Can an algorithm swing an election?

And how would we ever know either way?

A few years ago, not long after the Great Cambridge Analytica Scandal of 2017, I wrote a research paper for the Information Commissioner’s Office.

It was about future risks to elections posed by online targeting advertising. You can still read it, here. The Future of Political Campaigning.

Those ‘future’ risks are now present risks.

I’d been involved in the Cambridge Analytica story – I interviewed the CEO Alexander Nix and even the woman who ran Donald Trump’s advertising drive on social media. (‘Without Facebook, we wouldn’t have won’ she told me.) It was part of a two-part series called The Secrets of Silicon Valley.

Here’s a ‘fun’ clip from the show.

Theresa Wong, who ran Trump’s Facebook in 2016

Most people were wrong about Cambridge Analytica. It remains stubbornly fixed in the public mind that a mysterious army of Russian bots and ‘psychographic’ mind control moved the public to vote. There’s little evidence of that. Cambridge Analytica was a scapegoat for liberals struggling to see why so many disagreed with them about so much. The real story - and a worrying one too - was how elections were changing: becoming more datafied, automated, personalised. How rules designed to ensure fairness in easy-to-see television adverts or billboards couldn’t handle social media.

That was a less interesting story than Russian Bots hoodwinking the masses to vote for Brexit / Trump / etc. But the ICO did realise it, which is why they asked me to write that paper for them. In some ways it’s aged pretty well. There’s a lot in there about the probable evolution of deterministic targeting, automated message generation, location tracking, and how political parties would try to exploit it.

But there was one area I basically ignored, threw in the ‘too difficult to deal with’ pile. And which is probably more important than anything I did write about.