Public preview 02 · Source-mapped · Independent human review pending Concept dossier C-004

Published recordAs of

Atlas / Concept 004 / Ambient intelligence

Home & relationships
Interfaces & augmentation

The intelligent room

Selected fiction imagined a mind in the walls. The present is a room full of partial senses. This exhibition opens the walls.

Framing questionWhen does a helpful environment become an observing one?

House route / Stop 01

City edge · one household · browser exhibition

A normal house, one layer at a time.

This is not a tour of imaginary gadgets. It begins with an ordinary house, then follows what its systems can sense, infer, remember, control, and explain. The last stop separates equipment that exists now from the combinations that remain a five-year scenario.

House study H01 / Exterior orientationIllustrative CGI · capability boundaries appear in the tour

01 / Room Observatory

Interactive architectural study · renderer 01

The room is not a single intelligence.

The intelligence is scattered across sensors, models, memories, house rules, devices—and the people who live with them. Choose a verb to see what has to happen, and what the room should reveal, before we call it intelligent.

Room awake · composite view

Cutaway diagram of an intelligent room A project-created, materially AI-assisted isometric room diagram with switchable layers for sensing, inference, memory, decisions, action, explanation, and forgetting. PHYSICAL STOP OCCUPANT 01 ACTIVE APERTURES / 03 01 / SENSE CAMERA MIC CLAIM PLANE / INTERPRETATION 02 / INFER LOCAL MODEL RETENTION / SCOPE + TIME 03 / REMEMBER RETENTION LEDGER ASK ACT WAIT 04 / DECIDE POLICY + CONFIDENCE ACTION PLANE / REVERSIBLE EDGES 05 / ACT LIGHT DOOR BLIND RECEIPT / TRIGGER · RULE · EFFECT 06 / EXPLAIN ACTION RECEIPT NAMED STORE ENDS / OTHER EFFECTS UNRESOLVED 07 / FORGET DELETION RECEIPT
The Room, Opened · project-created, materially AI-assisted editorial diagram · illustrative, not a product schematic Meter-based scene · eight evidence-linked anchors · headset renderer not yet released

Stress-test the promise

Our starting room assumes one person, one vendor, and a working internet connection. Real homes rarely stay that simple. Change the conditions.

  1. 01SenseObserved capability

    The room picks up sound, motion, light, temperature, and device state.

  2. 02InferObserved capability · current limits

    Those signals become guesses: occupied, familiar, likely, unusual. Identity is a conclusion, not something a sensor simply sees.

  3. 03RememberObserved capability · current limits

    Events, preferences, and profiles can linger—often without a clear view of what was kept or how to change it.

  4. 04DecideProposed safeguard

    A person, rule, or model chooses: act, ask, wait, or do nothing.

  5. 05ActObserved capability

    A permitted command changes what people see, hear, feel, or can access.

  6. 06ExplainProposed safeguard

    People should be able to see what happened, why it happened, and who or what had permission.

  7. 07ForgetProposed safeguard

    Deletion needs a clear target and limit. A receipt can confirm an action, but not that every copy—or every learned effect—is gone.

We use seven verbs to read the room. They are not a literal flowchart: real systems loop, revise, and let one action become the next signal.

H01 / Full-scale scenario / City edge—2031

When the room begins to notice.

The house came first. The Roadster, one household robot, a spatial room, sensors, rules, memories, controls, and exits arrive later. Seven verbs expose the handoffs that make separate machines feel like one intelligence.

CGI street view of a full-scale city-edge house with rooftop solar, a red sports-car study beside charging equipment in an open garage, separate stationary storage, a headset dock upstairs, and a mower on the right-hand lawn.

The illustration loads only when you enter.

H02 / Lived-in fieldbook

Six scenes from one imagined household · one ordinary night

One house,
one ordinary night.

The full-scale view places the technology in an actual neighborhood. These close views show where it touches ordinary life: at the front door, across a shared kitchen, at the table, and inside the household record.

