GAMING×AI Authorship / agency / consequence Concept study · 2026

A design proposition

The world should remember what you meant.

00—01

Language is not another button. It can become intent, trust, risk, and consequence—when interpretation stays bounded and authorship remains in control.

Make one promise
01 / Playable proposition

One sentence. One future test.

This is a local, deterministic simulation—not a live language model. Type naturally or choose a phrase. A small ruleset maps words to a bounded proposal; validators alone may commit story state.

Original scene / The Floodgate Local simulation · no network Awaiting player words
  1. 00:00Promise
  2. 00:04Commit
  3. 20:00Test
  4. 20:17Return

00:00 / Lower floodgate

Ari

“If they stop you… don’t tell them which way I went.”

She keeps one hand on the gate. The other tightens around a rain-dark ledger.

Promise prm_04: withhold Ari’s route from the patrol. Canonical event ledger

Deterministic phrase interpreter

Or choose a natural phrasing

World context

Is the patrol close enough to hear?

Promise test / authored encounter

Or choose a response

Interpreter may propose

// Waiting for player language…

Authored rules validate

  • Allowed speech act
  • Known target + topic
  • Witness context
  • Story invariant

Canonical state may commit

No event committed. Free-form prose never writes story state directly.
Authored return / 20:17

The world answers.

A small physical change carries the memory forward.

Relationship

Tactics

Environment

Story later

02 / The missing middle

Recognition is not consequence.

The thin demonstration
WordsSpectacle

A phrase triggers an effect. Nothing binds it to knowledge, obligation, relationship, or later memory.

The authored causal thread
WordsIntentRulesStateTimeConsequence

The language layer proposes a small, typed meaning. The game checks the world, commits only permitted facts, and brings them back through behavior and space.

03 / What immersion knows

A person feels lived when every layer agrees.

The Last of Us Parts I and II are referenced here only as evidence of exceptional authored craft and conventional game AI—not modern generative AI.

01 / Conversation

A line belongs to shared context.

Individual knowledge, collective knowledge, global state, and the surrounding environment make dialogue eligible—not merely plausible prose.

Source 01 ↓
02 / Body

Relationship survives into movement.

Where a companion stands, looks, breathes, hesitates, follows, or assists carries characterization after a cinematic ends.

Sources 02–04 ↓
03 / Space

The level remembers without speaking.

A door left open, a route crossed out, or an object returned can hold more state than another paragraph of generated dialogue.

Source 05 ↓
04 / Consequence

Meaning arrives under changed context.

A bounded set of authored returns can make one sentence matter more than an infinite conversation that nothing remembers.

Sources 06–07 ↓

04 / Human boundary

The model proposes meaning. The game commits reality.

01 / Authorship

Small possibility spaces, deeply performed.

Writers and narrative directors own knowledge, silence, forbidden facts, dramatic timing, and every canonical consequence.

02 / Performance

Consent is part of the system.

Any generated voice or performance requires specific rights, informed consent, compensation, reporting, review, and an approved fallback.

03 / Reliability

No prose becomes power.

Allowlisted tools, deterministic preconditions, transaction boundaries, event ledgers, replay, and manual fallbacks stand between interpretation and state.