From the recent House of Lords AI report to the government AI Sector Deal yesterday, it’s been a busy time for artificial intelligence. We ask our Programme Director for Artificial Intelligence, Adrian Weller, to tell us more about what AI means to him and his vision for the Turing’s programme of research.
Artificial intelligence (AI) is a term that can mystify and mislead in equal measure. However, AI probably helped you get to work today. It’s in our phones. In airports. In hospitals. So-called ‘narrow’ artificial intelligence has become a ubiquitous part of modern society. Websites like Amazon and Facebook that tailor their content and recommendations are using interconnected networks of AI systems. Voice recognition. Search engines. Self-driving cars. All use AI. Simply put, artificial intelligence is about machines which act intelligently – typically making predictions or decisions about multiple aspects of the world in which we live.
AI is getting very good at these particular, narrow tasks, usually based on machine learning (ML); where a computer learns the mapping between inputs and a desired response, without being programmed exactly how to do this, in order to improve its performance progressively at a specific task. These systems typically require large volumes of ‘training’ data to be effective, and usually work best when such data has been carefully labelled (i.e. cleaned up, organised, and classified) by humans.
Many machine learning techniques have been around for decades but it’s only recently that we’ve seen an explosion in the development of AI products and research. This has been due in large part to massive increases in computing power and available training data.
These recent advances can lead to a tendency towards hype and hyperbole, so it’s important to understand what AI can’t do well yet. For example, it is still very difficult to learn abstract concepts in one context, such as playing a classic video game, and then to apply those concepts to perform well in a closely related context, such as the same game but with the main characters’ locations slightly altered. Machine learning training is also still poor at ‘unsupervised learning’, where data is raw and unlabelled. Furthermore, many ML algorithms are only as good the training data on which they learn – if training data contains biases, the algorithms produced are likely to be biased too. This issue is of particular importance in instances where algorithms are assisting in life-changing decisions, such as for parole hearings, loan applications, or university admissions. A growing body of technical research is being developed to help detect and remove these unfair biases.
With AI techniques and systems being used in ever more diverse facets of our lives, there are complex challenges that go hand in hand with leaps in innovation and development. These challenges include, but are not limited to: dealing with uncertainty, robust performance at scale, how to learn with little information, privacy and security, fairness, and transparency. These are among the areas that the Turing’s programme in AI will aim to tackle, by advancing world-class research into artificial intelligence, its applications, and its implications for society. The programme is built on the wealth of expertise, disciplines, and knowledge that we have across our network.
Exactly where the current trends in AI will take us is in the next few years is hard to predict. However, there are many opportunities where current know-how can be applied in a range of different domains. Examples include diagnosis and treatment design in healthcare, traffic control, supply chain management, and even looking for scientific breakthroughs, for example in drug discovery or astronomy. The Turing will be working on many of these different focus areas as we develop our AI programme.
As the national institute for AI, the Turing will contribute to leadership around this research. We will draw on our links with industry and government to help ensure that the UK remains at the forefront of scientific innovation, while building an ethical and regulatory framework for the use of AI that prevents misuse and inappropriate discrimination. This last point is key; whilst AI has the potential to improve our quality of life significantly, it is imperative that it is developed in a way that is fair and beneficial to all.