  1. Architectural evidence plate showing one standard 36-inch door leaf in plan beside a contact sensor; it separates the observed open state from identity and intent the threshold cannot know.
    Plate H02.1 / 18:14:07Project-created systems plate · measured illustration

    Threshold / Sense → Infer

    The door opens. That is all the threshold knows.

    A door contact reports one change; motion may suggest that someone is there.

    Calling it an arrival—and deciding who arrived—means combining separate signals into a story.

  2. CGI kitchen with two different chairs, an open book and warm task lamp at one place, a closed laptop and water glass at the other, and a shared wall dimmer.
    Plate H02.2 / 18:14:39CGI study · shared preferences

    Shared kitchen / Remember → Decide

    The routine remembers one person. The kitchen already belongs to another.

    One resident likes a bright hall. Someone already at the table wants the kitchen dim.

    Both facts can be true. The system still has to decide whose comfort wins—or stop and ask.

  3. CGI parts display with a wall light, radiator valve, speaker, and deadbolt, each connected by a yellow inlay to its own physical button.
    Plate H02.3 / 18:14:48CGI study · four separate actions

    Across the house / Act

    One command becomes four different changes.

    Light, heat, media, and a lock each receive their own instruction.

    Calling that “the room” hides the important part: every device can fail on its own, needs its own override, and leaves someone to deal with the result.

  4. CGI proposed explanation control with a rotary dial, stop button, blank paper receipt, and removable guest token on a side table beside a mug.
    Plate H02.4 / After the changeCGI proposed safeguard · explanation and override

    At the table / Explain

    If the room acts, it should be able to say why.

    The explanation should survive the app.

    A useful receipt would say what set the change off, which rule fired, what changed, and who authorized it. You should not have to hunt through someone else’s account menu to understand your own room.

  5. CGI empty hallway at night with a low path light, wall sensor, and a local controller with a removable memory module on the side table.
    Plate H02.5 / 02:14CGI study · a signal and its retained record

    Night hall / Sense → Remember

    At 2:14 a.m., a small moment can become a long memory.

    Walking down the hall is ordinary. Keeping a record of it is a choice.

    A sensor may register a crossing. It does not know why it happened. The harder question is whether that moment should enter the household’s history—and for how long.

  6. CGI open service enclosure with an empty memory socket and a mechanical disconnect, with the removed cassette, access token, and keys on the shelf below.
    Plate H02.6 / Next morningCGI proposed safeguard · stopping and removing

    Utility wall / Explain → Forget

    The exit is part of the room.

    Turn it off. See what it kept. Delete the event. Know what deletion cannot promise.

    Stopping the system and deleting its memory are different acts. A receipt can confirm that a named record was removed. It cannot prove that every copy—or every influence on a model—is gone.

02 / One minute, opened

Interactive event choreography · illustrative household sequence

One gesture.
Eight accountable events.

The feature begins with a room appearing to welcome someone home. Open that minute and the seamless gesture separates into signals, inferences, retained context, policy, commands, and consequences.

Scenario clock—At threshold

What it feels like

“The room welcomed someone home.”

One quiet gesture. No visible request.

60seconds
opened

Trace at rest. The complete sequence remains readable below.

  1. A signal changed. No identity yet.

    Responsible component · contact sensor Inspect Sense
  2. Signals suggest activity in two places. Signals are not identities.

    Responsible components · occupancy and audio sensing Open the sensing boundary
  3. Events are joined to a routine. This exact chain is not established by the sources.

    Responsible component · hub or automation service Open the present assessment
  4. The system assigns meaning. “Familiar” is an inference, not an observation.

    Responsible component · rule or model Inspect Infer
  5. A scenario preference appears. What was retained—and where?

    Responsible component · profile or routine store Inspect Remember
  6. The returning preference conflicts with the room already in use. Whose rule applies?

    Responsible component · household policy Open the shared-room conflict
  7. Separate commands leave the controller. Coordination does not supply judgment or consent.

    Responsible components · hub and device interfaces Read the Matter 1.6 profile
  8. The room changed. The reason, scope, and override should remain visible.

