How AI will affect what you see in your feed
And the problem it can't fix
One of my favourite pieces of reporting on creativity is a 2020 Uproxx interview with Jack Barth. You have likely not heard of him. After years of grinding out a living as a comedy writer he caught a big break at the age of 62 when he sold his first feature script, the story that eventually became 2019 Danny Boyle-directed, Richard Curtis-written Yesterday. You know the movie pitch: man wakes up as the sole person in the world who remembers The Beatles. Barth gets a ‘story by’ credit along with Curtis on the movie.
The piece outlines how Barth’s original pitch, that the protagonist can remember every Beatles song but still can’t catch a break as a musician, is taken and slowly developed into Curtis’ screenplay where the main character finds instant success before (spoilers) realizing success isn’t everything and giving it all up for love. The whole piece is worth a read, but it’s telling that Barth, the writer who never got his big break, sees the story going one way, and Curtis, the writer who meets Rowan Atkinson at the University of Oxford and has back-to-back success from there, sees it going another way.
I think about Barth every time I read about social media algorithms. He spent a career trying to get his work seen and his gatekeepers were studio executives and agents. But there are thousands of creatives - writers, videographers, artists, and musicians - starting out today who haven’t got near the level needed to approach human gatekeepers. They’re just trying to break out of their little corner of the internet and they are thinking about algorithms and AI. Thinking about algorithms and AI far more than any creative person should.
In a previous job, I managed a project rating social media videos, figuring out what was good, what was bad, and what was just meh. It was an interesting gig, trying to insert human expertise into an area that we’ve mainly given over to algorithms. And it highlighted just how difficult it is to create a formula for ‘good’, we generally just know it when we see it. Content can be effectively personalized for you, the For You Page on TikTok is still the best current example. But that’s an algorithm successfully gauging what will keep you watching, which is very different from ‘what’s good?’
So will AI figure that out? In most of Explainable’s output so far I’ve focused on what is getting produced with AI, not on how it’s distributed with AI. That’s mainly because the role of AI in social media distribution is black-boxed. Sometimes AI is highlighted as the shiny new thing when terms like ‘machine learning’ or ‘algorithm’ are more appropriate. Other times those same terms are used instead of AI to make a system seem less imposing.
Quick explainer
We’re just talking about what content appears in your feed here. So in that context:
Algorithm: A specific set of rules followed to distribute content, the rules can be updated continually based on machine learning input
Machine learning: A subset of AI that can refine distribution by learning from user interactions
AI: The application of human intelligence, or close-to-human intelligence, to automated distribution systems
At the moment algorithm and machine learning are generally why you see what you see in your feed. So what can more comprehensive AI do to improve that? I’m not sure it can do much. Because the one major missing element of the current systems is how to measure creativity. Every ‘decision’ made at present is based on consumption. Introducing expert human opinion into the loop, like the Consensual Assessment Technique (CAT), would be resource-heavy and very subjective.
But AI will still be used in distribution, even if it can’t improve on quality. The question ‘how do we highlight the good?’ is not generally being asked. The question ‘how do we keep people scrolling’ is being asked. AI will help with the latter, it may make the former even tougher than it currently is.
So there’s a third, narratively awful, version of Yesterday that could be made. The only guy with a memory of The Beatles back catalogue, stuck on six Spotify listeners and endlessly clicking refresh on his email alerts. A future where human curation is valued, in all its infuriatingly subjective glory, is a solution to that.
Small bits #1: Life finds a way
This 3D model of a Jurassic Park-style insect in amber, full video can be watched here, is deservedly picking up numbers on X. If you want to do more than admire it you can download the model here. It was made using VFX app Simulon.
Small bits #2: Didn’t stop to think if they should
Interesting project from Northwestern University where an AI program was capable of designing walking robots within seconds. Less Boston Dynamics and more sentient child’s bath toy. But still impressive!
Small bits #3: God destroys dinosaurs. God creates man. Man destroys God
Cool AI evolution video doing numbers on Instagram and Reddit from Sefa Kocakalay, AKA minelauvarat. But not doing numbers on YouTube because algorithm. See how I came full circle there?