Time Travel Family · Privacy Economics

The Third Kind of Time Travel

How AI compiles a past that never existed as any single record — and why the same collapse that opens your history closes everyone else’s obscurity.

I was talking about rsync. Samba flashed past the screen. My brain did the rest: Samba → Tridgell → Tridge → computer chess → Joel — and the fact that we had all once occupied a tiny Australian author scene.

I said the names out loud. Not carefully. Not as a research brief. A half-remembered meeting in Canberra, a mate in Chatswood, a private Java engine, a world championship where you had to be an author. Cool and slightly freaky in equal measure.

Five-ish seconds later the machine handed back a graph: Chompster, Bodo, Meep, KnightCap, an Australian Chess Federation newsletter that named me in print, a paper trail that puts Andrew Tridgell on a chess engine that learned its evaluation function from about 1650 to 2150 in a few hundred games.¹ The ingredients of that graph had been public for decades. The graph itself had not.

The ingredients existed. The compiled past did not.

That is not better search. Search returns documents. This compiles a world-state that never existed as any single record — and it does it inside the half-life of the thought that summoned it.

Two kinds you already have. A third you just felt.

If you have followed this series, you already own two temporal moves.

Future access — Cognitive Time Travel — is the deliverable that would have existed after weeks of work existing now, because calendar time was substituted with compute time.²

Temporal presence — Time-Shifted Proxy — is interaction with a person or state displaced by time: grandparents in the 1970s, Future You, a stand-in for someone the calendar will not let you meet.³

What happened with Scott / Joel / Tridge / Chompster / KnightCap / rsync / Samba is different. It is not reaching a future work state. It is not conversing with a proxy. It is:

Not a name-and-date lookup

The dry term for one piece of this is entity resolution — also called record linkage: deciding when records in different sources refer to the same real-world person or thing, often without a shared stable identifier.⁴ Temporal entity resolution adds the mess of identities and affiliations that change across decades; researchers treat cross-time linkage as a hard problem in its own right, not a trivial SQL join.⁵

Useful vocabulary. Incomplete story.

The actual traversal was multi-axis and mostly unspecified by the query:

Nobody asked for a seven-dimensional spatiotemporal graph join. Someone was talking about -W. The machine inferred which axes might matter, searched different corpora, resolved entities, and returned a shape while the original thought was still alive.

The query did not specify those dimensions.

Intelligence was never the documents

Imagine the same exercise in 1998. An analyst gets a half-sentence: Scott Farrell says he once met someone called Tridgell and Joel Veness through computer chess. Then the labour begins — organisations, participant lists, newsletters, spelling variants, disambiguation from every other Scott Farrell, engine names, dates, plausibility of the social geometry. Depending on archive access, that is hours or days of real work. I am not claiming any intelligence service “needed three days on Scott specifically.” We do not have that evidence, and invented agency numbers are theatre.

The economic point is sharper. Scott chatting about rsync would never have justified the intelligence expenditure. The join would not have been commissioned. The connection would have stayed dormant.

What changed is not merely that models got smarter. It is that intelligence-analysis-shaped compute can be spent on trivia at the speed of conversation. The threshold for what is worth thinking about collapsed. Cheap cognition makes questions exist that previously were not rational to attempt — a personal Version-3 cognition event sitting inside a chat about file transfer.⁶

Open-source intelligence strategy language from the U.S. Intelligence Community already treats the challenge as extracting actionable insight from vast and expanding open-source data — coordinating acquisition and developing tradecraft for that scale.⁷ Consumer AI is not “the CIA in your laptop.” The comparison that matters is economic: effort that used to require institutional allocation can now be pointed at an ephemeral human thought during the thought.

Conversational time is not a nice-to-have

Suppose the same facts had returned on Friday as a tidy report. Same sources. Same final graph. Completely different cognitive effect. By Friday the rsync thought is dead. Samba is no longer in working memory. You read a dossier and think: huh, cool.

Instead the reconstruction re-entered a still-live mental state. New cues landed on an open lattice. More private edges unlocked — “Joel lived down the road,” “we all met,” “tiny global scene” — and the graph recompiled. The machine got under the latency threshold of the thought. The result returned before the cognitive state that caused the query had decayed.

That is why “near instant” is not a UX boast. It is the difference between a report about your past and a past that can re-enter the same thought cycle that summoned it. The time machine, in this mode, is the loop.

The privacy face of the same artefact

You can frame the capability as recovery: AI can recover my forgotten world. Correct. Incomplete.

The inverse is the load-bearing half of this piece:

AI can recover everybody’s forgotten world.

Old mailing-list posts. Conference attendance. Chess results. Company newsletters. Ancient websites. Git commits. Usenet. PDFs. Staff directories. Old resumes. Archived databases. Names in meeting minutes. Each grain of sand was never particularly revealing alone. The protection was not secrecy. It was obscurity — the labour required to connect one fact to ten others.

That labour is disappearing.

“The facts were always public” no longer protects anyone the way it used to. Public is not the same as legible. Legibility is a function of join cost. Join cost fell. Privacy discourse that only models secrets will keep missing the boundary that is actually dissolving: obscurity as a security boundary.

This is not an argument against open research, journalism, or your own Life Wiki. It is a demand for intellectual honesty about symmetry. The capability you celebrate when it reconstructs you is the same capability that reconstructs them.

Compiled pasts are inference products

None of this requires pretending the reconstruction is a recovered film. False identity matches are a real danger — common names, incomplete records, aggressive inference. Provenance and uncertainty are not pedantry. They are what stops a fluent graph from becoming a confident wrong person.

Treat past-state compilation the way you would treat any other high-leverage inference system: show the exhibits, not just the verdict; keep the edges inspectable; assume error at scale. The dignity move in adjacent work is navigator-not-oracle — cue and evidence, not a pronouncement about a life.⁸ That posture is not optional once join cost collapses.

What to take with you

• Name the third kind. Future access, temporal presence, past-state compilation. Different moves. Different failure modes.
• Define intelligence operationally. Resolved entities and edges — not a pile of PDFs.
• Watch the threshold. When intelligence-shaped compute can be spent on trivia in conversational time, questions that were economically nonexistent start existing.
• Hold both faces. Personal recovery and the repeal of obscurity are one artefact, not two separate stories.
• Demand provenance. Compiled pasts are inference products. Fluency is not proof.

I still think “time travel” is the right umbrella. I would not replace it with a colder machine name. I would expand it — and I would stop pretending the privacy half is a footnote to the cool demo.