AI-generated images already fool people. Why experts say they'll only get harder to detect.

Two weeks before Donald Trump set foot inside a New York courtroom, images of the former president being tackled and carried away by a group of police officers went viral on social media.

The sensational images first appeared on Twitter days after Trump claimed his arrest was imminent. They then migrated to other platforms, amassing tens of thousands of likes and shares along the way.

There was just one problem: They were entirely fake.

The realistic-looking images of Trump's fictitious encounter with police, along with a wide variety of other fake images that have spread online recently, were created with image generators powered by artificial intelligence.

At the moment, it's still possible to look closely at images generated by AI and find clues they're not real. One of the Trump arrest images showed him with three legs, for example. And a fake image of Russian President Vladimir Putin being arrested showed only three fingers on one of his hands, a common issue with these sorts of images.

Fact check: Photos showing Trump arrested by law enforcement are computer-generated

But multiple experts told USA TODAY it's only a matter of time before there will be no way to visually differentiate between a real image and an AI-generated image.

"I'm very confident in saying that in the long run, it will be impossible to tell the difference between a generated image and a real one," said James O'Brien, a computer science professor at the University of California, Berkeley. "The generated images are just going to keep getting better."

With AI, fake images can be made easier and faster than ever before

The idea of making fake images is far from new. For years, experts have had the ability to create realistic-looking fakes using photo editing software.

"A skilled artist could create an image that you really need to study to determine if it's real or an artist's creation," O'Brien said.

But what used to take hours of effort and at least some level of expertise can now be done in minutes by someone with no training. All it typically takes to make an AI-generated image is a written prompt, limited only by the user's imagination.

The availability of the technology and the ease with which it can be used has made it possible for almost anyone to make realistic fake images, said V.S. Subrahmanian, a computer science professor at Northwestern University

Fact check: Experts say images showing Putin's arrest are fake

Subrahmanian pointed to a fake image of Pope Francis wearing a white puffer jacket as an example of an image created in obscurity that suddenly spread across social media, with many people believing it was real.

The image was created by a 31-year-old construction worker from the Chicago area who, after taking psychedelics, decided to create the images on Midjourney, a popular AI image generator, according to Buzzfeed News.

"It's kind of comical, but at the same time it's chilling because you know if you can do this, you can put the pope or any world leader, or any of us, in some kind of position that we would prefer not to be in," said Subrahmanian.

And the AI problem will get worse quickly.

The technology behind AI image generators is improving so rapidly, experts say, the visual clues that now give away fake images will disappear entirely in the not-so-distant future, making it virtually impossible to tell what's fake and what's not.

"People aren't going to know what to believe and what not to believe," Subrahmanian said. "People are going to end up believing things that are false."

O'Brien agreed: "Are we going to have to suspect everything of being fake? I think the answer is yes."

Education, transparency key to navigating AI landscape, experts say

Subrahmanian said AI's increasing complexity and ease of access exacerbate the existing misinformation problem, since many users already "don't always exercise a lot of judgment when they see something."

The primary blind spot is online claims that align with the user's view of the world, according to a study O'Brien worked on examining how people evaluated the authenticity of online images.

The study found participants with higher digital media literacy were more likely to be skeptical of images, but it also found those same participants were more likely to deem images credible if they aligned with the user's pre-existing views.

"We're manipulated by people showing us things we would like to believe," O'Brien said. "By showing us things we would like to believe, they're able to lead us in whatever direction they want."

Ari Lightman, a digital media professor at Carnegie Mellon University, said people need more education on the topic from a younger age, even if the technology gets to a point where the average person can't tell the difference between a real image and a fake image.

"We need to understand that we can't stifle innovation or creation, but there are some drawbacks," he said.

Fact check: Viral images of a beached great white shark are fake

Fact check: Images show artificial intelligence concept, not satanic hotel in Texas

There also needs to be more transparency about where information is coming from online, said Lightman, who suggested AI-generated images could be labeled with a watermark or a stamp.

While it's possible AI could be used to detect fake images, even that solution has its own limitations, O'Brien said.

"Then you just end up with an arms race," he said. "As soon as I build a new detector, people can go look at it, see how it works and figure out a way to make the generator trick it."

So, O'Brien said, instead of relying on technology to solve the problem, people's attitudes toward images they see online need to change. He compared it to the way most people now treat scam emails that promise large amounts of money.

People will need to develop a similar filter for social media, especially if an image – or a video, or any piece of media – seems too good to be true. And that change needs to happen quickly, O'Brien said, preferably before it becomes impossible to tell apart real and fake images.

"By then, I hope we've gotten to a place where we don't trust images as much," he said. "If not, we're in trouble."

This article originally appeared on USA TODAY: AI-generated images only going to get harder to detect, experts say