What do self-driving cars, nanobots and mapping out London have in common? They were all featured at the Royal Society on 12 April as part of an interactive exhibit about the applications of machine learning.

Adding to the eclectic host of exhibits on display, a panel of academic and industry leaders in machine learning discussed the subject and answered the public’s questions.

The discussion, available to watch here, was chaired by Marcus du Sautoy who spoke to us in our previous blog post on the subject.

Introducing machines that learn

Machine learning is a way of programming computers so that, instead of giving the computer a set of detailed instructions to follow, you can program it to be adaptive and learn from data.

Chris Bishop, Laboratory Director of Microsoft Research Cambridge, explained how the development of new algorithms, the availability of data and accessibility of computers with significant processing power had allowed machine learning to take off as a research field in recent years.

Maja Pantic, Professor of Affective and Behavioural Computing at Imperial College London, described how her research uses machine learning to analyse human facial behaviour. Our facial expressions show our reaction to what’s going on around us, as well as giving a window into our inner thinking, and can portray a huge variety of features including age, gender, identity and personality. Maja’s machine learning algorithm trains on data in the form of individual pixels of images, learning how to recognise patterns in images in the process.

It’s all about data

Data is crucial for training machine learning algorithms. With this in mind, the panel discussed the opportunities and issues surrounding the use of data. Chris Bishop spoke about the need to strike a balance between ‘locking-down’ our data, so it is unavailable, and allowing it to be used for projects that have social value. The panel also talked about how commercial value which lies in data is fast being realised by industry and that transparency of how this data is used is critical.

In the exhibition hall, the Future Cities Catapult demonstrated how it had made use of a collection of publicly available datasets at the London Data Store to visualise information about people living in London. Their algorithm reclassified the 33 London boroughs into 8 different categories, grouping neighbourhoods based on how people live, not where they live, with a view to informing how services could be delivered at a local level. The project is called Whereabouts London and aims to explore how open data can be used to help citizens see their environment in a new light.

Looking to the future

Machine learning has the potential to drive advances in a range of fields. One area in which the panel agreed machine learning could catalyse a revolution was healthcare; health data has been collected for a very long time and machines can spot patterns in this data which we cannot or develop new treatment methods. Sabine Hauert, Lecturer in Robotics at the University of Bristol, explained how machine learning could play a role in supporting medical applications for nanobots (incredibly small robots). Sabine and her team use an algorithm to train robots to communicate with each other so they can act with ‘swarm-like’ behaviour – could nanobot swarms delivering drugs be the most effective way to target cancer cells?

In the context of future technological developments, the panel discussed approaches to regulation. While concerns about robots replacing humans or Hollywood-style fears about artificial intelligence received a lot of attention, the panel noted that a range of more ‘mundane’ concerns needed to be discussed within the coming years. Such discussions would need to include machine learning researchers, but also bring in insights from social scientists and engage with the public, to inform how this technology would develop. As Chris noted, ‘we can make choices about how we control machine learning’ and that this is important to remember when negative press often surrounds its potential applications.  As with other technologies, machine learning could be vulnerable to attack, so the panel considered cybersecurity as a key issue as applications using this technology were put to use on a large scale.

As Marcus du Sautoy said on the evening ‘Machine learning raises social, legal and ethical questions, which is why the Royal Society is carrying out this project to identify the associated opportunities and challenges’. We hope this event gave a first look at these issues and exciting new technologies; watch this space for future events and activities.

Keep your eyes on In Verba for further information about the Royal Society’s machine learning project and future events.