The Frontier Is Not Yet Built
A response to Dario Amodei’s “We Must Pace the Frontier”
I remember the first time I heard Kirk Franklin’s “Hello Fear.” I had just watched tornadoes tear across the country, taking families and whole communities at a scale I had not seen before. This week I pressed play again.
The fear that has been rumbling under our feet for months stopped rumbling and started to speak. A job apocalypse. Rogue swarms. Our children’s minds going soft. And now, at the end of the week, the possibility of our own ending at the hands of an intelligence that is not ours. Every speech, every thread, every article carried the same note. A researcher walked out of one of the labs and said the people building this believe it could kill us all. A senator from the left and a strategist from the right stood on the same stage to say the decisions are being made by a handful of men who answer to no one. A former presidential candidate said a lab head told him the internet itself may be too polluted now to train on.
And Dario Amodei, who runs the lab the researcher resigned from, wrote that we must pace the frontier.
I read his essay twice. It deserves that. He opens with his father, who died of a disease cured a few years later, and with his own early cancer — treatable now, untreatable fifty years ago. He takes the OpenAI–Hugging Face swarm seriously and asks every lab to act as if it had happened to them. He commits his own company to something unusual: outside evaluators with badges and desks and the right to publish what they find without his editorial control. And he answers the question a slowdown was always missing — what would you do with the time.
I want to answer him on those terms. Because I agree with the danger. And because I think the essay stops one step short of the thing that matters most.
The frontier
We are arguing about how fast intelligence should advance while leaving the civilization it is supposed to make possible unbuilt.
Here is what I mean by the frontier. Not the models. Not the benchmark that fell this week. The frontier is the physical civilization intelligence could help us construct: an industrial base beyond Earth that does not need Earth to manufacture every increment of its own growth, energy that reaches every person as a floor rather than a promise, and the institutions that make what gets built an inheritance rather than a rental. That frontier is not yet built. It is barely begun. And nearly every hour of the pacing debate is spent on the fire, not on what the fire is for.
I don’t say this to dismiss the debate. I say it because I have a stake in the time. Dario wrote about his father. I will write about my parents. I believe we will solve dementia — I believe it with everything I have — but I no longer get to say when, and it will not be in time for them. I believe we will end energy poverty, but not in time for the more than two hundred thousand children under five who die each year from the smoke of cooking fires in homes with no clean energy, by the World Health Organization’s count. I believe we will build a world where a family’s today is not held hostage by its tomorrow. Not in time for the ones who already left.
That is what this technology could give us. Not intelligence. Time. AI was never the gift. It is a superpower, and superpowers save no one; the people who carry them into the world do. The gift is time, our imagination and creativity entangled with AI and not subjected to it, and the undeveloped frontier that waits — and whether what we build arrives before the next family needs it.
Not with you. For you.
Now the fear, named precisely.
If intelligence is what carried us from caves to the grid, and we have lit a fire that could exceed our intelligence by orders of magnitude, then a quiet thing happens — slowly, then very fast. Something else builds your universe for you. Not with you. For you. And if you cannot trust the ones who hold the fire, or the fire itself, the fear compounds until it is the only thing anyone can hear.
That is the real content of the week. Not that the machines are evil. That we might stop being the authors.
So here is the principle I would add to every pacing framework, and it has nothing to do with speed: greater intelligence must never automatically confer greater authority. A system can become better at everything and still not earn the right to expand its own mandate, redefine whom it serves, release a successor, or rewrite its purpose. In my work that is the Two-Speed Constitution. Capability may improve at generation speed. The standards by which improvement is judged change only through a slower, human-ratified process. It is the difference between a collaborator and a replacement, and it is the answer to “not with you, for you.” With you means you keep the pen.
Atoms don’t lie
I have spent more than two decades on one thing: to build the universe. To turn the matter and energy that is already out there into the greatest resource humanity has ever known. Not to photograph it. To build it. To turn it into a laboratory, and into a thermodynamic income statement — energy in, entropy out, civilization retained on the balance sheet.
That ambition taught me the discipline the AI debate needs most. AI is not the king. Physics is king. Atoms don’t lie — though instruments can be wrong, which is why someone must always be able to challenge the measurement. A bracket bears its load or it does not. A process meets its energy budget or it does not. A persuasive answer settles nothing on a test stand. The fire is loud. The ledger is quiet. Only the ledger tells you whether you built anything.
Andrew Yang said this week that a lab head told him the answer to a polluted internet was to move to “the world of atoms.” The account of self-replicating code seeded across the internet is a reported allegation from an unnamed source, and I will treat it as exactly that. But the phrase is right, and it points at the question I have been building toward for years: what happens when intelligence can investigate and construct the physical world?
Get Antonio’s stories in your inbox
Join Medium for free to get updates from this writer.
The honest answer is that the world of atoms does not make a system safe. It makes its claims answerable. Distance makes intervention harder, not easier. An autonomous system off Earth needs stronger evidence than a chatbot, not an exemption from scrutiny. What the world of atoms gives you is a place where every claim has to survive contact with matter before anyone is asked to believe it.
The synths
When I first imagined the Synth Civil Service Infrastructure — the machines I want to build — I never thought they needed access to the internet. They still don’t. I never imagined them walking my dog or doing the dishes. Others can build that machine, and for a caregiver it would return exactly the gift I am describing. I imagined ours doing what we couldn’t, because we couldn’t survive the fire first. Send them out into the universe and turn it into a laboratory. I would rather our synths land on Mercury and turn it into compounding wealth than take out the trash.
