Inside the Clubcard Panopticon: Why Dominic Cummings’ Seeing Room might not see all that much

The Russian ISS Flight Control Room (aka a “Seeing Room”) credit: Wikimedia

“A basic problem for people in politics is that approximately none have the hard skills necessary to distinguish great people from charlatans.”

Dominic Cummings, October 2016

I am, of course, late to this party. A week is a long time in Fervent Internet Speculation, and a lot of good things have been written about Dominic Cummings’ now-infamous job ad for “data scientists, project managers, policy experts, assorted weirdos”. If you’re reading this, I expect you’ll have read most of those, but in case you haven’t, I’d recommend Hannah Fry on why maths can’t solve everything; Jeni Tennison on digital twins and limitations of data; Nicolas Colin on the role of R&D in statecraft, and Tom Chivers on rationalism and Chesterton’s fence. Amazingly, I have something more to add.

I’m writing this because I might be one of the weirdos Cummings is describing: someone who thinks in systems and patterns, can read numbers as well as words, marry concepts, draw inexplicable triangles and nerd out about Donella Meadows. While I don’t share Cummings’ politics, I welcome the enthusiasm for science and technology in policymaking — but am deeply concerned by the naivety of the pitch.

Why everyone’s talking about Cummings’ job ad

There are two reasons, in particular, this job ad — or “call for expressions of interest” as Downing St termed it — has attracted a lot of attention.

The first is the language — which has proven to be an effective red rag to the liberal bulls, much, I suspect to Cummings’ delight. But the bluster and disregard for normal hiring protocols masks a fundamental misunderstanding of how to build “high-performance teams” (another of DC’s obsessions) and shows the limitations of a hobbyist approach to technology. Anyone who has shipped software knows that inclusive leadership and diverse, multi-disciplinary teams are simply better at delivering. At best this kind of hiring leads to a Cargo cult of misunderstood white men in expensive Japanese denim and the glorification of “prophets” like Shingy, at worst a mash-up of car-crash Uber office culture and an after-school wargaming club. While those kinds of teams are filled with performance, not much of it is very productive.

The second is the intent, which falls into the fascinating and frustrating category of “interesting but not-quite right”. There is undeniably a role in government for people who understand and can exploit the potential of data and technology for policy and strategic development, who understand how to run comms, and who can look to understand the possible and plausible alternative near- and far-future horizons. And there is a community of people who have those skills and want them to be used. If you’re not put off by (or, in fact, are attracted by) the “weirdo” label, it sounds intriguing and cool and important. After all, working at the heart of government in the gung-ho, disruptive way that Cummings sets will be familiar to anyone who’s worked at a blitz-scaling start-up. But the thing he wants this team to do is inherently flawed — and he’s going about it in the wrong way.

Building the Clubcard Panopticon

Reading around Cummings’ blog (particularly this post), it appears he wants to achieve the following things:

1. To visually and physically show “all” data to everyone who needs it to aid better decision making
2. To improve the velocity and, consequently, the productivity of policy teams
3. To make it easy and hassle-free to accrete and demonstrate evidence around an idea

So that:

1. The “assorted weirdos” can join the dots and come up with new ideas as quickly as possible
2. And the “deep experts on TV and digital” can bring the electorate along by communicating rapidly and effectively

Cummings’ enthusiasm for Brett Victor’s physical computing and for movie-set style Seeing Rooms show, I think, that he wants to get real somehow: to get beyond hand-waving and rhetoric to material proof of ideas and, perhaps, to proof of concept. He wants to use all of the available data to conjure the world in the cabinet room and govern in three dimensions. And while this might be quite the breathtaking ambition, it is also a dangerous folly.

In October 2016, Cummings wrote about VICS, the Voter Intention Collection System created by the Vote Leave team. It’s a CRM system (by the sounds of it, quite an effective one) that enabled the team to run a proper digital marketing campaign, the like that British politics had never experienced before. VICS allowed them to:

Serve about one billion targeted digital adverts, mostly via Facebook and strongly weighted to the period around postal voting and the last 10 days of the campaign. We ran many different versions of ads, tested them, dropped the less effective and reinforced the most effective in a constant iterative process. We combined this feedback with polls (conventional and unconventional) and focus groups to get an overall sense of what was getting through. The models honed by VICS also were used to produce dozens of different versions of the referendum address (46 million leaflets) and we tweaked the language and look according to the most reliable experiments done in the world (e.g. hence our very plain unbranded ‘The Facts’ leaflet which the other side tested, found very effective, and tried to copy). I will blog more about this.

