How AI Could Soon Take Over Elections

Artificial intelligence is already influencing the 2024 elections in ways that could reshape how campaigns are run, and voters are informed. One way is through language models that can generate text to nudge people towards specific actions, such as voting for a particular candidate or taking a specific action that supports a cause. Some social media platforms already use this algorithmic nudge to boost engagement and influence search algorithms.

But these kinds of tools can also be deployed in more sinister ways — for example, to manipulate voters and spread fake news or extremist narratives. Bots, autonomous accounts that can spread wrong political messages to create the illusion of public support, are already a significant problem in the social media ecosystem. For example, bots are responsible for the massive swarm of automated tweets during the 2016 presidential campaign to choke off dissent and sow division. And they are the source of much of the false information that dominated Twitter feeds in the days leading up to the French presidential election when they swarmed social media with leaked documents and fake news designed to depict Emmanuel Macron as a hypocrite and a fraud.

These tools are a danger to democracy because they can be used by anyone with access to the internet and an idea of what they want to accomplish. In addition, they can be manipulated and exploited by both political parties, who have a vested interest in making their candidates more popular, and by foreign governments with the potential to alter the results of a democratic vote.

The most dangerous kind of political AI is the kind that can be programmed to induce people to behave in specific ways, such as changing their votes. Such an AI could have the potential to radically expand and accelerate the kinds of behavior manipulations that political campaigns have been using since the early 2000s when researchers started finding that they could sweep large amounts of data about an individual — including their location, income, media habits, and social connections — and then find patterns that suggest whether someone will donate money or vote for a particular candidate. These are the things that a hypothetical AI called Clogger might be able to do on an immense scale, with the additional goal of actually changing people’s votes.

It’s possible to avoid this politically influential black box AI if politicians, campaigns, and consultants forswear its use, but that seems unlikely. And if a politically influential AI like Clogger is developed, the temptation to use it will almost certainly be overwhelming. If a consultant sees that it can change voters’ minds, that could be the difference between winning and losing. The most important thing we can do is to ensure that the technology has transparency and democratic control so that we know it isn’t being misused.