The Half-Life of a Mistake

First principles · Memory

The Half-Life of a Mistake

For all of recorded history, the forgetting was done for us — by paper, by distance, by the slow failure of human memory. Every second chance we ever institutionalized was built on that free labor. AI just made remembering the default, and mercy lost its subsidy.

John RectorEssay · about 6 minutes

The figure every merciful system contains t½ The half-life: how long it takes for half of a record’s weight to stop counting against you.

Credit records carry one by statute. Human memory carries one by biology. An archive read by AI carries none at all — unless somebody writes it in on purpose. That is the whole subject of this essay.

01The search you run on yourself

Tuesday · 11:47 p.m.

The final interview is Friday. Tonight you are searching your own name the way they will, and there it is on the second screen: the person you were twelve years ago, arguing badly about something you no longer believe, in a tone you no longer use.

You sit with it for a while. There is no one to appeal to. It is simply still there.

For almost the entire history of our species, that page would not have existed. Not because people were kinder, but because the world lost the file on its own. Paper yellowed. Ink faded. Witnesses died. Gossip decayed into rumor and rumor into nothing. Hermann Ebbinghaus measured the human half of this in 1885 and found that most of what we learn is gone within days — the forgetting curve, the first law of memory anyone bothered to write down. Decay was not a flaw in the system of record. Decay was the system of record, and it ran without being asked.

Every institution that ever granted a second chance was quietly built on top of that free labor. We did not merely tolerate the world’s forgetting. We trusted it, scheduled around it, and wrote it into law.

02Mercy by schedule

Look at the oldest and most durable rules humans have made, and you keep finding the same move: a clock attached to a record, and a date on which the record stops counting.

Deuteronomy commands that at the end of every seven years, debts be released — not renegotiated, released. Roman law developed praescriptio: claims that expire with time. Modern statutes of limitations say that for most offenses, the state itself loses the right to pursue you after a fixed number of years. American credit law puts a seven-year clock on most negative items and ten on a bankruptcy, after which they fall off the report by force of statute. Juvenile records seal. And the bankruptcy discharge — the Supreme Court said it plainly in 1934 — exists to give the honest but unfortunate debtor a new opportunity in life, unhampered by the pressure of preexisting debt.

Figure 01 Forgetting, by statute
The ruleWhat it lets go ofThe clock
Deuteronomy 15 Debts between neighbors, released outright. Every seventh year, on a calendar everyone could see coming.
Statutes of limitations The state’s right to prosecute, a plaintiff’s right to sue. Fixed terms of years, varying by offense; the clock is the point.
Fair Credit Reporting Act Late payments, collections, most negative items; bankruptcies. Seven years; ten for bankruptcy. Then off the report, by law.
Juvenile records What you did before you were formed. Sealed or expunged at majority in most American states.
Bankruptcy discharge The debt itself, for the honest but unfortunate debtor. At discharge — a date, stamped by a court, after which it is over.
All record, not argument: each row is settled law or scripture. Notice what they share — none of them asks whether you deserve the forgetting. The clock does not take applications. That indifference is the mercy.

Notice the design. None of these rules asks whether you have earned the release. The seventh year comes for the deserving and the undeserving alike. That is not sloppiness; it is the engineering insight. A mercy that must be applied for, argued, and judged is expensive and rare. A mercy that runs on a clock is cheap and universal. The clock is the only judge that never gets tired, never holds a grudge, and never asks for a brief.

Every second chance ever granted was co-signed by entropy. We supplied the principle. Decay supplied the enforcement.

03The quiet half of every clock

Here is the part nobody wrote down, because nobody had to. The statutes only did half the work. The other half was done by friction.

A court record from 1974 was public in name and forgotten in fact. It lived in a basement, in a county, in a filing system that answered only to someone with a car, a free afternoon, and a reason. The newspaper archive existed, but reading a decade of it was a job. Even after everything went digital, the old material sat below the waterline of any practical search — present, unindexed, effectively sealed by the cost of looking. Retrieval cost was the real statute of limitations, and it covered everything the written ones missed.

So the merciful world we inherited was never the product of the rules alone. It was the rules plus the basement. The law forgave on schedule, and friction forgave everything else.

04The default flips

Storage became free twenty years ago, and we noticed. What AI just did is different in kind: it made the archive legible. A machine that reads everything, cross-references everything, and never misfiles a thing has driven the cost of looking to zero — not for governments and investigators, who could always pay it, but for anyone, about anyone, in the time it takes to ask. The basement is gone. Every record now lives at eye level, forever.

And so, for the first time in the history of records, the default has flipped. Remembering used to be the effortful act and forgetting the automatic one. Now remembering is automatic and forgetting is the act — the thing that needs a budget, a policy, a proof, and a person responsible for it. Entropy has retired from the mercy business, and nothing steps in by default. A scoring model does not discount your past out of kindness; left alone, it uses everything it can reach, because everything it can reach is signal. The old mistake stops washing out and starts compounding.

The consequence most people miss is that the mercy in our institutions was never written where we could see it. It lived in the friction, uncodified, doing unpaid work. When the friction went, the mercy went with it — silently, with no vote taken and no law repealed. If we want it back, it has to be rebuilt on purpose, as engineering, with a date on it. The first attempts already exist: Europe’s courts ordered a search engine to unlist a man’s long-settled debts in 2014, and the GDPR wrote a right to erasure into law four years later. Those are not curiosities. They are the first bricks of the replacement — clumsy, contested, and correct about what is now required.