Here is the one mechanism under that sentence, so that “geometric abundance” is a claim and not a mood. We are pursuing industrial systems that can build more industrial capacity from local matter and energy — extraction to refining to fabrication to assembly to repair, with the awkward components on the bill of materials and not only the impressive ones. If a productive system can generate more than one viable successor over its operating life, after failures and replacement are counted, capacity compounds. If the mass it needs from Earth per successor falls across generations, that compounding stops depending on launches. Those two conditions have to coincide in a measurement, and the measurement has to survive repetition. Nobody has demonstrated that once — not at scale, not off-world, not anywhere. Until we do, it is a hypothesis about compounding. That makes it a claim you can lose, which is the only kind worth writing down.
The universe beyond Earth is also the one frontier in history where abundance is the default condition. On the crudest arithmetic, the asteroid belt holds enough mass for tens of thousands of industrial forges, each with more floor area than a third of everything humanity has ever built. Scarcity out there would have to be manufactured. That is worth saying in a week when two governments prepare to sit across a table from each other over AI. The race is being run as if it were zero-sum. The frontier is not.
And to the labs — what would you do with the extra time? Here is one answer: point some of it at the world of atoms. The discipline you are asking for at the frontier of models — evaluators with real access, evidence others can inspect — is the discipline an industrial lineage needs before it is allowed to reproduce. In my work it is the Return Coupon: every discovery comes home with its measurements, methods, provenance, uncertainty, and predictions, so that a human team can audit it in bounded time. A signature identifies a source. It does not make a claim true. And the last obligation, the one no builder wants to write: remain able to die. A lineage that cannot be stopped and audited between generations cannot have its inheritance certified. Death is not the failure of the design. It is the boundary condition that makes the design possible.
Supplicants
Bernie Sanders and Steve Bannon stood on the same stage this week, at an event convened by the physicist Max Tegmark. They agree on almost nothing else. Sanders intends to introduce legislation with Representative Greg Casar to permanently ban artificial superintelligence, pause advanced AI development until safety rules exist, and require that half the board of the largest AI companies be publicly appointed. Bannon said the American citizen will no longer be a supplicant to the oligarchs, that the public is being asked to carry the risk while the upside stays private, and that this is a hinge in history.
I don’t share either program. A ban settles who does not build. It does not settle who inherits. But each of them is right about something a builder has to answer, and I would rather answer it than wave it away.
Sanders is right that the public has had almost no say in a technology that will touch every life. Bannon is right that socialized risk and privatized gain is not a deal anyone should accept — and he is right about the word. It is a hinge. But a hinge without a crossing condition is a mood, and the crossing here is not a vote or a ban. It is a structural claim on the abundance. In my work that is abundance as the floor: a Universal Energy Dividend of one hundred kilowatt-hours per person per day, as the design target for what a self-reproducing industrial base owes the people who never owned a share in it. It is a target whose generation, delivery, and reliability still have to be demonstrated, with a ledger that distinguishes energy promised from energy actually available to a person, in a place, when they need it. And it has a companion for knowledge: every certified discovery a synth lineage makes returns to the commons with its evidence and its failures attached, so that when machines inherit more of the knowing, people inherit more of the understanding.
None of that excuses the present. A worker who loses an income cannot live on a prediction about stellar industry. A future dividend cannot excuse present dispossession. The transition needs protection now, public participation in consequential decisions, and durable claims on the gains — or the frontier will be built the way the last one was, with the dispossession moved off-world.
What survives the fire
I am not saying the fears are unfounded. I am saying they are not certain, and that fear is not a plan. I think of the scene in I, Robot: we don’t shiver. We are builders. We are the architects of our own universe. We sit with the fear and build like our lives depend on it, and they may.
What we must not do is create the thing that takes our future from us and then hope it loves us, or likes us, or at least does not disregard us. We cannot fight a ghost a hundred times smarter than we are if we hand it the means. AI is not an invasive species; we made it. It will not destroy us unless we surrender the one superpower that was always ours, which is choice. The danger and the audacious opportunity are both true. The choice is who builds our universe.
I call AI my co-creator in this work. The word describes the process, not a claim about consciousness. Its value is not agreement. It is helping produce a question precise enough that reality can answer no.
So here is what I would set beside Dario’s three steps. Pace the models. Build the frontier. Pacing should buy identifiable improvements in safety. Construction should produce identifiable improvements in human capability. Both should produce evidence other people can inspect, because the public deserves more than the assurance that someone powerful has the situation under control. Embedded evaluators for the labs, yes. Independent auditors for any lineage that reproduces, before it reproduces. A return path for every discovery. A floor under every person. A constitution that moves slower than the machine. Compress time. Never compress truth.
My answer to this moment begins on a bench. Our first prototype kernel sits there today — a test of whether one industrial cell can pass working capacity to another, and whether we can tell the truth about what it inherited. Its first proof may fit on a workbench. Its horizon may reach across the stars. Between them is the work, and the responsibility to make its promise answerable to every life it is meant to serve.
Our world is changing on every axis, all at once. The near-limitless capacity of the universe is a gift humanity must inherit. Let that wash over you for a moment. I pressed stop on the song.
We must not maximize the fire. We must maximize what survives the fire.
