Spotting the Fabrication: Techniques for Discerning AI-Created False News Content

Whether it’s fake images , deepfake videos, or ChatGPT, AI is upending the way we interact with the world — and our ability to determine truth. There’s no foolproof method for detecting bad information, but there are a few tools you can use to figure out if what you’re reading is real.

AI Can Detect Fake News — It Can Also Write It

Fake news is hard to fight. Inflammatory lies or misinformation that plays to people’s emotions and biases spread much, much faster than the truth. And while some AI tools are being used to fight this, others can just as easily be used to create fake content for a disinformation campaign.

It’s more than just fake articles; fake medical information, an AI photo, or a simulated voice clip can all get reported on as though it’s real news if the reporter doesn’t fact-check thoroughly enough.

How to Detect A Fake News Article

AI versus AI detection will likely be an ongoing arms race for the foreseeable future. As soon as one side improves, the other side innovates, upgrades, or changes tactics. But that doesn’t mean you should give up on trying to decide what’s real.

AI Tools

Something you use every day without thinking about it is already a sort of AI bullshit detector: your email spam filter. It uses AI text analysis to determine which emails are likely to be false due to their headlines, text, and/or contents of the message.

AI text detection programs can scan news articles in much the same way. It can flag articles that have headlines which don’t match the article, come from known disinformation websites or sources, and include unverified information.

Some AI-powered fact-checking tools include:

• The Factual : A newsletter, Chrome extension, and website that monitors and grades news articles.
• Grover : An AI program that taught itself what fake news looks like by producing its own fake news stories. It’s still in the prototype stage, but you can test it out with real news stories, or have it try and generate fake news copy. Its creators plan to use it to detect fake AI articles in the future and claim an efficacy rate of over 90%.

Tools like these can only do so much, so don’t rely on them completely. Open AI, the company behind ChatGPT, had to shut down its own AI detection tool because it wasn’t accurate enough .

Training Your Eye

It’s getting more difficult, but humans can still detect articles written by an AI content generator if they know what to look for. You probably won’t be 100% accurate, but you can at least train yourself to recognize misinformation.

Identify the source of an article that looks suspicious. Who posted it on social media? Was it an account you implicitly trust, or one that’s totally new to your feed? Have you heard of the publication that put this article out? Who wrote it, and what’s their background?

Look at the context of the article, or the person sharing it. Was a particularly inflammatory quote taken out of context and used to promote the article? Was that quote even in the article at all?

One of the most obvious signs of misinformation is emotional impact. If you see a headline that makes you immediately angry or sad, take a second and evaluate what that headline is actually saying. What is this article trying to do?

The creators of dis- and misinformation campaigns know very well that negative information spreads like wildfire, so it follows that they’d program AI to make the same type of content. Sharing content widens its reach and adds to its credibility, so think twice before retweeting that inflammatory article, even if you’re pointing out how ridiculous it is.

Humans Will Probably Always Be in the Loop

To recognize mis and disinformation more effectively will take multiple groups of people working together with technology. Until platforms like Facebook have effective methods for blocking it, or we have laws that stop its creation in the first place, the best thing you can do is stay up to date on the latest methods of detecting fake articles, AI-generated images, and deepfakes.

Also read: