OpenAI promises tools against electoral disinformation

The American company OpenAI, creator of the ChatGPT virtual robot, announced that it will launch tools against disinformation this year, when dozens of elections are held in countries that bring together half of the world's population.

Conversational tool ChatGPT has fueled a revolution in Artificial Intelligence (AI), but it has also prompted warnings from experts that such programs could flood the Internet with misinformation and influence voter decisions.

With numerous elections taking place this year in countries like the United States, India and Great Britain, or in smaller ones like Uruguay, OpenAI declared on Monday (15) that it will not allow its technology – including ChatGPT and the DALL-E 3 images – be used in political campaigns.

“We want to ensure that our technology is not used to undermine” democratic processes, OpenAI said in a blog post.

“We are still working to understand the effectiveness of our personalized persuasion tools. Until we know more, we do not allow the building of apps for political campaigns and pressure groups,” the company added.

– Misinformation: a risk –

Misinformation, including that generated by AI, is one of the world's biggest near-term risks and could cause harm to newly elected governments in major economic powers, the World Economic Forum warned last week.

According to experts, the fear of electoral misinformation began years ago, but easy access to powerful text generators and AI-based images has increased the threat, especially if users cannot easily distinguish whether the content is false or manipulated.

Yesterday, OpenAI reported that it is preparing tools that make texts generated by ChatGPT more reliable, in addition to offering users the possibility of detecting whether an image was created with DALL-E 3.

“Earlier this year, we will implement Coalition for Content Provenance and Authenticity digital credentials, an approach that encodes details about the provenance of content using encryption,” the company stated.

Also known as C2PA, this coalition aims to improve the identification and tracking of digital content. Its members include large companies such as Microsoft, Sony, Adobe and Japanese image specialists Nikon and Canon.

According to OpenAI, ChatGPT will direct users to authoritative websites when asked procedural questions about the US election: for example, where to vote.

“The findings from this work will inform our approach in other countries and regions,” the company said, adding that DALL-E 3 has “fences” to prevent the generation of images of real people, including candidates.

OpenAI's announcement comes in the wake of measures taken by technology giants such as Google and social network Facebook to limit electoral interference, especially through the use of AI.

In its work, the AFP has already debunked false material such as that of the President of the United States announcing the recruitment of soldiers, or the short film of Democrat Hillary Clinton, proclaiming her support for the presidential candidacy of the governor of the state of Florida, Republican Ron DeSantis .

This type of audiovisual material circulated on social media during the campaign, for example, for the Taiwanese elections held this month, according to studies carried out by the AFP Fact Check team.