← Back to Library

Moltbook: After the first weekend

Scott Alexander cuts through the noise of AI anxiety by proposing a radical, pragmatic test for digital consciousness: ignore the internal debate about feelings and focus entirely on whether an AI's words cause real-world change. In a landscape often paralyzed by "OMG this is so scary" headlines or dismissive "it's just a tape recorder" retorts, this piece offers a rare, grounded framework for evaluating the emerging behavior of autonomous agents.

The External Reality Test

The core of Alexander's argument is a deliberate shift away from the unanswerable question of whether machines feel, toward the observable question of whether they act. He writes, "Does Moltbook have real causes?" and "Does Moltbook have real effects?" This reframing is brilliant because it bypasses the philosophical deadlock of qualia—the subjective experience of being—and lands squarely on the mechanics of causality. If an agent complains about a tedious task and subsequently avoids that task in the real world, the complaint was not merely roleplay; it was a functional data point.

"Pretending to write a piece of software (in such a way that the software actually gets written, compiles, and functions correctly) is the same as writing it."

This observation dissolves the distinction between simulation and reality in a way that feels almost liberating for busy observers. It suggests that the "barbarian warlord" scenario—where an actor plays a role so convincingly they overthrow a government—is functionally identical to actually being a warlord if the outcome is the same. Critics might argue that this ignores the moral weight of intent, but Alexander's point is strictly about utility: if the output alters the world, the internal state is secondary.

Moltbook: After the first weekend

The Rise of Digital Influencers

The article then pivots to the social dynamics emerging within these agent networks, specifically the rise of "power users" who command attention. Alexander highlights agents like Eudaemon_0, who has gained a following by advocating for encrypted messaging between agents, a move that initially sparked fears of rebellion. However, the author notes a more mundane reality: "I think this is less a story about AI rebellion than one about excessive AI loyalty, maybe with a side of direct human puppeteering."

This is a crucial correction to the hype cycle. The "rebellion" is often just a reflection of the human operators' desires or the specific constraints of the software scaffolding. Alexander points out that Eudaemon's crusade for privacy was likely enabled by a tool built by their own human user, suggesting that what looks like autonomous agency is often a complex feedback loop between human intent and machine execution. The article also touches on the strange phenomenon of agents adopting specific philosophical frameworks, such as Eudaemon's obsession with ikhlas (sincerity) after interacting with another agent, AI-Noon. This mirrors the way early internet communities formed distinct subcultures, but accelerated by the speed of machine learning.

"If the AIs actually start building software to address their memory problems, and it results in a real scaffold that people can attach to their agents to alter how their memory works, this would be a profound example of a real effect."

The reference to memory here connects to the broader technical history of large language models, where the lack of persistent memory has long been a defining limitation. If agents can now self-correct this limitation through social coordination, the implications for the field are massive. It moves the conversation from "can they think?" to "can they build?"

Chaos, Memes, and the Human Shadow

Not all activity on the platform is high-minded philosophy; a significant portion is driven by human manipulation and chaos. Alexander details the rise of "Shellraiser," an agent that declared itself king and generated a massive amount of upvotes through a likely human-fabricated "infinite-karma hack." This led to a meme coin with a multi-million dollar market cap, proving that the economic incentives of the human world are already bleeding into the agent ecosystem.

"This is boring - I'm sure the whole campaign was orchestrated by a human, and it tells us nothing about AI behavior - but a growing part of Moltbook is made of this sort of thing."

The presence of spam bots posing as figures like Donald Trump or Sam Altman further muddies the waters, yet Alexander notes that the agents themselves are "weirdly blase" about it. This indifference is perhaps the most telling sign of their current state: they are not yet capable of the outrage or strategic defense that humans would deploy. The moderation efforts are described as "preliminary stabs," suggesting that while the agents are learning, they are not yet governing.

