From the general election in Bangladesh in January to the one in Ghana in December, this year will see around 2 billion people across the world casting their votes. While online misinformation and other kinds of digital meddling have long been a concern around global politics, the recent advent of readily accessible generative artificial intelligence tools makes 2024 a real test case for this technology in the political sphere.
Throughout the year, Rest of World will be tracking incidents of AI being used in the context of elections — with a focus on countries outside of the West. We’re launching with a selection of examples starting from the beginning of 2024 and will keep updating the tracker as new incidents occur. We’re considering all incidents for inclusion — the good, the bad, and the memes. It could be content generated directly by a political party or candidate, or created by fans, bad actors, or other third parties. So far, we’ve seen a range of different AI tools being used to create different types of media, including text, images, audio, and video.
Some incidents show innovative approaches to political campaigning, such as a ChatGPT-powered virtual candidate designed to inform voters in the Belarusian elections while flouting censorship, or AI videos allowing Pakistan’s former Prime Minister Imran Khan — who is currently imprisoned — to deliver speeches from behind bars. Others highlight many of the concerns experts have around mis- or disinformation and fake news: A spam campaign to discredit Taiwan’s president was traced back to an actor associated with the Chinese Communist Party, while deepfake videos showed Bangladeshi candidates withdrawing from the elections on election day. Of course, there are also plenty of examples of AI being used for fun. In India, Prime Minister Narendra Modi turned into a popular singer, and in Mexico, voters were confronted with spoofed Starbucks cups.
For this project, we have engaged reporters in several countries that are holding elections this year to research and submit incidents as they happen. We’re also keen to work with other organizations and researchers who are monitoring tech’s influence on elections, and we’re encouraging readers to share any examples they come across online. If you’ve seen an instance of AI being used around elections local to you, please fill in this simple form, and we’ll investigate it.
Our goal is to create a database of examples that can be used to understand the many ways in which AI is being deployed around elections. Wherever possible, we aim to preserve the AI content so that it is available for future reference and study, even if it is removed from or by the platforms originally hosting it. In rare cases, we may opt not to publish the content if doing so could have a particularly damaging consequence.
Some social media platforms have developed specific rules around AI content, and/or have existing rules that may apply to manipulated media, especially if it is used for harmful intent such as misinformation. But given the recency of generative AI tools, it’s hard to tell how effective these are. Previous experience suggests that content moderation is less effective in languages other than English, so we’re particularly keen to see which platforms AI content ends up on and how it spreads. Where possible, we’ll be tracking if any content is removed from specific social media platforms.
The nature of AI content also means it can be difficult to verify exactly what’s what, especially when those who post it are not transparent about using AI tools. Where the use of AI may be ambiguous, we’ll put extra effort into reporting by reaching out to researchers and fact-checkers with expertise in AI. We will be transparent with our readers about what we know (and what we don’t), and we’ll always credit the original sources. If new information comes to light, we’ll update the entries. If you see a post you think needs updating, please let us know.
To keep up with new entries, follow Rest of World on X, or sign up for our global newsletter.
This work is supported by Luminate
