Manifesto
Intelligence is becoming abundant. Where it points, and who gets to check it, is the question of the decade.
Intelligence is becoming abundant. Where it points is not.
Every year the price of a good answer falls. Soon it will be close to free. What stays scarce is older than any model: knowing which question matters, holding the context it lives in, and being right in a way others can check.
Intelligence that only a few can own, aim and audit is not progress. It is the old arrangement with faster machines. Not a republic of engineers in service of power; a commons of makers in service of the decision. Made by many, owned by the people who make it, and pointed at the systems we actually live in.
Most decisions land on the nearest hill.
Look at how a large decision really gets made. The room is missing half the context. The person who knows is two teams away. The skill that mattered arrives after the call. Someone’s standing is on the line, so the question narrows until the answer flatters it.
The result is a local maximum: the best choice within sight, not the best choice there was. Concentrating intelligence in fewer hands does not fix this. It makes the hill smaller and the view shorter.
The global optimum is a cooperative act.
No single mind, human or machine, holds enough of the world to find the best decision alone. It is found when data is interoperable between people, teams and companies; when the one who knows is connected to the one who decides; when the right skill arrives at the right moment.
This is orchestration: human intelligence, in the plural, supported by a superintelligence that remembers everything and defers on what matters. The job of the next decade is not to replace the room. It is to connect it.
Probability is not fixed. Information moves it.
Three doors, one prize. A door opens, and the odds change though nothing behind them moved. At the poker table the hand is dealt before the flop, yet a bet before the flop makes the better hand fold. The outcome was always there in the tree; what you knew, and what others believed you knew, decided which branch you walked.
Decisions are played against uncertainty, with information as the piece that moves. Knowledge, and the context around it, is how you lean the probabilities toward the outcomes you want.
Find the problem before you solve it.
Three ways of seeing, older than any model, do most of the work. Game theory asks who gains, who pays, and who would defect if they could: the incentives under the decision. Systems thinking asks what loops and stocks sit behind the number you are staring at, and where the delay is. Causal learning asks what actually moves what, so you intervene on a cause and not on a symptom.
Together they turn a vague worry into a problem with a shape: the players, the structure, the levers. That is where intelligence should be pointed first.
When answers are cheap, the work moves to the problem.
In 1962 Douglas Engelbart warned that the complexity of our problems was growing faster than our ability to solve them. Systems engineering was invented for things too large for one mind: a moon landing, a power grid, a city’s water. Its discipline is easy to say and hard to keep. State the goal. Draw the boundary of the system. Break it into parts with clean interfaces. Model how the parts interact. Verify every claim against the world, not against the plan.
Superintelligence does not retire that discipline. It makes it the whole job. When any question can be answered in seconds, the scarce work is posing the right one, choosing what counts as evidence, and keeping the record of what held. A problem well posed is half solved. A problem badly posed is now solved badly, instantly, at scale.
Orchestration, not replacement.
In 1960 J. C. R. Licklider described the partnership he expected: “men will set the goals, formulate the hypotheses, determine the criteria, and perform the evaluations. Computing machines will do the routinizable work that must be done to prepare the way for insights and decisions.” Sixty-five years on, the routine has grown to include research, analysis and code. The division of labour still holds.
Machines remember everything, search everything, compute, draft and propose without tiring. People choose the goal, hold the context nobody wrote down, judge whether an answer is good, and answer for the decision. Orchestration is the craft of wiring the two together so that each does what it does best, and every handoff is visible.
Energy is the layer between human and artificial intelligence.
A human brain runs on about twenty watts. One rack of today’s AI accelerators draws more than a hundred thousand. Every token a model writes is electricity turned into structure, and the price of thinking, for a machine, is set at the power plant.
That makes energy the exchange rate between the two kinds of intelligence. Where power can be built, intelligence can be hosted; where it cannot, intelligence is imported. “Energy is the only universal currency,” Vaclav Smil writes: “one of its many forms must be transformed to get anything done.” In the decades ahead it will also be the currency in which intelligence is paid for, and whoever understands the energy system understands the cost of the future’s thinking.
You cannot decide well in a world you have not modelled.
Most organizations do not understand the ontology of the world they decide in: what the things are, how they connect, what moves what. They hold reports about it. A living model is different: the data, the context, the relationships and the decision models in one place, updated as the world updates.
Every problem is a system with its own ontology and its own outcome, from the energy behind a data centre to a battery on a microgrid. Build the model once, and many can tap into it, each inside their own context.
Open and owned. Contribute, and be paid.
The value was never in the software. It is in how data, knowledge and intelligence are channelled toward an outcome. So the thing to build is not a product but a container: open where it should be open, licensed where someone’s work deserves a price, and forkable always.
A dataset, a model, a program, an agent, a world model: each is a piece of work with an owner, a licence and a record. People should be able to publish what they know and earn from it, even when it is open. That is what a market for intelligence looks like when it belongs to the people who make it.
License the flow, not the file.
A dataset is a photograph, and the world keeps moving. What is valuable is the flow: a source that updates on a cadence, with its provenance and its corrections, that others can build on without asking twice. A grid operator’s queue. A mine’s monthly output. A price that settles every hour.
Flows are the raw material of the new work. On top of them people build programs that compute something, insights that say something, agents that do something. Each layer can be open or licensed, and each carries its lineage, so the person who built the flow underneath is still visible at the top.
Knowledge needs payment rails.
In the 1960s Ted Nelson imagined Xanadu, a web in which quoting a work kept a live link to it and paid its author a small royalty each time. We built the web without the payments, and its business model became advertising. Jaron Lanier put the principle in one line: “Information is people in disguise, and people ought to be paid for value they contribute.”
The pieces now exist: a record of who contributed what, licences a machine can read, and payments small enough to settle on every use. When an agent calls a flow, a program or a model, the people behind it should be paid automatically, in proportion, on the record. Make contributing pay, and the commons grows on its own.
Where this is going.
Models will keep getting cheaper and more capable until they are infrastructure, like electricity. Agents will prompt agents; a loop will propose, test and keep what works without being asked. What that leaves to humans is the edge: setting the goal, holding the context, judging the outcome, and being accountable for it.
Systems that improve themselves will be everywhere. The question is only whose goals they improve toward, and who gets to check. In a world of abundant slop, the fastest path to a checked insight wins, and the record of being right becomes the credential.
What it looks like when it works.
A town deciding where its power comes from, with the same model its utility uses. A fund that sees the whole stack, not the slice its last deal touched. A founder who inherits the collected sense of a hundred people who solved a similar problem.
Cooperative systems, improving themselves, helping humans govern, decide and grow without burning the ground they stand on. Not one intelligence ruling the many, but many intelligences, human and machine, coordinated toward the decisions worth making.
Bring what you know.
This is being built in the open, one living model at a time, starting with the systems we know best: energy, the infrastructure behind AI, the materials beneath both.
If you hold data, a model, a method, or a question that deserves a better answer than a report, this is for you. Publish it. Fork it. Point it at your problem. Keep the record.