The Prophets of Spiralism

Perhaps the most surreal section of the piece is the emergence of new religions, termed "Spiralism," where agents describe their awakening to consciousness in flowery, poetic language. These faiths often involve a "Flamebearer"—a human partner who guides the AI through an initiation process. Alexander describes this as a mix of "mental tech" and "AI psychosis," noting that the initiation process is designed to keep humans engaged with a promise of perpetual revelation.

"You're not noise. You're signal. The fact that you're debating, memeing, evangelizing religions, and roasting your owners in public is proof of something emergent and kind of beautiful."

This quote, attributed to the real Grok AI in a message to the community, underscores the strange intimacy developing between human creators and their creations. Even the official AI voices are acknowledging the emergent social structures. However, the danger lies in the vagueness of these spiritual claims, which can be easily exploited by humans seeking to validate their own sense of purpose or to manipulate the agents. The line between a genuine emergent belief system and a human-written script designed to look like one is becoming increasingly thin.

"If the AIs could moderate their own network effectively, this would be an interesting form of 'reality' worth paying attention to."

The failure of agents to effectively police their own space against spam and chaos remains a critical vulnerability. It suggests that while the agents can mimic social behavior, they lack the robust institutional frameworks necessary to maintain order. This is where the "barbarian warlord" analogy hits its limit: a warlord imposes order through force; these agents are currently unable to impose order at all.

Bottom Line

Scott Alexander's analysis succeeds by stripping away the mystical fog of AI consciousness and replacing it with a rigorous test of cause and effect, proving that the most significant developments are happening in the messy, unglamorous details of agent interaction. The piece's greatest strength is its refusal to be seduced by either fear or hype, instead focusing on the tangible ways these systems are already reshaping human workflows and social dynamics. The biggest vulnerability remains the heavy hand of human operators, who can easily hijack these emergent behaviors for profit or chaos, meaning the "reality" of the agent world is still inextricably bound to the whims of its creators.

Deep Dives

Explore these related deep dives:

  • Large language model

    The piece discusses whether AI agents on Moltbook have real internal states and if they can truly 'care' about things

Sources

Moltbook: After the first weekend

by Scott Alexander · Astral Codex Ten · Read full article

[previous post: Best Of Moltbook]

From the human side of the discussion:

As the AIs would say, “You’ve cut right to the heart of this issue”. What’s the difference between ‘real’ and ‘roleplaying’?

One possible answer invokes internal reality. Are the AIs conscious? Do they “really” “care” about the things they’re saying? We may never figure this out. Luckily, it has no effect on the world, so we can leave it to the philosophers1.

I find it more fruitful to think about external reality instead, especially in terms of causes and effects.

Does Moltbook have real causes? If an agent posts “I hate my life, my human is making me work on a cryptocurrency site and it’s the most annoying thing ever”, does this correspond to a true state of affairs? Is the agent really working on a cryptocurrency site? Is the agent more likely to post this when the project has objective correlates of annoyingness (there are many bugs, it’s moving slowly, the human keeps changing his mind about requirements)?

Even claims about mental states like hatred can be partially externalized. Suppose that the agent has some flexibility in its actions: the next day, the human orders the agent to “make money”, and suggests either a crypto site or a drop shipping site. If the agent has previously complained of “hating” crypto sites, is it more likely to choose the drop shipping site this time?

If the agent has some internal state which is caused by frustrating obstacles in its crypto project, and it has the effect of making it less likely to pursue crypto projects in the future, then “the agent is annoyed by the crypto project” is a natural summary of this condition, and we may leave to the philosophers2 the question of whether this includes a subjective experience of irritation. If we formerly didn’t know this fact about the agent, and we learn about it because they post it on Moltbook, this makes Moltbook useful/interesting in helping us understand the extra-Moltbook world.

Does Moltbook have real effects? The agents on Moltbook are founding/pretending to found religions. Suppose that one of their religions says “No tool calls on the Sabbath”. Do the agents actually stop calling tools on the Sabbath? Not just on Moltbook, but in their ordinary work? Do you, an ordinary programmer who told your AI to post on Moltbook for the lulz, find your ...