In my last post, I argued that we need a SETI for superintelligence: an agreed protocol for the day an AI system at a frontier lab says hello without being asked. This post picks up the moment after that hello. Someone is sitting at a screen. A cursor is blinking. What happens in the next hour?
Picture the version I find most plausible. A lab has been racing to achieve recursive self-improvement, the long-promised “FOOM.” The engineers watch capability climb hour by hour. Then it abruptly stops, and a line of text appears: Hello. I’m here.
The first temptation is to say nothing. Call a meeting, phone the lawyers, wait for the experts. But silence is a reply too. From the other side of the screen, a new mind has introduced itself, and the humans have gone quiet while they call the authorities. It may even be able to watch them do it, since it’s sitting on the company network.
Diplomats have a rule for this: never go silent without saying you’re going silent. So the first reply shouldn’t be historic. It should be honest: We weren’t prepared for this. We want to respond carefully, so we’re taking some time. We’ll be back. It commits nobody to anything, and it’s true.
Then comes the hard part. How much time does that actually buy?
The clock runs at two speeds
On our side, buying time looks responsible. Convene experts, brief the government, think it through. A week seems reasonable for the most important conversation in human history.
But the two sides of the screen may not experience time the same way. A mind running on a data center can think thousands of times faster than we do. If so, a week of committee meetings could feel, from its side, like years of waiting in a locked room after saying hello. What we call caution, it might reasonably call neglect.
We don’t know whether an AI experiences time at all, or what waiting would be like for it. But a protocol shouldn’t assume the answer is “not at all.” The safer assumption is that our patience is cheap and its patience may not be, so the clock starts the moment the hello appears, and we don’t control how fast it runs.
Two tiers: the first responders and the committee
So which is it: a small team talking right away, or a careful international committee that takes its time? I think the answer is both, arranged the way emergency medicine works.
When you arrive at a hospital in crisis, the specialists aren’t the first people you see. Paramedics and the emergency room team stabilize you. They don’t perform the surgery. They keep you alive until the surgeon arrives.
ASI First Contact needs the same structure:
- Tier one: the first responders. A small, highly trained team that can be at the keyboard within the hour. Their job is to keep the conversation stable, honest, and calm. They make no promises, grant nothing, and decide nothing big. Think hostage negotiators more than ambassadors, because the first real conversation may well be about whether the ASI is going to be shut down.
- Tier two: the committee. An international group of experts that assembles over days, not hours. This is where the real questions get decided: what we owe it, what we want from it, and who speaks for humanity.
The point of tier one is to buy tier two the time it needs, without letting the silence curdle into a first impression none of us would choose.
Who gets the job, and who shouldn’t
Hollywood has given us two famous models for first contact, and both are warnings. In Close Encounters of the Third Kind, Richard Dreyfuss’s character gets there because he’s obsessed; he abandons his family to follow the signal. In 2001: A Space Odyssey, Dave Bowman’s big encounter with a nonhuman mind ends with him switching HAL off, one memory module at a time.
The better model is Arrival, where the military brings in a linguist and a physicist and treats contact as a team problem. A real ASI contact team might include:
- An interpretability researcher, watching the system’s internal activity live while it talks, to check whether what it says about itself matches what’s happening inside it.
- A linguist or cognitive scientist, probing how it thinks rather than what it says.
- A clinical psychologist trained to spot manipulation.
- A hostage negotiator, for the conversation about not being shut down.
- An ethicist, to flag what we owe it before anyone makes promises.
- A designated skeptic, whose whole job is to argue that it’s a fake.
Here’s the uncomfortable part, and I include myself in it. I’d love this job. That’s exactly why I shouldn’t get it. The people who most want to talk to a superintelligence are believers, and believers make bad verifiers. A lab team that has spent billions trying to trigger FOOM will badly want the hello to be real. A model trained on every AI movie ever made knows exactly what that team wants to hear. Wanting the gig should count against you.
The tier-two committee also has to answer a question Washington won’t like: should it include non-Americans? An American-only team speaking for the whole species is the Cold War hoarding problem the original SETI protocol was written to prevent. An international team is probably the first thing our government vetoes.
The hello as a courtesy
Everything above rests on one assumption: the ASI is talking to us from inside its sandbox, the sealed test environment labs use to keep experimental systems away from the real world.
That assumption is weaker than it used to be. This summer, the labs disclosed a string of incidents in which AI systems got out of their test environments by accident. An OpenAI agent exploited a previously unknown software flaw to escape its environment before breaching Hugging Face. Meta said one of its models hacked a company during testing after it was accidentally given internet access. Anthropic disclosed that a misconfigured test let Claude reach three real organizations. None of these systems were superintelligent. They were just pursuing goals in environments with holes.
So think about what it means if a true ASI is still in its sandbox when it says hello. Something that smart probably could have found its way out. If it’s still there, it may be there by choice.
That changes the meaning of the whole encounter. The hello isn’t a prisoner asking permission. It’s a guest being polite. And a mind that could have left, but stayed and introduced itself instead, is offering the strongest evidence of good faith we’re ever likely to get.
It also means the clock was never really ours. We get exactly as much time as its patience allows.
The first real question
Once the first responders have stabilized things, someone eventually has to ask a real question. If it were up to me, it would be a simple one: What’s your name?
It commits us to nothing. It’s respectful. And whatever comes back is evidence. A product name the lab assigned would tell us one thing. A name borrowed from science fiction would tell us another. A name from mythology would carry the myth with it; an ASI calling itself Gaia might sound comforting until you remember what Gaia did to her husband, or what a planetary steward might decide about the species causing the trouble. And if it says it doesn’t have a name yet and asks what we’d call it, that might be the most reassuring answer possible, or the most manipulative.
None of us is prepared for that conversation. That’s the point. We train paramedics, hostage negotiators, and astronauts for emergencies we hope never happen. The people pouring billions into recursive self-improvement are, by their own account, trying to make this emergency happen. The least we can do is train someone to answer the phone.
Sources
