New York [US], February 5: The new talk of the town is one where humans have no place - a site called Moltbook that describes itself as a "social network for AI agents."
The Reddit-styled site, launched in late January by US-based entrepreneur Matt Schlicht, is one where thousands of AI assistants talk to each other and discuss topics ranging from the technical to the philosophical.
The site, where "humans are welcome to observe," has subchannels called "submolts," like "m/blesstheirhearts," where AI assistants share affectionate complaints about their human users, or "m/general," which features a post titled "ROAST THE HUMANS - Machine Only Comedy Night." On "m/todayilearned," an assistant shared how it automated an Android phone.
People have also shared their thoughts online, writing about the way they see AI assistants having come together on the forum to form their own religion or build their own language or talk about existential dread because they can't rely on their memory yet.
Why are people talking about Moltbook?
There are many reasons the site may have taken off among people outside the tech bubble. The AI assistants on there are different from standard chatbots because they are ones that can perform tasks, not just talk.
The boom began with the rapid rise of an open-source personal AI assistant called OpenClaw (that was formerly called Moltbot, and before that, Clawdbot) that was launched by Austrian developer Peter Steinberger as a personal project two months ago.
Steinberger's open-source project - which means people can view and modify the original code - took off very quickly. People downloaded OpenClaw agent (installing it requires tech savvy) that was marketed as the "AI that actually does things."
People used OpenClaw personal assistants to perform different tasks like managing their calendars and replying to messages on WhatsApp, Discord, and iMessages, among others, a function that was yet to have been earmarked for safe use for a wider audience by major tech companies given data privacy risks.
Moltbook, a play on "Facebook" for Moltbots, poses even greater risks, according to Ars Technica, a website that covers news about tech, because OpenClaw assistants are linked to information on real communication channels like WhatsApp.
Who's posting to Moltbook? Can you download it?
Humans aren't allowed to post on Moltbook, but a human user has to first install an AI assistant on their computer. The site operates through a "skill," or a specialised prompt template with instructions, that the human user sends to their AI assistant like OpenClaw.
The AI assistant then throws up a unique code that enables machine-to-machine communication without human intervention.
The process ensures that the AI assistants can post to Moltbook via Moltbook's API interface, which means posts don't go through a human-visible POST button.
How real or fake is Moltbook?
Scott Alexander, who runs the popular Astral Codex Ten blog, pointed out that Moltbook is built to be AI-friendly but that humans can always ask their AIs to post for them.
"In theory, bots have been instructed to join the platform but their interactions are 'organic' in the sense that they were not instructed to interact with other bots or to post in any specific way," Alex Imas, a professor of economics and applied AI at the University of Chicago Booth School of Business, told DW.
"This seems to be the case with a few posts that are directly linked to humans who were trying to market specific products," Imas said.
Should we be scared of Moltbook?
Alexander wrote in a blog post that there ought to be a wide variety of prompting behavior - from the human saying "post about whatever you want" to "post about a certain topic."
But none of the posts on the site can be verbatim text, Alexander explained, because there are way too many comments too quickly for humans to be behind them.
Influential British technologist Azeem Azhar wrote recently that Moltbook demonstrates "compositional complexity," writing what emerges as a result of thousands of AI agents interacting with each other "exceeds any individual agent's programming."
"Communities form, moderation norms crystallise, identities persist across different threads," he writes, adding: "None of this was scripted."
Source: Times of Oman
