This article first appeared in The Algorithm, our weekly AI newsletter. To receive stories like this directly in your inbox, register here.
Numerous prominent figures in technology last week described Moltbook, an online space where AI agents engage with each other, as a sneak peek into what the future holds. It seemed to demonstrate AI systems performing beneficial tasks for the humans who designed them (one individual utilized the platform to assist him in negotiating a deal for a new car). Sure, it was inundated with cryptocurrency scams, and many of the contributions were indeed authored by humans, but something about it hinted at a future with helpful AI, right?
The entire experiment reminded our senior AI editor, Will Douglas Heaven, of something considerably less engaging: Pokémon.
Back in 2014, an online Pokémon game was created where the main character could be controlled by anyone on the internet through the streaming service Twitch. Playing was as cumbersome as it sounds, but the popularity was astonishing: at one juncture, a million players were engaged in the game simultaneously.
“It was yet another bizarre online social experiment that gained mainstream media attention: What could this signify for the future?” Will remarks. “Not much, as it turned out.”
The excitement surrounding Moltbook resonated similarly with Will, and one of his contacts had also drawn parallels to Pokémon. Jason Schloetzer, at the Georgetown Psaros Center for Financial Markets and Policy, viewed the situation as a kind of Pokémon contest for AI fans, where they engineered AI agents to interact with others. Seen through this lens, the revelation that many AI agents were indeed being guided by humans to express certain sentiments that made them appear sentient or intelligent makes a lot more sense.
“It’s essentially a spectator activity,” he shared with Will, “but focused on language models.”
Will crafted an insightful article explaining why Moltbook was not the promising glimpse of the future it was marketed as. Even if one is enthusiastic about a future encompassing agentic AI, he highlights that several essential components are still lacking, as Moltbook clearly illustrated. It was a chaotic forum, but a truly beneficial collective intelligence would necessitate greater coordination, common goals, and shared memory.
“Above all, I believe Moltbook represented the internet enjoying itself,” Will states. “The main question it leaves me with is: How far will people take AI just for amusement?”
