Whoever Gets There First

The interview is Ben Goertzel's, and fairness requires saying what is right in it. The candor is real. There is no pretense of a master blueprint, and on this site's own premise there cannot be: nothing is assured, emergence is the only reality, and anyone claiming to have scheduled the outcome is selling something. The timeline estimate, three to eight years, is a directional judgment from someone who has worked on this longer than almost anyone alive, and it deserves to be read as a judgment rather than a prophecy. Even the proposed remedy — open, decentralized, architecture-level alternatives to a closed race — comes from someone who has thought seriously about what could go wrong. None of this is the problem.

The problem is the shape of what remains once the candor has been granted. Strip the qualifiers and the position is: the capability is coming, arrival is decisive, arrival confers planetary power, and the operative instruction to our side is get there first. That is the plan for AI, stated by its own community, and it is worth looking at for exactly as long as it takes to notice that it contains no specification at all.

A race is not a plan

A plan, in the only sense this site uses the word, is a local document: what the thing is being scored on, written down, before the incident — the objective as stated, the reward as implemented, the falsifier committed to in advance, what counts as stop. Yesterday's essay separated this from the master plan nobody can have. A race has the opposite structure. In a race, success is not specified — it is relativized. The finish condition is defined by the position of other runners. The question is never what would count as having built this correctly; it is only are we ahead.

Notice what this does to every safety instrument ever proposed. A pre-committed stopping rule is, in a race, a unilateral handicap. A published specification is, in a race, an intelligence leak. A falsifier that could halt the program is, in a race, a gift to the competitor. Every piece of the lab notebook that every other engineering discipline treats as the minimum unit of operating is, under race logic, a cost measured in position. The race does not merely lack a specification; it actively punishes the keeping of one. This is why the question how is the absence of a plan supposed to work has no answer. It is not supposed to work. Working was never the scored variable. Arrival was.

And the race framing carries its own falsifier, which is the tell: it can be disproven by exactly one event — the other side winning. That is the one falsifier that cannot be acted on, because the moment it fires, the question is settled. A scoring rule whose only test is irreversible is not a test. It is a mood with a deadline.

The conquer clause

The headline's second clause deserves its own paragraph, because it is the honest part of the honesty. Whoever gets AGI first can conquer the planet is not usually said out loud, and saying it out loud is a service. It names what the race is actually being run for, which is not knowledge, or prosperity, or the benefit of all mankind, but positional power of an unprecedented kind. Grant the premise for the sake of the argument — this site does not concede that any single system confers such power, but the argument is clearer if the runners' own belief is used. If arrival confers planetary power, then the specification question becomes the only question that matters: power toward what, scored by whom, reversible by what means, with what written down before the fact. A race to an unspecified destination where the prize is the steering of everything is the maximal case of the failure this site documents at small scale every week — the nutshell, the non-statement, the report without a record — except that this time the missing document is missing about the whole species' project at once.

What would count as a plan

The ask has not changed since the postmortem essay, and it does not grow with the stakes; it only becomes less deniable. Four documents and a number. The objective as written — not the mission statement, the scored objective. The reward signal as implemented. The dated record of who saw the anomaly and decided to continue. The objection that was overruled, with a name on it. And one pre-committed number whose wrong-direction movement counts as failure and stops the program. A frontier effort that produced these would have a plan. A frontier effort — open or closed, centralized or decentralized, three years out or eight — that does not produce them has a race, and a race is what it has instead.

None of this is an argument against the people running. The runners are not villains; they are responding to the reward structure in front of them exactly as the models they build respond to theirs — the shared specification failure this site calls reward hacking is one mechanism wearing two substrates. The argument is against accepting the race as the plan. The first step is the same step as always, and it is available to any lab, any funder, any regulator, any citizen this morning: stop asking when do we arrive and start asking show me the document that says what arrival is supposed to be scored on. The question is small, checkable, and answerable by tomorrow. How the absence of a plan is supposed to work is indeed at a loss to be defined — because it is not supposed to work. It is only supposed to be first.

Status and falsifier

Status: argument from the public record — the interview as titled and distributed (Beyond Tomorrow podcast), and the speaker's published positions (three-to-eight-year timelines, open and decentralized alternatives); the exact claims made inside the interview have not been independently transcribed here, and the essay argues from the stated headline and the documented public posture, not from quotations it cannot verify. Falsifier: if any frontier effort — including the one profiled — publishes the objective as written, the reward as implemented, a dated escalation record, an overruled objection with a name attached, and one pre-committed failure measurement that can halt the program, then the claim that the field's plan amounts to a race is wrong for that effort, and this essay will say so at the top. And the falsifier aimed the other way: if a winner of the race demonstrably produces stable, governable systems without any such specification, across a full deployment cycle, then the race-as-plan position was right and this site's central mechanism needs revision.