As 2 billion people head to the polls around the world, a key concern for many is the role new AI may play in election campaigns. This technology can exacerbate problems with disinformation and online harassment.
So far, there is still not much evidence that AI will play a significant role in this year's elections, despite some well-known examples. But elections in the Global South may remain hidden, with many news outlets focusing on votes in the US, UK and European Union. Rest of World, a nonprofit online publication covering the impact of technology outside the Western world, aims to address this issue with its new project, the 2024 AI Elections Tracker, launched on April 16. Masu.
The tracker is a continuous log of examples of AI being used in election-related uses. Examples are flagged to the team, verified, and reported before a summary is added to the tracker timeline with context and a link to the original image, video, or post. Rest of World's features director, she spoke to Victoria Turk. She describes this project in her own words.
1. Project overview
Victoria Turk: “The aim is to collect examples of AI being used in the context of elections around the world. To give you a little background, 2024 is a big year for elections. is eligible to vote. This comes after a huge boom in highly accessible generation AI tools such as ChatGPT, AI image generators, and audio and video tools.
There has been a lot of discussion about what this means for democracies in terms of how these tools can be used or abused in campaigns and the spread of misinformation and disinformation. Ta. We wanted to look at what's really going on and go beyond theoretical concerns to see what's happening on the ground. We investigate how countries around the world use generative AI tools before and after elections. ”
2. The problem you wanted to solve
VT: “From a top-level perspective, we all know that generative AI tools are being used before and after elections. What we want to know is real-life examples of this happening, especially outside of Western countries. I think there's probably a lot of attention paid to English-language content around the US elections and the UK elections.Western media, in particular, aren't paying much attention to elections in Latin America, Africa, or Asia.
We continue to do this throughout the year and hope that in the future we will have a database of content that we can reference to see overall trends. That database is open to other people so they can come back and see what happened in different places.
It is necessary to say that we are not only focused on the negative uses of these tools. So when people think about elections and generative AI, they automatically think about disinformation, disinformation, deepfakes, and deliberately misleading voters. We're seeing some of that, but there are other creative uses for this technology as well. In some cases, we see political parties and campaigns using this technology to enhance their reach to voters, disseminate campaign materials, and provide information.
We saw this from former Pakistani Prime Minister Imran Khan's statements. His party was using his AI to allow him to video address his supporters despite being in prison. In Belarus, we saw opposition parties create AI candidates powered by ChatGPT that could answer people's questions in a way that might be dangerous to an actual opponent in that political context.
I've also seen a lot of similar fun stuff. We don't think about whether this is good or bad because memes and people create memes that make fun of the opposition or celebrate the candidates they support. To capture what's going on. ”
3. Problems they encountered
VT: “A lot of this content is written in local languages, so it's important to have journalists who speak the language and are familiar with the situation. We can't do it without that knowledge. Verifying whether it's actually AI is a big challenge, and sometimes it's very obvious. People often say they created something using AI. They're very upfront about it, sometimes even touting their mastery of using these new tools.
In other cases, they may not acknowledge the use of AI or, in the case of deepfakes, may be attempting to pass off content as legitimate when it has actually been altered. If that's the case, you'll need to do a little detective work to find out if the content was generated by AI. ”
4. How did they find the solution?
VT: “So far, we have found that there is no one-size-fits-all solution for performing these checks. There are several tools that can help, but I don't think any of them are foolproof, so we're using the following resources as much as possible:
We use some of the online tools available, but we also go to meet researchers in the field in person to get their opinions, and visit local fact-checking organizations to see what they think. Masu.
In some cases, we look for digital clues. Therefore, the mouth movements may not match the audio, which could suggest that the video has been tampered with. But it may be more situational. If you are aware of the local political situation, you will know that it is probably very unlikely that the particular case shown in the video actually happened, for example. So we do more traditional reporting and the people in the photos were there at the time.
We also built update functionality into our trackers from the beginning. We understand that stories may change or more information about an entry may become available, and we feel it necessary to add that information for our readers. ”
5. How did we test the tool?
VT: “We launched with the initial idea that this was something that would grow. A few things happened between early January and the April 16 launch date that led us to launch this. We are currently updating new entries as we find them. We are trying to accommodate as many as possible.
The number of entries is expected to rise and fall with elections, so as elections take place, we expect to generate more content related to a particular election.
At the moment, most of the entries are about Indonesia, Pakistan, and Bangladesh, where elections have already been held. As the Mexican and South African elections approach, we expect to see more happening before and after the elections. This is a new format for us and unusual in that it's a project that we work on consistently and keep changing throughout the year, so it will be interesting to see how it develops. ”
6. How the tool actually works
VT: “I think one of the things we have at Rest of World compared to a lot of publications is a global team and a global network of contributors, so to find and report on entries, , we rely on it a lot” for this tracker. There are journalists in almost every region except the West. North America and Western Europe are not covered. We have a network of contributors in different countries tasked with finding examples of AI being used during or around elections in their countries.
They can pitch and report those examples to us, see that AI is being used, who's doing it, why they're doing it, what tools they're using. We're trying to find out details like how the content is spreading. They are all entered into one tracker. This is important. Because a lot of this content can be written in local languages and has local political backgrounds that one person sitting in one country doesn't get anywhere in the world. This is because knowledge of this may be required.
We are also collaborating with other organizations in this area and are encouraging people to share with us what they see, making sure we cover as many bases as possible. We're sharing it back with them to make sure.
Finally, we're asking our readers to submit examples they've seen on social media. We try to get as many examples as possible so we can examine them and see if they meet the criteria for inclusion in our tracker. ”
7. How can we improve our tools?
VT: “The first thing we want to do is collect as many examples as we can. Even on the first day of launch, we were really pleased to get some entries submitted by readers to be included in the survey. That's great because we're going to have more incidents, so it's important to keep that momentum going to make this as useful as possible.
We're interested in seeing how people end up using it and what kind of content we have. Currently, it is possible to filter content by month, country, and platform where the content is spread. We're interested in how useful that filtering technique is. Maybe it can be developed in the future.
One of the parts of this that we thought was really important from the beginning is preserving the content that we're talking about here. Much of this content may be removed by the platform or the people who originally published it. We think it's very important to preserve content so that when people come back to see it in the future, it's meaningful and they can see what it was exactly. We are working to include it.
If we judge that there is a possibility of causing special damage, we may refrain from posting it. But in general, we want to build a library that can be referenced by readers, researchers, and anyone who wants to use this tool.
We’re also interested in tracking how platforms approach election-related AI content. Some have developed special rules for this. Others are still working on them or changing them along the way. We want to track how different platforms react to this content: whether they remove it or label it. Are we seeing certain phenomena on certain platforms? That's also what we try to include in our entries. ”
8. How this can help newsrooms
VT: “We’re just putting everything out there so people can see it and use it for themselves. Others writing about the AI phenomenon around elections can use these examples as inspiration. This can be a useful resource if you have a variety of publications focusing on different regions or specific technology trends and want to look up specific incidents related to them. The more entries we have, the more interesting the data analysis can be by looking at the data together to see if certain trends emerge.”