AI is heading to the polls. And the election campaign is changing its approach

Artificial intelligence is not making its way into election campaigns through the front door, with a robot candidate or a campaign speech written by ChatGPT. It is entering through the back door, where messages are crafted, data is organised, voters to be contacted are selected, and decisions are made about what to say to whom. And that is precisely why this change is more significant than it appears.

In the United States, at least 220 adverts in this election campaign contain AI-generated material, and over 137 million dollars has already been spent on airing them. Some are caricatures, others are aggressive manipulations: candidates are shown doing or saying things that never happened. But the most interesting part is the one that is least visible. Eighty-three per cent of American political strategists and advisers report using artificial intelligence on a regular basis. Not to create science fiction. To run election campaigns.

AI is used to sift through databases, personalise messages, gain a better understanding of voters and make door-to-door canvassing more efficient. Democrat Abdul El-Sayed, in Michigan, spent $170,000 on Daisychain, a platform that converses with voters via text message. Swing Left has volunteers record voice memos after door-to-door meetings and then uses AI to analyse hundreds of conversations. Three staff members working for the Democrat Saikat Chakrabarti have even used AI to recreate almost all the software needed for the campaign.

The paradox is that the more AI becomes involved in campaigns, the more politics can return to being a human endeavour. If an algorithm can handle the tedious work – organising data, building tools and sifting through mountains of information – candidates can devote more time to what no machine can truly do: looking voters in the eye and winning them over.

Of course, there is another side to the story. More and more voters are receiving political information directly from chatbot summaries: this has already happened to at least 38 per cent of voters. As a result, campaign teams have begun optimising their websites not only for Google, but also for artificial intelligence, in an attempt to influence what chatbots will say about the candidates. The Economist calls it ‘sloptimisation’: a sort of SEO for machines that are becoming political intermediaries.

The problem is that machines get it wrong. A study cited by The Economist found incorrect, incomplete or outdated answers in more than a quarter of basic questions about the mid-term elections. And some have already realised how to exploit this weakness: websites masquerading as news outlets are being created to feed chatbots with information favourable to certain candidates. To this are added deepfakes and fake videos. Of spending on adverts featuring AI-generated content, 88 per cent has so far come from Republican campaigns and groups close to the party.

The lesson, however, is not that AI will destroy democracy. Nor is it that it will win elections in place of the candidates. It is simpler than that: artificial intelligence is becoming an infrastructure of politics, just as television, opinion polls, the internet and social media were before it. It can make a campaign smarter or more underhand; it can help to understand voters better or to manipulate them more effectively. The technology itself does not decide which side to take. But those involved in politics have already decided that they can no longer do without it.