    Responsible component · accountable interface Inspect Explain

One seamless moment. Separate responsibilities. The mapped sources document components that can sense, infer, coordinate, and act. They do not verify this exact minute, its timing, or one coherent household mind.

Scenario reconstruction · not a product trace, latency test, or description of one vendor system. Apple, Google, and Matter evidence is not presented as one combined product. Read the originating scene

03 / Two people, one room

Shared-space conflict study · illustrative household conditions

A room has no neutral default.

The system can guess correctly who arrived and still make the wrong choice. These four cases show how a smooth routine becomes a real argument about comfort, privacy, and control.

Overview Two reasonable preferences; one shared environment.

Person A / arriving

Bright hall.
Warmer rooms.

Basis
Prior preference
Authority
Not established

Person B / already here

Dim kitchen.
Cooler air.

Basis
Current explicit condition
Authority
Affected resident

Correct inference · unresolved permission

  1. ActApply the routineOne current condition is displaced.
  2. AskSurface the conflictPeople decide; automation interrupts.
  3. DeferPreserve the occupied zoneThe arrival routine waits or narrows.
  4. Do nothingLeave the room unchangedNo hidden resolution is imposed.

Every branch moves comfort, privacy, time, or control toward someone.

The same conflict, without the diagram.

Illustrative shared-room condition register
ConditionPeople and inputsAuthorityIf the routine actsPossible overrideUnresolved question
LightBright arrival scene / dim occupied kitchenPrior preference meets present needThe current condition is displacedAsk, narrow the zone, or leave manualWho authorized the wider scene?
ClimateWarm arrival / cool occupied zoneStored preference meets current settingThe comfort cost moves to someoneConfirm, hold the zone, or do nothingWhen may memory alter shared space?
PrivacyPersonalization / no indoor identity inferenceOne opt-in affects another personProcessing may extend beyond the account holderAsk, narrow sensing, or disable inferenceHow can one person opt in alone?
AuthorityPrimary account / affected residentTechnical control is not household consentThe account rule governs the roomShared approval, defer, or require a ruleWho may authorize, veto, and revoke?

04 / Local until it is not

Interactive boundary plate · documented fragments, not one product

Local until it is not.

A feature can run locally while the rest of the chain reaches elsewhere. Follow the cuts to see where the evidence holds—and where we simply do not know.

All paths Accountability domains arranged as a reading instrument, not literal topology.

Five accountability domains and three evidence paths. Their position explains questions of responsibility; it does not describe a universal product architecture.
  1. Path A / Documented local

    A task can be local. The system may not be.

    Apple documents on-device models. Google documents local execution for some automations. These are separate records, not one pipeline.

    Not established
    Local generative household interpretation, complete offline behavior, or the location of every prompt, output, log, and telemetry stream.
    Network-cut result
    No whole-system result inferred.
  2. Path B / Disclosed remote component

    A remote component has a network edge.

    Apple documents three server-based models running on Private Cloud Compute. Apple’s PCC requirements say request data must not be retained, including through logging or debugging, after a response; this is a vendor-stated boundary, not an independent audit or a feature-routing map.

    Network-cut result
    The generic remote segment is interrupted. No named feature inherits this state without documented dependency and fallback behavior.
  3. Path C / Unstated

    Silence is not local.

    Google’s Gemini for Home report lists devices and services but does not state execution location, network dependency, offline behavior, or fallback for its named AI capabilities. Matter coordinates supported components; it does not supply one universal local/cloud architecture. Other unstated processing, account, telemetry, retention, and fallback boundaries remain unknown.

    Network-cut result
    Unstated behavior remains unknown before and after the network cut.

05 / The memory residue

Interactive deletion-path study · proposed safeguard

The record can go.
The residue remains a question.

Deleting one named record is not the same as erasing every trace. This diagram follows an illustrative request and marks what the receipt can—and cannot—prove.

Illustrative request / H-0214 Remove the 02:14 hallway event from the visible household history. Scenario object—not a product control or legal result.

All objects and uncertainties are shown. Trace the request to emphasize its named boundary.

