Achieve transformational development and use of the Internet of Things
AI is expected to contribute to the way information can be intelligently assimilated and utilised across a range of business sectors and services.
C3: Deliver intelligent technologies and systems
AI will allow smart tools and intelligent technologies to take the Connected Nation beyond flows of data and turn it into physical action. Multidisciplinary research involving social scientists will enable such tools to be acceptable, usable and ethical.
H3: Optimise diagnosis and treatment
AI can contribute to the development of sophisticated, personalised models that will enable individuals and clinicians to plan treatment holistically, based on a range of possible, and increasingly accurate, predicted outcomes.
P4: Drive business innovation through digital transformation
AI could contribute to the development of autonomous and intelligent technologies that can transform business models and services.
- Commons Select Committee, Robotics and Artificial Intelligence Inquiry - Written and Oral Submissions, (2016).
- EPSRC, Analysis of Research Excellence Framework (REF) 2014 data and EPSRC Knowledge Maps, (2014).
- Community and user engagement (individual input and group feedback).
- EPSRC, Future Intelligent Technologies (FIT) Workshop, (2015).
- Robotics and Autonomous Systems Special Interest Group (RAS SIG), Strategy and Landscape Documents, (2016).
- The Royal Society, Machine Learning Conference Report (PDF) and ongoing policy project, (2015).
- Nesta, Machines that Learn in the Wild (PDF), (2015).
- IT Jobs Watch, Tracking the IT Job Market, (2016).
- EPSRC, Output from the EPSRC Speech Technologies exceptions process (2015).
- BEIS, Industrial Strategy White paper – building a Britain Fit for the Future (2017)
- Growing the artificial intelligence industry in the UK, (2017)
- BEIS, Artificial Intelligence Sector Deal, (2018)
- Lords Select Committee on Artificial Intelligence – evidence and report, (2018)
Research area connections
This diagram shows the top 10 connections between Research Areas within the EPSRC research portfolio. The depth of the segment relates to value of grants and the width of the segment relates to the number of grants shared by those two Research Areas. Please click to see the related Research Area rationale.
Visualising our Portfolio (VoP)
Visualising our portfolio (VoP) is a tool for users to visually interact with the EPSRC portfolio and data relationships.
EPSRC support by research area in Artificial intelligence technologies (GoW)
Search EPSRC's research and training grants.