Josh Smith, Senior Researcher at the Centre for the Analysis of Social Media, Demos
Targeted online advertising has produced a fundamental shift in the ways in which political campaigns are conducted. The ability to divide the electorate into precisely shaped groups, lovingly adjusted by inferred characteristics and patterns of behaviour, has helped parties and campaigns reach ever smaller groups with the right message, and measure the response in real time.
The importance of digital messaging to parties can be seen in their budget sheets. In 2015, the Electoral Commission found that 23 per cent of the total spend for the election went toward digital campaigns, and in the run up to the EU referendum, Dominic Cummings estimates that Vote Leave ran around one billion targeted adverts, mostly via Facebook.
In this, campaigns are following the market. Across the political spectrum, electoral politics has become, in the words of US academics Chester and Montgomery, “fully integrated into a growing, global commercial digital media and marketing ecosystem that has already transformed how corporations market their products and influence consumers”.
The capabilities offered to parties by technology are set to change over the very short term. In 2018, Demos produced a report for the Information Commissioner’s office sketching out how advances in data analytics, AI and micro-targeting might affect political campaigns. We found a sector buzzing with possibility.
One company promises to help find sympathetic voters through their phone’s presence at political events, and follow them home with advertising. Another offers ‘voter file enhancements’, providing electoral rolls fleshed out with data on income, occupation, education, likely marital status, ethnic and religious identification, magazine category subscriptions, pet ownership and so on.
As the character and volume of data increases, along with our capacity to make inferences from it about voter mood, preference and behaviour, strategies for targeting are likely to become ever more sophisticated and complex.
Incoming technology has implications not only for the accuracy with which campaigns can direct messaging, but also for the basis on which that targeting is performed. Take, for example, the rise of ‘deep learning’. Crudely put, one powerful form of algorithm, a ‘neural net’, can be used to work out how to turn inputs – e.g. detailed voter profiles – to outputs – e.g. the wording to use when contacting them. Deep learning takes this process one step further, not only automating the process of working out how to get from voters to vocabulary, but also deciding, itself, which information on a voter should be used to inform this content.
As voter profiles become increasingly detailed, use of this technology by campaigns could lead to their targeting people based on characteristics, such as health, race or sexual preference, which campaign organisers may not be aware they are using. This could be true even where protected characteristics are not stored, as machine learning makes it possible to infer detailed demographic information based on seemingly innocuous data. For example, MIT has used deep learning to accurately infer a user’s age and gender simply from data on the timing and duration of their phone calls.
Deep learning is just one example of the technologies which are changing campaigns. Political groups may soon be able to incorporate data gleaned from ‘wearables’ (such as smartwatches) and the ‘internet of things’ into their messaging, or to algorithmically generate entirely new slogans and designs. The direction of travel is towards campaigns being able to address each voter individually, based on their specific profile; advertising to an audience of one. This individual targeting risks breaking down public political discourse – having a conversation about what a party stands for in the run up to a campaign is difficult if your neighbours and family have seen completely different messages.
This could, of course, be a shot in the arm for politics. A detailed and accurate view of the electorate could enable parties not only to better allocate restricted messaging budgets, but also to develop policies more in tune with the lives and concerns of people they will affect. For this to happen, however, rapidly developing profiling techniques will need to move beyond their status as a marketing tool, and be understood, and thoughtfully implemented, by policymakers.
Making individually crafted, algorithmically-honed offers to people is a sound marketing technique, but politicians need a way to ensure those offers are followed up. At present, it is difficult for campaigners themselves (let alone regulators) to know on which basis they are targeting voters, and what they are telling them.
It is imperative that these technologies are used to improve our political process, and understood by their users as well as their targets. Otherwise, we run the risk of becoming an electorate resigned to a politics which says the right things but never fulfils its promises, singled out for persuasion in ways we do not expect or understand, and to which we cannot meaningfully consent.