Six distinct records, effects, or accountability questions surrounding one illustrative deletion request.
  1. 01 / Inside illustrated scope

    Raw event

    Hallway occupied · 02:14

    Requested outcome: remove from the named visible-history store.

    Illustrative target · not proof of storage architecture
  2. 02 / Separate object

    Derived preference

    Prefer low light after midnight

    A related preference is a separate object. This diagram does not assert that one event created it.

    Outcome not established
  3. 03 / Separate object

    Profile association

    Routine associated with Resident A

    Removing an event does not establish that a resident or household profile changed.

    Outcome not established
  4. 04 / Unknown

    Possible learned influence

    Prior update or tuning effect?

    No specific influence is established. The dotted path asks whether one exists and whether it can be reversed.

    Existence and reversibility unknown
  5. 05 / Proposed safeguard

    Audit entry

    Request H-0214 · scope stated

    An accountability record would have its own purpose, access, and retention boundary.

    Separate record
  6. 06 / Limited proof

    Deletion receipt

    Visible-history event marked removed

    A receipt can report the named action. It cannot, by itself, prove what happened to backups, shared copies, profiles, or learned influence.

    Proof limited to stated scope

A reasoning diagram—not a deletion guarantee. This is not a product trace, vendor architecture, legal determination, or claim that every pictured object exists in a named system.

Current evidence supports a narrower conclusion: local inference does not establish that telemetry, logs, backups, or outputs remain local; smart-home research documents users’ varied and sometimes unclear understanding of how household data was collected and used; lifecycle guidance makes end-of-life support a product concern.

Artificially.cloud’s notice describes this website’s own request-data boundary. It is not evidence about a household product.

H03 / Five-year house

A conditional 2031 field study · evidence window 16 Oct 2026

The future lives in the utility room.

Five years is close enough that the studs, the budget, and the service call still matter. This is the same 3,500-square-foot house at the edge of the city—not a lab and not a billionaire compound. Most of it is ordinary. The speculative step is that real components begin to coordinate locally, explain what they did, and give way when someone disagrees.

  • Current

    A component or capability can be bought, deployed, or used now, though access and quality vary.

  • Emerging

    Real standards, products, or research systems exist; ordinary household availability is not established.

  • 2031 forecast

    A conditional design extrapolation for this exhibit—not an issued forecast, promised arrival date, or product claim.

CGI utility room with one robot folding a towel, a red first-generation Roadster charging in the adjacent garage, and a separate home battery and network cabinet.
Plate H03.1 / Utility core and garage2031 scenario · materially AI-assisted CGI

01 / The working edge

The glamorous part is that the laundry gets folded. The credible part is everything bolted to the wall.

A single mobile robot shares the utility room with the systems that make the house work: heat pump, filtered air, water isolation, battery, wired network, local compute, and the old shelf where spare parts accumulate. In the garage, the Roadster charges like a car—not like a magic battery for the building.

Boundary: The robot is a five-year extrapolation from research systems, not a claim of safe unattended domestic competence. The Roadster is not shown powering the house; bidirectional support for that exact vehicle and installation would have to be verified.

RoomHeadset off

CGI multipurpose room with a docked headset and two controllers, open floor space, a closed door and physical wall control, with neighborhood roofs and a skyline beyond the window.

ViewInside the headset

CGI concept of the same room seen through a headset, with a translucent robot service model inside a marked floor boundary; no physical robot or person occupies the room.
Plate H03.2 / The room and its digital layerCGI room / simulated headset view · materially AI-assisted

02 / Spatial room

The headset comes off. The room still has to work.

Most days it is a spare room with a reading chair and somewhere to put your books. Put on the headset and a digital model of the household robot could appear in the clear floor area, ready to inspect without moving the real machine. The door, seating, and physical controls still matter. Someone should be able to interrupt, and taking the headset off should be enough to return to the room.

Boundary: Headsets and immersive media exist now. A socially graceful room that coordinates lighting, sound, safety, and household interruption is the forecast layer. No headset pose, room scan, or bystander recording should enter the household memory by default.

