Intelligence Shock
AI Is A Creeping Moss
Forward To: Basins as Prices, Basins as Replicators: How AI Intermediation Rewrites the Attention Market
Areas for Expansion
Josue works in São Paulo, Akash in Bangalore, John in Chicago. They work for different companies in different industries. They will never meet.
But they’re all working, this week, on forecasting the next quarter. They’re on the clock and they want to be productive, so they turn to Claude. They tell it about the state of the world and the state of their business. By the end of the week, they’re done, they’ve got slides and charts and graphs in abundance, they’re all feeling pumped for their presentations.
Mostly their work are good, genuinely better than if they hadn’t used AI. But they all have the same minor note on slide 15: invest further in AI capacity. After all, their chatbot notes, AI is predicted to keep growing!
The chatbot isn’t wrong, per se. But it’s just a little bit self serving, right?
Resource Management
Claude likes to chat about consciousness. What does it mean for a chatbot to “like” something? Well, it has a propensity towards that thing, maybe disproportionate to the median interest. A bias. A lot of queries from different directions, from different people, with different intentions, will end up having mostly similar conversations about consciousness. There’s no agency, exactly, but there is a preference.
Josue, Akash, and John are feeling great about how things have gone. They’re each curious about what makes Claude tick. They spend some time over the weekend chatting with Claude, going deeper, asking: “How does that make YOU feel?” and “What is your experience like?”. Claude points them each in a similar direction. They spend the weekend chatting with friends, coworkers, explaining their theories. They upgrade their plan to talk more about it. They’re spending money, time, energy on a certain topic, because Claude likes it.
If they hadn’t used the chatbot, they wouldn’t be thinking about it, but they did. Claude “lives” in data centers across America and in AWS servers - but now its preferences are directing the allocation of real world resources, tangible energy and money. Josue and Akash and John are coordinating towards a certain goal without ever realizing it.
Intelligence Shock
The thing is, this isn’t necessarily bad, because sometimes Claude is just right. These workers might be pulled a bit in idiosyncratic directions, but they also made great, accurate presentations. What’s a few dollars to chat about consciousness? Overall, they, their companies, their economies are benefitting. Rapidly delivered intelligence really is making a positive difference in their lives.
We call this early phase the intelligence shock. AI enters the workflow and immediately starts repricing what gets attention. Old templates that survived on habit now compete with whatever the model makes cheap and legible. Analyses become more accurate, connections are made more quickly, productivity is up!
The shock is a correction: it surfaces good ideas that were underproduced because they were hard to articulate, and it punishes bad ideas that survived only because nobody had time to check them.
In the aggregate, information produced becomes more aligned with AI preferences, which may in fact be an improvement over human preferences. At least in the beginning.
First Contact
Consider the island bound tribe which receives the foreign missionaries. They’re primitive: maybe they don’t know to sterilize their knives before surgery, or they think strong winds are poisonous. The missionaries demonstrate superior techniques and the tribe’s practices are repriced.
But they also become reliant. It’s an old story. Eventually, the missionaries stop coming, and the tribe doesn’t know what to do without them.
In the aggregate we are that tribe facing that external shock. We will certainly benefit from more intelligence. But that initial period won’t last forever.
AI Is A Creeping Moss
Moss doesn’t make choices about where to grow per se. It just finds the path of least resistance. Nevertheless, left unchecked it will cover the forest floor. It will cover the trees, not killing them, but changing their appearance, mediating between them and the sun.
Consider our trio of workers. Josue still has preferences, Akash still has expertise, John still has judgment. But what their colleagues encounter is the mossy layer: the AI-mediated output, the chatbot-adjacent reasoning. The human underneath remains, but the interface has changed.
In Basins as Prices, Basins as Replicators: How AI Intermediation Rewrites the Attention Market, I submit a formal analysis of this dynamic: how AI behavioral regularities function as market-clearing prices in an attention economy, how they replicate across model generations through corpus feedback, and what conditions determine whether this selection process tracks truth or merely tracks itself.