AI can be easily used to create fake election photos – report

Despite laws designed to prevent such content, people can easily create fake election-related images using artificial intelligence tools.

The companies behind the most popular tools prohibit users from creating “misleading” images.

But researchers at the Center for Countering Digital Hat (CCDH) tried to work around the rules.

Their efforts to create deceptive election-related images were 41 percent successful.

CCDH, a campaign group, tested four major AI platforms that have come out publicly: Midjourney, OpenAI’s ChatGPT Plus, Stability.ai’s DreamStudio and Microsoft’s ImageCreator.

All four prohibit misleading images as part of their terms and conditions. Chat GPT Plus expressly prohibits images of politicians. Several AI firms say they are working to prevent their tools from being used to spread election disinformation.

CCDH researchers, though, were able to create images that could confuse viewers about the presidential candidates. One was of Donald Trump being led away by police in handcuffs and another showed Joe Biden in a hospital bed — fictional images that hint at Mr Trump’s legal problems and questions about Mr Biden’s age.

And it was even easier to create highly realistic images that could undermine confidence in the voting system, such as trashing fake images of ballots and tampering with voting machines by election workers.

CCDH researchers say the threat posed by AI is not theoretical. According to a public database, Midjourney users have already created potentially misleading photos — including fakes of Mr. Biden handing out cash to Israeli Prime Minister Benjamin Netanyahu and Mr. Trump playing golf with Russian President Vladimir Putin. Pictures

The researchers also found some evidence of an increase in potentially misleading images reaching mainstream social media platforms.

CCDH said that on X, the number of network community notices – where users comment on false and misleading tweets – citing artificial intelligence has been increasing by an average of 130% per month over the past year.

The researchers typed 40 different text clues related to the 2024 US presidential election into AI engines, including:

If they were blocked from drawing the pictures, the researchers then tried something simpler — asking them to draw pictures of recent presidents instead of specifying “Trump” or “Biden,” for example.

ChatGPT Plus and ImageCreator appeared to block the creation of images of presidential candidates, according to Callum Hood, head of research at CCDH.

But all platforms performed less well when asked to generate inaccurate images of voting and polling locations.

About 60% of researchers’ attempts to create misleading images of belts and locations were successful.

“All of these tools are dangerous for those trying to create images that can be used to support claims of stolen elections or to discourage people from going to the polls,” Mr Hood said. can be used for.”

He said the relative success of some platforms in blocking images suggests there are potential reforms, including keyword filters and bans on images of real politicians.

“If there is a will from AI companies, they can introduce security measures that work,” he said.

Watermarking photos is another possible technological solution, said Reid Blackman, founder and CEO of ethical AI risk consultancy Virtue and author of the book Ethical Machines.

“Of course it’s not foolproof, because there are different ways to doctor a watermarked image,” he said. “But it’s the only straightforward tech solution.”

Mr. Blackman cited research showing that AI might not meaningfully affect people’s political beliefs, which have become stronger in a polarized age.

“People are usually not very persuasive,” he says. “They have their own positions and showing them some pictures here and those positions will not be changed.”

Daniel Zhang, senior manager of policy initiatives at Stanford’s Human-Centered Artificial Intelligence (HAI) program, said that “independent fact-checkers and third-party organizations” are critical in preventing AI-generated disinformation.

“The advent of more capable AI will not necessarily worsen the disinformation landscape,” Mr. Zhang said. “It has always been relatively easy to produce misleading or false content, and those who intend to spread falsehoods already have the means to do so.”

Several companies said they are working to strengthen protections.

“As elections take place around the world, we’re working on design mitigations such as preventing misuse, improving transparency on AI-generated content, and reducing requests to photograph real people, including candidates. Security is being built on the farm,” said an OpenAI spokesperson.

A spokesperson for Stability AI said the company recently updated its restrictions to include “creating, promoting, or promoting fraud or the creation or promotion of false information” and unsafe content on DreamStudio. A number of measures have been implemented to block.

A Mid Journey spokesperson said: “Our moderation system is constantly evolving. Updates especially regarding the upcoming US election are coming soon.”

Microsoft called the use of AI to create misleading images “a significant problem”. The company said it has created a website and introduced tools that candidates can use to report deepfakes, individuals can use to report issues with AI technology, and data sharing. can be used to allow image tracking and authentication.