The Right to Interrupt

A system earns the right to be ignored the way a loan is retired: one uneventful payment at a time. That was the argument for trust as amortized attention. But being ignored is only half a protocol. A system that can never speak again is not trustworthy; it is mute. The other half — the harder half — is the interrupt: the moment when the machine that has gone quiet for a thousand days must reach back into a human life and say, this one is yours.

The zero-attention economy is usually described by its silence. Absorption swallows the predictable, attention is released, the dividend is paid. What that description skips is the return channel. No absorption is total. Reality keeps a remainder, and sooner or later the remainder arrives at a human being — through an alert, an escalation, an alarm at 3 a.m. The question that decides whether the whole system works is not how well the machine stays quiet. It is what the machine is allowed to say when it finally speaks, and whether anyone still believes it.

The right to be ignored and the right to interrupt are the same right, read in opposite directions. A system earns both together or holds neither.

The Reality Equation gives the interrupt its job description.

Everything a system can predict, it should absorb. What it cannot predict — the residual, the genuine surprise — is precisely the material that belongs in front of a human, because human attention is where unresolved surprise becomes judgment and action. So an interruption, properly understood, is a delivery of surprise. It is the system saying: Reality has produced something my model did not cover, and the uncovered part requires you.

This gives us a clean test for every alert, notification, ping, and escalation in our lives: how much surprise does it actually carry? The overnight page that says a payment processor is failing in a way no runbook anticipated carries a great deal. The app badge that says someone you follow has posted again carries almost none — the event was not merely predictable, it was engineered. One is a delivery of surprise. The other is prediction wearing surprise’s uniform.

Call the second kind what it is: counterfeit surprise. The engagement economy mints it industrially. Every artificial urgency, every red badge on a routine event, every “someone liked your comment” is a predicted occurrence costumed as an exception, because the costume is what gets through the door of your attention. Counterfeit surprise is to the attention economy what counterfeit currency is to a monetary one — individually trivial, collectively fatal to the unit of account.

Hospitals learned this before software did. Clinical monitors generate so many alarms — the overwhelming majority of them requiring no action — that clinicians tune them out, and the rare alarm that matters dies in the noise. The safety literature calls it alarm fatigue. The economics of attention calls it something more familiar: inflation.

When interruptions are free to issue, every subsystem issues them in its own defense. Each engineer, each vendor, each compliance layer adds one more alert, because the private cost of a missed event is high and the public cost of one more ping seems to be zero. But the cost is not zero. Every low-value interruption debases the currency in which the high-value one will someday be denominated. Print enough counterfeit surprise and the real surprise — the one the entire zero-attention economy exists to route to a human in time — buys nothing when it arrives. The page goes unanswered. The alarm gets silenced by reflex. The system fails not because it stayed quiet too long but because it spoke too often about too little.

This is why the interrupt cannot be treated as a free channel. It is a draw against the surprise reserve — the rehearsed, ready attention an organization keeps for the day an ignored system hands a problem back. Reserves that can be drawn against without limit are not reserves. They are rounding errors waiting to be discovered.

So what does a legitimate interruption owe the person it claims? If trust is the loan a system repays through uneventful operation, the interrupt is a clause in that same loan — and it can be written down. A machine that wants a durable right to interrupt a human should be able to satisfy six conditions:

1. Novel. The interruption carries surprise the system could not absorb — not a predicted event dressed as an exception, not a routine occurrence with a red badge on it.
2. Actionable. It names a decision only a human can make, and says plainly what is being asked. An interruption without a decision attached is a status report, and status reports belong in logs.
3. Timed. It arrives while action can still change the outcome. Too early is speculation; too late is a post-mortem. Both are withdrawals from the reserve with nothing purchased.
4. Priced. The issuer pays for being wrong. False alarms must cost the system standing — a demotion of its right to interrupt — exactly as a borrower’s missed payment costs creditworthiness.
5. Auditable. It leaves a record of why the model was surprised, so the next version can absorb what this version could not. An interruption that teaches nothing will be repeated forever.
6. Terminal. Once resolved, it retires. The exception becomes rule, the rule becomes absorption, and the channel falls silent again — quieter than before, because the reserve was spent on schedule and replenished with trust.

Notice what the contract implies: the right to interrupt is rationed by reputation. A system that has interrupted honestly — rarely, accurately, in time, at a named decision — accumulates the standing to be believed instantly the next time. A system that cries wolf spends that standing down to zero, and no architecture diagram can restore it. Attention, once again, behaves like capital: the interrupt is a withdrawal, and only uneventful operation makes deposits.

There is a version of this discipline that predates software by centuries. Militaries ration the right to wake a commander. Hospitals triage. Aviation spent decades deciding which cockpit warnings deserve a voice, which deserve a light, and which deserve nothing at all. Every mature high-stakes profession has converged on the same discovery: the interrupt is a sacred channel, and it stays sacred only if almost nothing is allowed through it.

The zero-attention economy will be built by teams that treat their escalation paths the way aviation treats the master caution light — and it will be unbuilt by teams that treat them the way consumer software treats the notification tray. The difference is not technical sophistication. Both use the same webhooks and the same pagers. The difference is whether anyone is accounting for the currency: whether each interruption is audited for the surprise it actually carried, whether false alarms cost their issuers anything, whether silence is treated as the asset it is.

Here is the closing thought, and it is the whole argument in one sentence: the right to interrupt is earned by not using it. A machine acquires a claim on human attention the same way a quiet colleague does — by years of speaking only when it mattered, so that the day it raises its voice, the room goes still. That stillness is the most valuable response the zero-attention economy can produce. It cannot be bought, badged, or A/B tested into existence. It can only be deserved — one uneventful day, and one honest interruption, at a time.