Sam Jeffers is a senior fellow at the Dais, a public policy and leadership think tank at Toronto Metropolitan University, and co-founder of the election transparency project Who Targets Me.
In 2024, there will be at least 65 elections in 54 countries, and even more when you count individual countries’ votes for their leaders in the European Parliament. Like the circling of some great democratic comet, we won’t see the likes of this again until 2048. And as we enter a year where more than two billion people will vote, Silicon Valley and the political system are panicking about the impact of artificial intelligence on elections.
In March, 2022, a deepfake video of Ukrainian President Volodymyr Zelensky declaring his country was “surrendering” was posted online. In May last year, an AI-generated photo of an explosion at the Pentagon caused U.S. markets to dip briefly. In September, AI-generated videos and audio purporting to demonstrate election rigging circulated in Slovakia just as citizens were headed to the polls, and in October, fake audio of Keir Starmer (likely Britain’s next prime minister) berating a staffer spread across British Twitter for an hour or so.
All are cited as examples of the risks to democracy and truth that AI can pose. But it’s important to note that each was quickly debunked, and they never made their way into more authoritative news sources, where they might have had a wider impact.
Currently, generative AI produces convincing, but short-lived, fakes. The quick debunking in these cases also shows us that information, disinformation and misinformation, as piecemeal phenomena, generally don’t matter very much on their own. If they are to be dangerous, they need to hitch themselves to a wider, more convincing narrative.
Over the past few years, we’ve seen the American right create these narratives in the form of conspiracy theories such as Pizzagate and QAnon, as well as the MAGA movement (all without the use of AI). Fortunately, to have a similar effect, AI-generated content will also need skilled human operators, with knowledge across politics, news and distribution. Without these skilled humans, generative AI isn’t autonomous enough to create, spread or sustain stories like the ones that have taken hold in the U.S. Just as most human political campaigns fail, AI-enabled ones will, too. But we may not be far off from a world where they succeed.
We also need to consider how generative AI content reaches us. Currently, this mostly happens through social networks, which actually place further limitations on AI’s ability to have an impact on elections in 2024. Why? Because the contemporary media ecosystem is incredibly fragmented, and is only becoming more so. As the previously concentrated user group from Twitter/X fragments into separate communities on Threads, BlueSky and Mastodon, and as Facebook’s users head to Instagram, TikTok and beyond, this fragmentation slows things down and limits the effects of bad content to ever-smaller corners of the internet. It makes detection and explanation of the phenomena more difficult, but it also makes it more marginal.
These three factors alone – the short-lived impact of most fakes, the need for them to be attached to larger narratives and the fragmentation of our information diet – are our first layer of “natural” protections against the potential harms of generative AI when it comes to elections.
But we shouldn’t rely solely on these features of the landscape to protect us. Instead, we should take a pro-active position in favour of democracy and elections as human activities that are by and for people.
We can do this by limiting excessive technological personalization in election campaigns, whether by AI, social media or paid ads. Many voters do want to learn what a politician can do for them, but they also want to know what’s being offered to everyone else. Just as candidates can’t speak to each of us individually, nor should thousands of AI agents be able to do so on their behalf. We must prevent this from becoming a normal practice for political campaigns.
We must also be mostly able to trust that someone is who they say they are online. Anonymity plays an important role in voting and we should cherish it when it’s needed, but otherwise, it should be discouraged in general, particularly when votes are at stake. We should expect to know, in almost all cases, who is saying what, and therefore what motivates their position. AI should never hide, or be hidden from us.
Social-media platforms must play an active role, too. Not all of them are going to end up as big players in generative AI (though all of them are trying to be), but as the largest carriers of the information we see, they will need to help users understand which content is AI generated, monitor for large-scale manipulation of accounts and comments, prevent co-ordinated AI harassment of politicians and civil society, and allow for proper research into and auditing of their systems.
Together, we need to push back hard against those who try to harm democracy, a problem that has increasingly come from within in recent years. Disinformation often comes from the very top of the tree, as do polarization and fragmentation. Where leaders lead, society, at least in part, seems to follow. We should set a higher standard, and collectively draw the line that shouldn’t be crossed, by calling out politicians and campaigns who use AI to mislead us or our fellow citizens.
The AI frenzy is very much here, and when it comes to threats to democracy from generative AI, we should be prepared, even if the risks don’t quite end up matching the hype. We should also prepare ourselves with a particular perspective – one that puts democracy, rather than AI’s progress, at the centre of our thinking.