H03 / Evidence register

Twelve systems. Twelve places to stop pretending.

The image is a synthesis; the evidence is not. Each entry names what is real, what is still uneven, and the line this house does not cross.

  1. 01 / Domestic roboticsEmerging

    One robot, with a job description.

    Research robots can already chain manipulation tasks such as collecting and folding laundry. Here, one shared machine works in the utility room instead of following people through every room.

    Boundary
    A laboratory task is not proof of safe, reliable, unsupervised household work. Stairs, children, pets, clutter, wet fabric, maintenance, and recovery remain hard.

    Physical Intelligence π0 research

  2. 02 / Yard and waterCurrent

    The outside robot stays outside.

    A robotic mower follows a mapped work area while drip irrigation responds to weather and several soil readings. The lawn still has thin spots, leaves, and edges a person has to trim.

    Boundary
    A mower is a narrow-purpose appliance, not a second household robot. One soil probe cannot represent a whole property, and weather-aware watering still needs calibration and winter maintenance.

    Husqvarna Automower 410 iQMatter 1.5 soil-sensor support

  3. 03 / Spatial mediaCurrent

    A dedicated VR area is mostly good architecture.

    High-resolution stereoscopic video, spatial audio, hand tracking, passthrough, and room-scale experiences are available. The house adds the unglamorous pieces: clear floor space, a seated mode, a visible dock, washable face interfaces, acoustic control, and a physical way out.

    Boundary
    Fidelity is not comfort, shared presence, universal accessibility, or privacy. A room designed around one headset can become obsolete; the safe floor and ordinary controls should remain useful.

    Apple Immersive Video technical overviewW3C WebXR Device APIApple Vision Pro privacy overview

  4. 04 / Local intelligenceCurrent

    The house can hear some things without calling home.

    Fixed voice commands can run on modest local hardware; larger on-device models are arriving in consumer systems. The utility cabinet gives them power, cooling, storage, and an ordinary network cable.

    Boundary
    Local inference does not prove local logs, telemetry, backups, accounts, or outputs. Open-ended speech and multimodal reasoning need more compute than a handful of household commands.

    Home Assistant local voice documentationApple on-device model research

  5. 05 / InteroperabilityEmerging

    The future is a shared vocabulary, not one brand.

    Matter and Thread let more controls, sensors, locks, energy devices, cameras, and appliances describe themselves across ecosystems. Two border routers keep the low-power fabric from depending on one box.

    Boundary
    A published standard is not universal implementation. Commissioning, multi-admin behavior, firmware support, and feature depth still vary by product and platform.

    Connectivity Standards Alliance: Matter 1.6Thread 1.4 feature paper

  6. 06 / Energy and mobilityCurrent

    The Roadster charges. The house negotiates everything around it.

    A home battery, heat-pump water heater, smart panel, solar inverter, and managed charger shift flexible loads without making the resident decode a utility tariff at dinner.

    Boundary
    Managed charging is not vehicle-to-home power. Grid programs, interconnection, equipment, and the exact car all determine whether bidirectional charging is possible.

    U.S. DOE managed and bidirectional chargingTesla Powerwall operationTesla charging during an outage

  7. 07 / Air and waterCurrent

    The best automation prevents an ordinary bad day.

    Particulate, carbon-dioxide, temperature, and humidity trends can change ventilation and filtration. Flow monitoring, point sensors, and a motorized main valve can catch some leaks before the basement becomes the story.

    Boundary
    Consumer sensors drift and cross-react; they are not clinical or regulatory instruments. Automatic valves can miss leaks or close by mistake, so testing, a manual lever, and a plumber still matter.

    U.S. EPA low-cost indoor-air monitorsU.S. EPA leak and flow-monitoring guideMatter 1.3 water and energy devices

  8. 08 / Privacy and access2031 forecast

    Permission becomes part of the wall.

    Camera shutters, microphone power cuts, room-level permissions, local-only indicators, tactile switches, and a mechanical key sit where residents and guests can actually find them.

