Will AI Create More Fake News Than It Exposes?

(Bloomberg Opinion/Tyler Cowen) -- The best large-language models can already write like humans, especially if prompted properly. Photos and images can be faked at low cost. Yet-to-be-released technology can create convincing voice simulations. There are signs that some academic papers contain traces of GPT-4. If even professors are faking it, then surely the dam has burst.

In other words: As the AI revolution spreads, so will the AI-enabled fakes. This is an inevitability, but it can be managed — so long as consumers and producers make significant adjustments in the way they use the internet.

The post-AI internet will have a lot of free, advertising-supported content full of AI fakes, designed to appeal to the lowest common denominator. The first question is whether anyone will even notice. After all, there are plenty of low-quality and unreliable outlets out there already, mostly not driven by AI.

Technology to detect AI fakery will evolve and improve, and there will be an arms race. So long as the detection works, the free, ad-supported internet will continue. Readers, listeners and viewers will be able to use their own AIs to find content they like — and a percentage of them may prefer the fakes. After all, the Weekly World News still has an audience.

Related:AI Basics: A Quick Reference Guide for IT Professionals

Still, even assuming the fake detectors work, mainstream news sites will face additional competition. Alternative sites will spoof their content, adding twists — sometimes partisan, sometimes sensational, sometimes sponsored by foreign governments. Some of the most effective competitors might be 98% legitimate, but a 2% fake rate along some critical dimensions, such as coverage of foreign wars or salacious personal scandals, could be significant.

Of course institutions will evolve to limit the scope of these problems. The best and most authentic material will probably end up in highly curated sites, available only on a subscription basis. Perhaps those sites will occasionally make mistakes and post fake material, but their authenticity will be their major selling point.

These sites will work hard to create the kinds of content that AI cannot easily spoof. For instance, columnists with well-known personalities will become more important, relative to more generic but still first-rate writers. Even if an AI can copy the style of Paul Krugman, for example, it cannot be Paul Krugman, and many of Krugman’s readers care what he thinks about an issue. They don’t want the AI cloned version, no matter how high its quality. So media outlets will do more to promote the personal brands of their authors.

These sites will also make their content smarter and more complex. That will make it harder for free, AI-driven sites to produce derivative versions. To give a simple example, Sora, a service from OpenAI, can make a good two-minute video, but making much longer videos may not be cost-effective. So some media outlets might start publishing longer videos.

Or how about podcasts with lots of rapid back-and-forth on complex issues? Again, it’s hard to imagine current versions of AIs engaging in that kind of debate. There will be an arms race here too, with subscription-funded media always trying to stay ahead of the AIs.

The bad news is that journalists will have to work harder. The good news is that subscribers will end up smarter and better-informed. This part of the arms race will point in the direction of ever-higher quality.

Media outlets can also be expected to move further into the business of live events. Holograms aside, it is hard to imagine the AIs doing anything comparable, not for a long while at least.

Over time, as the AIs continue to improve, hybrid subscription sites might emerge, with both human- and AI-produced content. They will delineate who (or what?) was producing which content, rather than trying to obscure it. Articles authored by humans might even have an attached AI bot that could offer further detail.

People are right to worry about AI fakes. I share their concern. But I also have faith that the media will be innovative enough to adjust to these new realities.