The Second-Cheapest Way to Find Out

Thinking has always been a cost-saving device. That is not a diminishment; it is the entire dignity of the thing. A creature that must try everything learns by dying, and most of what there is to learn kills you exactly once. So evolution built simulators — nervous systems that could run the experiment inside, where failure costs a flinch instead of a life. Karl Popper, giving the first Darwin Lecture at Cambridge in 1977, compressed the whole achievement into one sentence: “Let our conjectures, our theories, die in our stead!” Daniel Dennett later gave the design a name — Popperian creatures, minds that test the move before making it — and put us at the top of his tower for it. The reason you deliberate, the reason there is a you in the room at all, is that for all of history the real trial was expensive. A forecast is what finding out costs when you cannot afford to find out.

Every apparatus of considered judgment is downstream of that one price. The meeting, the memo, the committee, the pro-con list on the back of an envelope, the five-year plan — all of them are simulators, built and staffed because sampling the world was slow, costly, or irreversible. We do not argue about which headline will work because arguing is pleasant. We argue because, for nearly all of human history, finding out meant spending something we could not get back. Thought was the second-cheapest way to find out. The cheapest way — just try it — was almost never affordable. We built civilization on the runner-up.

Now watch the price move. A capable AI stack does not make you smarter about which of two drafts will land. It makes the question unnecessary, because running both has become cheaper than discussing either. Wherever an action is cheap to take and cheap to retire — a draft, a price, a route, a schedule, a block of code, an order of operations — the honest answer to what do you think will happen? is becoming why are you asking me? Ask the world. It answers within the hour, and it has never once been wrong about itself.

Be precise about what loses its job here, because it is not the denominator. The Reality Equation puts expectation underneath everything — R = A/E — and E is unconscious, inherited, already in the room before you are. No meeting ever set it, and no cheap trial will unseat it. What the cheap trial replaces is the conscious ceremony of prediction — the part of deliberation that was always a committee doing a slow impression of the machinery underneath. And notice the quiet gift in the exchange: a trial is an action, an action leaves artifacts, artifacts become the record, and only the record retrains the forecaster. The person who runs ten cheap trials a day is not bypassing their expectation. They are feeding it — down the only road that has ever reached it, at a speed no amount of deliberating ever managed.

For all of history, the cheapest way to find out was unaffordable. We built everything on the runner-up.

The consequence most people miss is where judgment goes when prediction stops paying. It does not evaporate. It concentrates. The price of a trial did not fall everywhere — it fell exactly where actions can be rerun. But nobody lives the average: you live one path, in order, once, and along that path sit a small number of moves that cannot be taken twice. The hire. The resignation. The word said in the kitchen. The trust spent. The name on the building. The year given to the wrong thing. No fleet of agents can run next year both ways and show you the readout. Where the trial cannot enter, the forecast is still the only instrument there is — and so the territory of human judgment is not shrinking so much as hardening. The rerunnable world is leaving its jurisdiction. The irreversible world is becoming all of it.

Figure 01

What leaves the desk, and what remains

• Rerunnable · ask the world The trial takes these
  • the headline
  • the price point
  • the draft
  • the route
  • the send time
  • the layout
  • the code path
  • the order of steps
  Cheap to take, cheap to retire. Debating these is now slower than answering them.
• One-shot · yours Judgment keeps these
  • the hire
  • the resignation
  • the word said
  • the trust spent
  • the partner taken
  • the name chosen
  • the year given
  No rerun exists. The forecast — conscious and inherited both — is the only instrument that reaches them.

What I think happens next

1. The forecast meeting dies wherever the trial is cheap.

   Debating a testable question in a conference room starts to read the way long division performed beside a spreadsheet reads: not wrong, just a performance of effort the tools have outgrown. Meetings split into two honest kinds — readouts of what was run, and judgment on what cannot be — and the second kind gets smaller, slower, and far more senior.

2. “Can it be rerun?” becomes the first question asked of any decision.

   Before who owns it, before what it costs — is a failed attempt cheap to retire? Organizations will keep explicit registers of their one-shot decisions the way they now keep risk registers, and everything off that register will be delegated to the stack by default. Reversibility becomes the working border of the job description.

3. The prestige of being right in advance splits in two.

   In rerunnable domains, prediction becomes a parlor trick — the readout arrives before the argument finishes, and the confident forecaster is remembered as the person who delayed it. In one-shot domains the opposite happens: the person who judges well where no trial can go becomes the scarcest input in the economy, and track records on irreversible calls become the credential that matters.

4. Deliberation survives as ceremony — and starts being named honestly.

   Many meetings were never simulators; they were rituals for distributing ownership of an outcome before anyone would commit to it. Those survive, because a readout assigns accountability to no one. What changes is the cover story: once the trial is cheaper than the talk, a meeting about a testable question has to admit what it is actually for.

5. The people who ask the world pull away from the people who ask the room.

   Every cheap trial is an action, and actions compound into a record no amount of opinion can match. A person who routes a hundred small questions a month through real trials is manufacturing history at a rate a deliberator never will — and only history retrains the forecaster they carry. The gap between the two kinds of person stops being a difference of style and becomes a difference of calibration, and it widens every month.

The desk I kept

I have written that my two favorite prompts are four words long — you do it, and you decide. Some months in, I can report what those four words actually did to my desk. They did not empty it. They sorted it. I used to open a document to decide between two versions of a thing; now I say run it both ways and read what came back. What remains in front of me has stopped being a queue and become a short list of things that can only happen once — fewer decisions than I have ever carried, and heavier. The last line is still me. But the line is not made of tasks anymore. It is made of moments that will not come again.

Already true

Popper’s sentence is record — Darwin College, Cambridge, November 8, 1977. A generation of product teams already replaced which-will-work arguments with live trials wherever traffic made trials cheap; that is what testing culture is. Agentic stacks are now extending run-both-ways past web pages into drafts, prices, schedules, and code — and a scheduled stack runs trials while its owner sleeps.

What has to happen

The stack has to become trusted to run trials unsupervised, because a trial that waits for permission inherits the cost of the meeting it was supposed to replace. The price of real trials has to keep falling outside pure software. And the readout has to stay honest — a trial quietly rerun until it flatters its author is not a trial; it is the old ceremony wearing a lab coat.

Where I am probably wrong

Reversibility may be a fiction of small stakes. Every trial leaves artifacts in a record that does not forget, and an audience run through a hundred cheap experiments learns it is being experimented on — the aggregate of rerunnable trials may itself be a one-shot decision about trust. And deliberation may hold a second job economics cannot see: a meeting distributes ownership of an outcome, and a readout assigns it to no one. If accountability needs somewhere to pool, the conference room survives on that alone — and my first prediction dies of it.

What can only be lived

The meeting about the headline will sit on calendars for a few more years, out of habit, the way hand arithmetic sat in ledgers after the machines arrived. When it ends early — and it will end early, because the readout got there first — watch where the room’s attention goes. Not to the winning headline. To the two items at the bottom of someone’s list that no slice of any audience can answer. For four billion years, minds deliberated because they could not afford to find out, and theories died so that we did not have to. Now almost everything can be tried, and what cannot be tried stands out with terrible clarity: the part that can only be lived, once, in order, by you.

That part was always yours.