Alan Turing Institute annual lectureWhen Harry Powell shared, that he was planning to attend this year’s Alan Turing Institute annual lecture, I naturally asked him to blog for us.

Harry is a very experienced Data Science leader, having created & led Data Science teams at several businesses, including Barclays. So, I was interested to learn more about his interest in a lecture from Prof. Sandy Pentland of MIT.

As someone with a background in R&D, MIT is one of those institutions you learn to admire & listen out for, as a source of innovation. But the planned lecture, on the topic of “Better Living through Trusted Data” was also engaging. I’ve shared before on the importance of ethical dilemmas for Data Scientists, especially those who want to achieve social good through their work.

Anyway, I’m delighted to share that Harry took up my challenge & joins us as our latest guest blogger. Here is his debrief on what sounds like a great lecture. After that I’ve also embedded the video of the entire lecture, so you can see for yourself why Harry is so enthusiastic. Over to Harry…

The work of Sandy Pentland and his team at MIT

Yesterday evening I had the pleasure of attending the Alan Turing Institute public lecture by Sandy Pentland of MIT, “Better living through trusted data” in London, and going to dinner with Sandy, John Taysom and others afterwards to continue the conversation. You are probably aware of Sandy, through his work at the MIT Media Lab and his book “Social Physics” on which much of his talk was based. Sandy is a very engaging speaker and I think everyone got a lot out of his presentation. I thought I would share with you some of my thoughts on what he said.

Sandy and his team have been working for a while, on the way that ideas diffuse across networks of people, to create a form of collective knowledge and intelligence. One which is greater than the sum of the parts. He has taken the new technology of big data, to model these relationships, in a way that was not easy previously and to then test the results against real life examples. It turns out that your productivity is strongly influenced, by the diversity of your network. These connections allow new ideas to be formed. They can be exchanged and tested. The number of connections, is often a much more powerful predictor of effectiveness, than conventional machine learning features such as demographics.

Sandy then extends this kind of analysis, to organisations and society. For example, noting that those at the lowest end of the socioeconomic distribution tend to have a very narrow circle of connections, and those at the very top have a very wide circle connections, entrenching economic division. In terms of using these ideas to improve the world around us, a rather attractive and simple idea is that mechanisms and structures, which improve the diversity of ideas, are beneficial. So, make sure that your department in your business has a free coffee machine available to all your colleagues: the footfall that it will drive, and the conversations it will generate, will make your team a top performer.

From Machine Learning to System Modelling

So what’s new about all this. Isn’t this obvious? Haven’t sociologists been saying this forever?

I guess that is true but I think the team at MIT are bringing three things to the field:

  • an ability to gather and compute individual-level data;
  • rigorous mathematical models that reflect natural human interactions;
  • a focus on prescriptive quantitative conclusions about how to improve the world.

I haven’t seen a whole load of the maths, but what I have seen seems plausible and has undoubtedly delivered impressive results.

My reflection on the talk overall is that it heralds a move away from simply using machine learning to predict optimally given a social system, to a set of models which through an explicit modelling of the system itself, will enable us to change to that system in order to improve outcomes.

This is because conventional academic work has been restricted, by computation, data and an academic preference for closed-form solutions. You get Nobel prizes for those! So, the only interpretive models have been equilibrium models, at an aggregate level. But, of course, one cannot set policy for aggregates, nor for equilibria. They are both idealised abstractions, which do not exist in real life.

Alan Turing Institute annual lecture – Harry’s takeaway

It is great to move beyond aggregate statistics, to structured networks with interactions explicitly modelled. But, the conventional static analysis of networks (such as measuring betweenness and centrality), is inadequate to deal with the dynamic nature of the creation of connections, and the diffusion of information & ideas between people.

It is this willingness to study, local changes in large networks, and to generate the data and maths required to do it, which marks out Sandy Pentland’s work. That’s what makes it interesting and potentially important. The internet of things generates more and more information, about each of us as nodes, in a grand social interaction. So, we can expect more and more of this way of looking at the world, and the disconnect between micro and macro models will diminish.

Alan Turing Institute annual lecture – see for yourself

Thanks to Harry for those reflections on what certainly does sound an interesting & fruitful field for more experimentation. Given the quality of this thinking, I’ll certainly watch out for attending the Alan Turing Institute annual lecture next year.

If Harry’s summary has peaked your interest & you’d like to learn more about Sandy’s work and how such Data Science can be applied in our world, you’re in luck. Here is the video of the full lecture by Sandy. So, get yourself a coffee, settle into a comfy chair & enjoy having your eyes opened to a new approach: