Starting point

I want the evolution of machines to give people more room to shape their own lives. If we derive the human future only from what machines can do, the question of how we want to live is pushed out of the discussion. I would rather argue about what the future ought to be, and help build it.

Everything else on this site tries to make that conviction precise. It leads to narrower questions: what an organization actually means before it automates; who holds authority when agents act; how an organization can learn from what it does; when a prediction stops observing the world and starts producing it; where all of this should live while models keep changing; and what kind of civilization these systems add up to.

The argument in six steps

Each step is argued in full in an essay. Read in this order, they form one line of thought—from civilization to the everyday mechanics of meaning, authority, learning, and causality, and finally to the layer that has to hold them.

  1. Distinctiveness and common ground

    Progress is a cycle: irreproducible perspectives contribute safely to a foundation society can share, and that foundation makes room for the next difference. Contribution is not extraction—distinctiveness needs both the freedom to share and the freedom to keep. Machines should accelerate this cycle, not replace the people inside it.

    EssayThe Evolution of Machines and the Human Future

  2. Meaning as infrastructure

    Every automation runs on a hidden input: what the organization means by a customer, a completed project, an authoritative source. When that meaning stays implicit, automation turns ambiguity into infrastructure. The remedy is a small, living agreement about the meanings that carry consequence.

    EssayMeaning Before Automation

  3. Authority as a system

    Agents turn AI from an interface into delegation, and delegation moves responsibility. An approval button is not authority. Authority needs purpose, evidence, constraints, rejected alternatives, and a named owner for choices that cannot be undone—with room for the agent wherever a mistake can still be reversed.

    EssayHuman Authority in the Age of Agents

  4. Learning as connected consequence

    An organization learns only when decision, authority, execution, and observed outcome stay connected strongly enough to change the next decision. When outcomes arrive late or noisy, the honest record keeps the open question instead of manufacturing a lesson.

    EssayBuilding Companies That Learn

  5. Prediction as intervention

    Once a forecast allocates staff, credit, or scrutiny, it helps create the future it will be judged against. Accuracy after allocation can be a record of the decision loop rather than a reading of an independent future. The question is also which future became easier to occur.

    EssayDoes a Prediction That Allocates Observe the Future?

  6. Meaning that outlasts the model

    Models generate and agents act, but neither keeps what an organization needs to remain itself. The durable layer is closer to an operating system: it preserves meaning, authority, memory, and causality across models, so that replacing a model becomes an experiment instead of a reset.

    EssayNot a Model, but an Operating System

Principles

Meaning before mechanism.
Meaning is not documentation added after a system is built. It is the substrate from which a trustworthy system can be built at all.
Models before automation.
Before it is asked to act at speed, a system should be able to say what it knows, what it is allowed to do, what proof it must keep, and when it must stop for a person.
Evidence before confidence.
Forcing a lesson onto delayed or noisy evidence manufactures certainty. Record what remains unknown, and let the open question return when a similar choice does.
Human authority where consequences are real.
Being good at a task is a reason to be entrusted with it. It does not confer the authority to decide what others should live for.

Premises I do not accept

  • That human worth depends on outperforming machines. If it did, it would shrink every time machines improved.
  • That as machines become more capable, we no longer need to ask what people want. The more futures become possible, the more it matters which ones we choose.
  • That an approval button at the end of a workflow is the same as human authority.
  • That a rise in accuracy after deployment proves a model read the future.
  • That an organization's intelligence is whichever model it happens to run.
  • That common ground requires everyone to hold the same values. It should provide the conditions for living together with different ones.

Open questions

These are live. Some are the questions I am working on now; others are limits the essays name without resolving.

Current questions

  • How can meaning be represented computationally?
  • What should remain under human authority as agents gain autonomy?
  • How can an organization become a system that learns?
  • What does it take to return human distinctiveness to society's common ground without losing it?

Unresolved in the essays

  • When distinctiveness enters common ground, who decides what is safe—without turning safety into another name for the existing order?
  • Should future AI be treated as a subject with experiences and interests of its own? I do not deny the possibility; I have not resolved it.
  • How can an organization learn from outcomes that arrive years later, mix several decisions, or are too noisy to isolate a cause?
  • How can a shared layer of meaning allow local variants without fragmenting into incompatible definitions?

Systems as experiments

Ideas come first. Systems are where they meet reality, and companies are the current vehicles for building them. Each system below tests a claim; none of them is the identity of this site.

How these ideas are held

I try to start from a real question, to distinguish observation from hypothesis, argument, and principle, to make consequential assumptions visible, and to preserve the conditions under which a claim may fail.

These are working positions, not a finished doctrine. When evidence or argument changes my view, this page will be revised rather than defended.