Once AI can create endless viral videos, good luck switching off social media

Imagine humans … watching humans …watching TikTok … forever

Column "I finally had enough," a friend recently confessed. "Lying in bed very late at night, watching one video after another on Instagram. Suddenly: midnight. How did that happen? I put the phone down, turned off the light. And couldn't sleep. Mind still whizzing after all those videos. Finally, I got up, turned on the light, reached for my phone … and deleted Instagram."

It's difficult to imagine that she's alone in feeling the need for some boundaries, given how hard it can be to tear our eyes away from the endless play of brief, punchy videos that social media parade before our eyes.

That difficulty is intentional. We know that social networks and video platforms work hard to keep you watching just one more video … then one more … and one more … ad infinitum.

Ever since TikTok became the fastest app to reach 100 million users, Facebook (err, Meta) has been running scared – doing everything it can to replicate the TikTok experience. Long dominated by text and cat photos, social media began its convergence with video.

Today, content length marks the biggest difference between Netflix and Instagram. Because of their novelty – and because you can swipe them away the moment any feelings of boredom arise – short-form videos have proven to be more engaging, and more "bingeable". Mix that with "recommendation engines" that continuously optimize user interaction for "stickiness" and – voila! – an app that users simply can't look away from until their bodies, pushed beyond natural limits, finally send signals that override the brain's desire for more stimulation.

With so many people seeking the gratification of Likes – or that other dopamine hit, advertiser dollars – social media services have built a deep catalog of content to rate and test against their audiences. Videos that hit the sweet spot travel up the charts until – in a matter of hours – they become global hits. And the creators – if they're both smart and lucky – establish themselves in the entertainment firmament.

That rags-to-riches fantasy of creativity and success feels profoundly out of step with the postmodern world of early 21st century media. Things may have worked that way once – back when artists were "discovered" – but these days a media powerhouse cannot afford to leave anything to chance.

Last decade that used to mean test marketing – packaging the artist in different ways to find the best approach to bring them to the public - but even that now feels a bit old-fashioned, slow and prone to error. Why can't we just optimize the whole thing, head to tail, engineering something specifically for media consumers?

That's the hidden-in-plain-sight secret of the world's biggest bands – all from South Korea – each featuring a preposterously large troupe of generically-interchangeable-yet-individually-themed members, everything about them delivering a delightful yet empty experience.

That formula has been all-but-perfected, meaning it can only be a matter of time before the formula becomes an algorithm.

The explosive reach of Generative AI has, in just the last ten months since the launch of OpenAI's DALL-E, led to a galaxy of related applications.

Consider Riffusion, capable of generating an endless stream of "original" music generated by text prompts. Or the Motion Diffusion Model, generating an endless stream of human animations from text prompts. Then there's Meta's Make-a-Video – an endless stream of video, also generated by text prompts.

Surely some clever entrepreneur is already hard at work putting them all together to write a song, animate a human "singing" it, then place the artificial artist within a truly awe-inspiring video.

All without any actual human talent involved.

Would such a piece of synthetic pop be good? A worthy question – but one entirely beside the point.

These automated systems will be able to generate millions of releases every day, slipping them into social media feeds, then observing the results.

The unsuccessful will be ruthlessly culled while the fit are just as tenaciously promoted. For the price of a few programmers (and a significant cloud computing bill) any media company will have all the "stars" they'll ever need, generating an endless stream of "good-enough" hits to keep the whole operation ticking over.

If all of this sounds a bit familiar, a similar argument appeared on this site three months ago. In "You get the internet you deserve" Nicole Hemsoth described an analogous cycle of generative AI systems gumming up the works within the web's knowledge space, triggered by an exponential increase in the number of low-quality – but highly optimized – "facts". It may hit SEO first, but the rest of the internet looks to be following close behind.

We can already see this in "Nothing, Forever", the endless Seinfeld episode that streamed on Twitch, and HackerFM, a "podcast" about current tech topics – written and voiced by Generative AI. These are simply the first, crude droplets in what will quickly become a vast ocean of content created by machines with one goal: optimizing engagement.

Before this year ends, our social media will be a combination of human and synthetic short form videos, each vying to be more engaging than the last. But humans have limits. (Vexingly, they also expect to be paid for their work.) The machines don't. They will outproduce and overwhelm any level of human contribution, relying on user surveillance and analytics to improve the "stickiness" of their productions.

And they will never, ever stop. ®