And doubtless these methods were improved upon in the recent general election.

This kind of CRM is very effective when used well, and it sounds like Vote Leave used it very well indeed. But it is not a machine to predict or manipulate the future, and any underlying data science is probably pretty similar to that used by Dunnhumby to create Tesco Clubcard in 1994. The clever things here are the quality of the execution and the category shift. Cummings writes about the importance of hiring physicists in this blog post, but I’m not sure what the physicists did. If the physicists were running voter segmentation and A-B testing ads, it’s fair to say Vote Leave were getting charged for some services they weren’t really using.

This faith in data seems to also drive the desire to build a panopticon. Cummings has drunk gallons of Brett Victor’s kool aid: he believes it’s possible, in fact desirable, to write evidence into an argument as easily as it is to add an emoji. But anyone who’s accidentally sent an aubergine to their mother-in-law knows, it’s pretty easy to choose the wrong emoji, or to misinterpret its meaning. And how will competing data points be chosen? Who will chair the sorting group of Universal Truth that decides which data are good enough to be automagically written into evidence? The creation and selection of emoji requires governance; believe it or not, it gets pretty fraught at the Unicode Consortium, because all information represents an opinion or a particular perspective on the world.

What is most surprising is that Cummings believes a panopticon is possible; that all data is available; that evidence is neutral and unbiased and somehow right. As a history graduate, Dominic Cummings should have an innate understanding of the founding premise of data ethics: all data are influenced by the circumstance of their collection. There is no universal truth: all datasets are incomplete, all datasets are biased.

One of my go-to graphs is always Mary Meeker’s annual update on how much of the world’s data is structured enough to be useful: the latest estimate is 16%. I can only imagine how much (or indeed, how little) of the data at the UK Government’s disposable is timely or well structured enough to be coherently and uniformly transmitted into a Seeing Room. Unless these Seeing Rooms are incredibly focussed on specific tasks (for instance, optimising parts of the transport network; launching a rocket on Mars; building a bridge between two countries), it is possible that the data at their disposal will be incomplete at best, and out-of-date and incoherent at worst.

The job of work to format and bring together sufficient public data sets to make data-driven decisions possible is non-trivial and at least multi-year. I can find no mention of setting up governance structures to ensure data is accurate, that bias is mitigated, and errors corrected and spotted, or — indeed — any awareness of data ethics at all, despite its mention in Boris Johnson’s rather flowery speech at the UN in September 2019. Perhaps — and who knows? — the Centre for Data Ethics and Innovation are on stand-by provide this.

But the most baffling omission is that Cumming’s job ad does not mention interface designers. Both a Seeing Room and Brett Victor’s Data Land rely heavily on interface design and visual representation. Cummings quotes heavily from Michael Nielsen on the importance of good interfaces, including,

‘such an interface makes it easy to have insights or make discoveries that were formerly difficult or impossible. At the highest level, it will enable discoveries (or other forms of creativity) that go beyond all previous human achievement. (DC’s bold)

But the list of freaks and geeks needed to transform government doesn’t involve anyone to craft those interfaces. Data visualisation is a professional discipline, not a naturally occurring phenomenon.

So we have the prospect of possibly transformative policy decisions being driven by incomplete and biased datasets, managed by a team unconcerned by “‘gender identity diversity blah blah’”, and rendered in haphazard and incomprehensible arrangements because no one has designed them. This is not a transformative approach to policy-making; it is a classic transformation money pit.

By Cummings’ own admission, he is not a technologist: he describes himself as “an arts graduate interested in these subjects but not expert”. There are lots of experts out there who could help design and build a high-performance team that helps policymakers make better use of data science, and if Cummings is as good as he thinks he might be, he will stop being a hobbyist who wants to be friends with the cool sci-fi kids, and find some professionals to do the job instead.