Mirco Musolesi's research portrait

Mirco Musolesi's interest in privacy came about as a by-product of his interest in modelling human behaviour at both individual and collective scales using emergent forms of data: the digital traces we leave behind as we navigate the world. Taken together, these streams of data offer the opportunity to test, even prove or disprove, theories of human behaviour that have been in play for years - but they also give a rich and intrusively intimate picture of the life of the individuals concerned.

These traces are many and varied, some deliberate, some inadvertent. A modern smartphone, for example, has microphone, cameras, accelerometer, gyroscope, multiple radios (Bluetooth, WiFi, cellular), and many sensors that measure levels of proximity, ambient light, and other environmental factors. All of that may be functioning and generating data before you even begin to look at the more intentional streams users generate as they post to social media, check in from their location, and interact with friends and contacts via email, text messages, or audio/video services.

"I got interested in privacy," Musolesi says, "because I had to use these datasets and realised what it's possible to do. And then I think also that it's our responsibility as researchers to try to use the data in a responsible manner."

In his 2014 paper, Big Mobile Data Mining: Good or Evil?, he examined the ethics of using these fine-grained, large-scale, longitudinal datasets to carry out new and interesting studies of behaviour.

In the paper, "The idea was that if you use these new forms of data, especially from mobile phones, you can get a very rich picture of human behaviour." Despite the privacy implications, "I would argue that the benefits are much bigger than the costs here, and I would say also that the most important thing is that people should be aware of the possibilities of what can be done with the data." Still, he says, this kind of work should not be stopped by privacy concerns: "The idea is not to collect more data, but to use the data that's already there. It's very difficult to reverse the course of action now."

One of the obvious applications for work that studies patterns of human behaviour is security. "I believe that the models we develop might also be used for security in the sense that we can understand malicious behaviour and start to understand and identify potentially dangerous or malicious individuals." This isn't, he hastens to add, an imminent goal or even a current project. Rather, because people don't change a lot in terms of the places they visit regularly, he is using prior information about patterns to try to find the identity of users ("identity" here may mean any of a number of aspects, not specifically a particular real-world identity as we typically talk about it).

The word "profiling" carries with it uneasy connotations of prejudice and exclusion. However, in his work studying location data, by "profiling" Musolesi means a picture of the places individuals or groups of people visit and their activities.

"Sometimes we need to profile users, and when it's necessary I think the methodology is extremely powerful." With that power, however, again comes responsibility: "When we release a dataset we need to obfuscate, but in general I would say these two aspects - identification, or profiling for security, and privacy - are connected, and there's a really fine balance there."

Musolesi's journey into computer security and privacy began with an interest in mathematics and computing in Bologna, Italy, where he grew up. Like many Italians, he stayed in his home city for university, where he studied electronic engineering, always staying on the mathematical and computational side.

"I really like abstract models and how to capture things with mathematics," he says. A research fellowship at UCL led to a master degree and PhD in computer science. For the latter, he studied delay-tolerant networks, that is, those that may be intermittently connected or must overcome long delays in order to function. Examples include the deep-space Internet, communications in rural areas, and, more interesting, Bluetooth communications. Using Bluetooth, an individual may pass a message directly to another to carry to a third party they're likely to meet - a store-and-forward mechanism similar to the way the old Fidonet bulletin boards passed on email, "but with people".

The idea of exploiting people's patterns of mobility in order to optimise communications without a central infrastructure was of particular interest to Musolesi, and it led him to work on modelling social networks and mobility. "And then I got really interested in the prediction of social conduct and this was my PhD."

In a postdoc at Dartmouth, he worked on sensing behaviour using mobile phones. "I was lucky that we were among the first groups doing that," he says. "In 2007, we built CenceMe, one of the first sensing systems, based on the Nokia N95. The resulting paper, Sensing meets mobile social networks: the design, implementation and evaluation of the CenceMe application, was presented at the 2008 ACM conference on embedded network sensor systems. Following work at the Cambridge University Computer Lab and lectureships at St Andrews and Birmingham, Musolesi has returned to UCL to work on data science, digital traces of human behaviour, and social dynamics - work that now has important applications in security and privacy.

His most recent work, presented at the 2015 ACE opening meeting, is part of an ESPRC-funded project called The Uncertainty of Identity. In this project, Musolesi uses data from social networks to try to model and understand spatio-temporal patterns. The privacy news: location profiles are even more revealing than we thought.

"The general idea of that paper," Musolesi says, "is that given the practical availability of these mobility and GPS traces, it's quite easy to identify users with just a few data points. The innovative part of that series of papers was about the fact that it's not just that you get a dataset of points and give me three points and I can identify a person, but that I can train a machine learning algorithm on a training set and then if you give me new points I will be able to associate these points with a certain individual. This is different from saying I have full knowledge - it takes much less information than previously thought. It mathematically proved that people have routines - they repeat patterns over time, as do animals."

Part of his work involves applying models initially developed for studying animals to human behaviour. "It's quite interesting, because you can really understand that the mobility patterns are quite similar." Being able to build mathematical models of such patterns has obvious applications in areas such as transportation planning.

Musolesi first discovered the correspondence while working on his PhD. "I became quite fascinated by trying to model contact patterns and mobility between individuals, and then I discovered that there was this ongoing work in zoology and ecology on animals." There's no real word for this field of study; Musolesi suggests calling it "theoretical biology".

In his current work, Musolesi is interested in understanding security and privacy for ubiquitous computing; the Internet of Things will embed millions of always-connected devices into the fabric of the city and its buildings. "I want to try to understand the risks and implications of using these devices to build models for potential identification - and obfuscation - of the data coming from them," he says, "and also other uses for them such as authentication."

Despite the privacy challenges, Musolesi remains optimistic, both about the benefits of this approach and about researchers' understanding of the responsibility involved. "For example, one use of data is in extremely interesting applications for intervention in epidemiology in Africa. There are applications of these datasets for good things, and that's something we should consider when we talk about privacy and the problems of mining them." There are, as always, trade-offs to be made, and the balance looks different in different contexts. Still, he says, privacy is an important value, and something that shouldn't be given up. As a result, ethics is becoming a more important concern in computer science and statistics, much as it has been in past scientific endeavours such as building the atomic bomb. "Since this is about the everyday lives of people, ethics is becoming a central concern, and it should be part of the practical education of computer scientists."

This page was last modified on 06 Jan 2016.

Dr Mirco Musolesi



Room: 213, Pearson Building


+44 (0)20 7679 0567


m.musolesi [AT] ucl.ac.uk