A hiring screen · March 2031

Your daughter applies for her first real job. The employer’s model has read everything she ever posted — it cannot help it. But the screen shows its decay function next to her name, because the law in that state requires it: nothing older than five years carries weight, and the system can prove it didn’t.

She gets the interview. The person she was at fifteen stays fifteen.

05Five predictions

  1. Records get born with death dates

    Retention policy becomes decay policy. The respectable archive of 2030 writes an expiry into a record at creation, deletes on schedule, and can prove the deletion happened. “We keep everything” stops sounding diligent and starts sounding feral.

  2. Proof of forgetting becomes a product

    Deleting is cheap; demonstrating it is not. A market grows in attestations that a system no longer carries you — audited, dated, and worth paying for precisely because copies are easy and trust is not.

  3. Scoring systems publish their half-lives

    Anyone who ranks people — lenders, insurers, employers, admissions — gets forced, by courts or by customers, to answer one question with a date: when does this stop counting against me? “Never” becomes an answer you can lose a lawsuit over.

  4. The seven-year clock generalizes into AI law

    What the credit bureaus were forced to do in 1970, systems that score people from personal history will be forced to do broadly: age the data out on a fixed schedule. The regulatory fight of the 2030s is not over access to data. It is over half-lives.

  5. Never forgetting becomes a liability

    People learn to route around counterparties whose memory has no mercy. The bank, the platform, the employer that provably lets the past expire wins the customer who has something to live down — which, given time, is every customer there is.

06Where I stand in this

I should be the last person making this argument. I keep one of the larger personal archives I know of — thousands of posts, published daily, in public, on purpose. My whole model of how the world works runs on history: actions leave artifacts, artifacts become history, and history is the only thing that ever retrains an expectation. I want the archive. I have argued that a personal AI stack must sleep — replay, consolidate, prune, rehearse — and when I wrote it I framed pruning as an engineering function, a performance matter. I now think I undersold it. Pruning is the mercy function. And I have argued that forgetting is the depreciation schedule that keeps the world payable. This essay is the institutional half of that claim: precisely because the record is my asset, I need its clock to be honest. A record that never stops counting is not history anymore. It is a sentence.

07The ledger

Already true
Negative credit items age off by statute at seven and ten years. Juvenile records seal. The GDPR’s Article 17 right to erasure is in force, and the Google Spain ruling has had search results delisted since 2014. Meanwhile plain-language search over any archive has collapsed retrieval cost to nearly nothing — the basement is already gone.
What has to happen
At least one major jurisdiction imposes decay schedules on automated scoring of personal history, the way the Fair Credit Reporting Act did for credit in 1970. An audit market emerges for proof of deletion. Scoring systems begin disclosing recency weighting the way food discloses ingredients. Watch the insurers and the employment screeners first; they are the most exposed.
Where I am probably wrong
The whole piece bets that institutions will use everything they can reach unless stopped. If stale data turns out to punish accuracy — if models score better by discounting the past on their own — then the market builds the half-life without mercy ever being invoked, the legislation I predict never needs to pass, and this essay mistook an optimization detail for a moral crisis. And if deletion proves physically impossible at scale because copies propagate faster than erasure, then engineered forgetting fails at the physics, not the policy; weighting survives, the death-date prediction does not. I would rather state those failure conditions than hedge the claim.

08The job is ours now

Go back to that Tuesday night, the page on the second screen, the person you no longer are still arguing in your name. In the old world, you could wait. Paper yellowed on its own schedule and asked nothing of anyone. The world’s forgetting was the one mercy that never required a decision.

That world is over. The file will not lose itself. From here on, every second chance that exists will exist because somebody built it — a date written on a record, a weight that falls with time, a deletion that can be proved. Decay was the oldest mercy we had.

We just took over the job.

Background

  1. Hermann Ebbinghaus, Über das Gedächtnis (Memory: A Contribution to Experimental Psychology), 1885. The forgetting curve.
  2. Deuteronomy 15:1–2. The seventh-year release of debts.
  3. Local Loan Co. v. Hunt, 292 U.S. 234 (1934). The “honest but unfortunate debtor” and the fresh start.
  4. Fair Credit Reporting Act (1970), 15 U.S.C. § 1681c. The seven- and ten-year obsolescence periods.
  5. Court of Justice of the EU, Google Spain v. AEPD and Costeja González, C-131/12 (2014).
  6. GDPR, Article 17 — the right to erasure, in force since May 2018.

Related

Author: John Rector

John Rector is a Charleston-based entrepreneur, author, and AI strategist. He co-founded E2open, the supply-chain software company acquired for $2.1 billion in 2025, and in 2026 opened Charleston AI, a 3,000-square-foot lab that helps people and organizations understand and use artificial intelligence. He is the creator of The Reality Equation — a lecture series, book, and curriculum exploring attention, prediction, and how reality is experienced — and the author of more than two dozen books. He writes and speaks widely on artificial intelligence, attention, and the future of human work.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.