TikTok, the Generative AI Behemoth (of sorts) - And What Comes Next
Will synthetic influencers soon start calling the shots?
Unlike fully generative systems like ChatGPT and Midjourney, TikTok doesn’t synthesize media out of thin air. But few of its videos would ever be created if it weren’t for the platform’s success and massive audience. That audience is built by an algorithm that almost telepathically chooses the unique flow of videos that each user will personally find irresistible. In this sense, TikTok’s videos -- like ChatGPT’s prose and Midjourney’s images -- are effectively summoned into existence by a powerful AI.
Of course, videos on TikTok (and its pivoting mimickers like YouTube and Instagram) aren’t responses to the unique desires of specific users. But most of them are responses to the aggregate desires of many users. Those desires are revealed through the comments, likes, and view counts of the current crop of videos. Metrics that are beacons and bug lights for armies of attention-hungry creators deciding on what video to produce next.
Fresh memes, styles, and formats constantly emerge. Many would never be conceived of if they hadn’t emerged from this AI-driven stew of experimentation, mimicry and mutation. When a new trend emerges, armies of creators hop onto the wave, competing for likes in the latest arena that viewers are suddenly into.
The result is the fastest engine ever built for detecting, shaping, and satisfying the media cravings of a mass audience. The bandwagon is seductive enough to get countless giant brands and studios chasing TikTok trends as frantically as the most insecure wannabe influencer. And the flywheel of viewers, creators, and the almighty algorithm generates oceans of video that would otherwise not exist.
For all of this, TikTok is but a shadow of its future self. Or of its successor, if regulators kill it or some hot new thing dethrones it. Because imagine this engine becoming a fully realized generative AI. A TikTok that can direct, shoot, and produce its own videos in moments. That can A/B test flurries of variants for each video across demographics and psychographics. That tweaks videos constantly as the metrics flood in, personalizing and optimizing to ruthlessly maximize addiction. Or product sales – or recruits for the next digital political cult.
For all their influence, creativity and occasional flashes of genius, today’s TikTokers can’t create videos continuously or instantly. They can’t afford to create dozens of video variants for A/B testing. Even if they could, they’re not data scientists. And even if they were, even the mightiest influencer lacks the power to demographically cross-test blizzards of videos in statistically optimized batches. Those controls lie strictly in the hands of the closed master algorithm.
Sophisticated online advertisers have had some of these powers for years. And the mightiest distributed intelligence in history – today’s market economy – has upvoted the results with a half trillion dollars in spending each year. Imagine that kind of budget behind a generative platform that immaculately optimizes media for addiction, purchases or persuasion.
Though I’m sure a fully generative TikTok will serve up customized media on demand for single-user consumption like Midjourney, its main draw will still be viral hits that resonate with large groups of people. Much of TikTok’s allure comes from its community features, and from experiencing things alongside millions of others. Plus the selection algorithm that drives it all is powered by feedback compiled from many users who have seen the same thing.
This means there will still be a huge role for influencers -- and that a fully generative TikTok could usher in a golden age for them. The savviest ones could drive audiences to unheard of levels of scale and engagement by using generative tools to radically increase their output, and then optimize it based on the signals gleaned from a lot more shots on goal.
But things could go very differently -- because a fully generative TikTok may have no need for human influencers. Ones who earn big revenue splits. Who can’t constantly produce at digital speeds. Who can’t be shot in several exotic situations simultaneously. And who can’t repeatedly alter their eye color, skin tone, and outraged exclamations until the optimum level of engagement is reached on the 85th video of the day.
Synthetic influencers will be capable of all that. And not only will they work for free, but will be wholly owned, valuable and salable assets of the platform that creates them.
As with all generative AI scenarios, it’s too early to have much of an inkling of how this will play out. If things go well, it could uplift billions of us by entertaining and educating us in magical ways we can’t yet imagine. If they go badly, it could boost the division and disinformation sewn by today’s social media to exponential extremes.
But however you cut it, this looks like one of the biggest business opportunities of all time. And there’s no chance it will be allowed to flourish on American soil with a Made in China stamp on it.
"… If things go well, it could uplift billions of us by entertaining and educating us in magical ways we can’t yet imagine…"
The described scenario of massive auto generation and auto hyping to bring the optimum happy hormone producing content (junk?) to the masses only brings a Wall-e kind of world to my mind.
But I'm curious as to other opinions or insights about that doom(?) scenario.
China is not the problem, people who use these technologies in the bad way are.