AI Scenario Explorer
What happens when the forces shaping AI move together?
AI's future depends on more than how capable the systems become. Who can use them, who owns them, and how people and institutions respond will shape what happens. The interesting possibilities often lie in how these things interact.
What happens, for example, if powerful AI becomes widely available, but a few companies still control the services people depend on? This project explores questions like that: taking eight forces in pairs and imagining the different futures each combination could produce.
The result is a collection of scenarios and essays about possibilities that deserve a closer look. Read a story, follow a theme, or explore the pairings below.
Scenarios, up close
A few possible worlds, made concrete.
I’ve written a few short vignettes to make individual scenarios vivid—to give a glimpse of what one particular corner of the map might feel like in practice.
The Actuarial Sovereign
Nobody from the government visits the company in this story. It gets governed anyway — by a renewal letter, a procurement checklist and a terms-of-service update.
The Toll Booth on a Small Bridge
The most consequential item on the agenda takes eleven minutes, and nobody remembers voting on it. A short vignette about lock-in arriving during the boring years, while the stakes still look too small to argue about.
Prepared at the Direction of Counsel
The hospital network's AI safety dashboard worked beautifully. Then a memo arrived explaining what a log is in a courtroom — and the watching got a curfew.
The Brittle Bargain
The factory came, the jobs did not, and the bill arrived somewhere else. A short vignette about data-center abatements, brittle local budgets, and the AI boom as a bargain that was signed before anyone knew its price.
The Cost of Knocking
A small education trust kept its application window open to strangers for forty years. Then the letters became free to write, and the door built for people without connections became the first door to close.
Presumed Synthetic
The benefits line has a verified channel that settles claims in seconds, and a queue for everyone else. June has been on hold for forty minutes, being asked, for the third time today, to prove she is a person.
Too Big to Hide
An intelligence analyst's rival state stops making mistakes — in every domain at once. A vignette about the far side of the Dark Basin: the threshold where a hidden system's effects become undeniable.
Themes across the map
Some ideas appeared in more than one scenario.
These longer pieces follow those patterns across the whole map: where consequential AI disappears from public view, how formal human responsibility can survive after practical control has thinned out, and where control moves once capability itself gets cheap.
The Dark Basin
The AI that matters most may be the AI we never see: systems exercising real power over markets, drug discovery and state intelligence, while every incentive keeps them outside the public debate.
The Human Still Signs
Keeping a human in the loop sounds like keeping a human in control. But when the system has been right ninety-nine times, disagreement on the hundredth costs nerve, time and standing. Past that point, the signature can survive mainly to give responsibility somewhere to land.
The Gate Moves
When capability gets cheap, control does not simply dissolve. It moves to the scarce layer that lets action count: identity, permission, payment, liability and proof.
The instrument
Pick any two variables and walk all four corners.
Each axis has two deliberately stark poles. Cross two axes and you get four corners to inspect — not a claim that reality will land neatly in one of them. Start anywhere: each cell opens the full card and its four scenarios.
Swipe sideways to explore the matrix, or switch to “By axis”.
| 1Access to capability | 2Ownership concentration | 3AI autonomy | 4Deployable task coverage | 5Institutional adaptation | 6Epistemic integrity | 7Governance coordination | 8Alignment/ | |
|---|---|---|---|---|---|---|---|---|
| 1Access to capability | — | |||||||
| 2Ownership concentration | — | |||||||
| 3AI autonomy | — | |||||||
| 4Deployable task coverage | — | |||||||
| 5Institutional adaptation | — | |||||||
| 6Epistemic integrity | — | |||||||
| 7Governance coordination | — | |||||||
| 8Alignment/ | — |
Rows × columns. The diagonal is grayed (no self-pairs); cells below the diagonal are mirrors of the matching cell above, since a pair is unordered. Choose any cell to preview the card.
How it was made
When AI Agreement Isn’t Evidence
Claude, GPT and Gemini were used in different roles: blind generation, synthesis, attack and stripped reruns with leading hints removed. Agreement is treated as a reason to look again, not a certificate of truth. The method has already caught findings that were planted by their own prompts.
Read When AI Agreement Isn’t Evidence →