    Boundary
    NIST defines capabilities, not one master household cutoff. The unified privacy station shown here is a design proposal. A software mute is not the same as removing sensor power.

    NIST IR 8259 Rev. 1Connectivity Standards Alliance: Aliro 1.0Apple accessibility features

  9. 09 / Health and careEmerging

    The bedroom gets an off switch before it gets a sensor.

    An opt-in bedside radar can estimate sleep and movement without a camera. A later system might notice an unusual fall-like event and ask before escalating to a named person.

    Boundary
    This is wellness support, not diagnosis or emergency certainty. Pets, bedding, another sleeper, room geometry, sparse radar returns, and consent all complicate the signal.

    Google Research sleep/wake radar studyPrimary mmWave activity and fall study

  10. 10 / AppliancesCurrent

    The kitchen coordinates. It does not cosplay as a chef.

    The induction range, hood, air system, washer, dryer, dishwasher, refrigerator, and water heater can expose state and energy use while retaining physical controls.

    Boundary
    Interoperability does not establish robot cooking, reliable food recognition, automatic shopping, or safe remote start. The resident can always run the machine now.

    Matter 1.2 appliance supportMatter 1.3 kitchen, laundry, and energy support

  11. 11 / Domestic networkCurrent

    The invisible house still needs labeled cable.

    Fiber, Ethernet, a PoE switch, local storage, a small AI appliance, UPS, ceiling access points, and Thread border routers make the network a serviceable part of the building.

    Boundary
    Wi-Fi 7 adds capacity, not perfect coverage; 6 GHz attenuates quickly through walls. Critical fixed systems still benefit from wire, ventilation, updates, and a documented fallback.

    Wi-Fi Alliance: Wi-Fi Certified 7Thread Group: Thread 1.4

  12. 12 / Repair and envelopeEmerging

    The house fabricates the drawer clip, not the water pipe.

    An enclosed desktop printer makes jigs, cable guides, adapters, and low-risk replacement parts. Electrochromic glazing and a matte e-paper panel quietly manage glare and household information without turning every surface into a screen.

    Boundary
    This is not a replicator. Printed gas, mains electrical, structural, medical, or pressure-bearing parts require qualification. Smart glass switches slowly, varies by climate, and should keep a conventional shade.

    NIST additive-manufacturing qualification reportU.S. DOE electrochromic window overviewE Ink Spectra 6

One plausible house, not one available product. The plates combine current components, emerging systems, and clearly labeled 2031 extrapolations. They do not establish that one vendor can deliver this house, that the parts will interoperate reliably, or that the resulting system would be affordable to every household.

Project-created · materially AI-assisted illustrations · primary sources linked above · independent human fact and media-rights review pending

Continue to the evidence record

06 / Time Spine

Related visions, separated by evidence.

Filter by time, then choose an entry to open its Evidence Margin. Keyboard: Tab enters the chronology; arrow keys move between visible entries. Everything is also readable in the list view.

12 chronology entries visible

Archive When a work or idea entered the record

These are the dates when works appeared or events happened. They do not move as the present moves.

Now cut · 12 Oct 2026 A dated view of current evidence

This is what the evidence showed on 12 Oct 2026. It will age.

Research windows Dates to check the evidence again

These are review dates, not promises about what arrives. No outcome, probability, or website timeline is forecast here.

  1. Read context

    2001: A Space Odyssey. Included as cultural context, not evidence that the film predicted the technology or determined its technical design. The later name “Hal” for MIT’s successor room is documented separately.

  2. Read context

    Star Trek: The Next Generation (1987–1994). An official franchise account identifies “Manhunt” and “Cost of Living” as episodes in which Lwaxana Troi directly addresses the Enterprise-D computer. This is a bounded example of the voice interface—not a claim of technical influence.

  3. Read context

    Weiser’s The Computer for the 21st Century is a ubiquitous-computing vision, not a description of generative AI, an omniscient assistant, or a unified intelligent room.

  4. Read context

    Michael H. Coen’s AAAI-98 paper describes the MIT Intelligent Room as a research platform for natural, multimodal human-computer interaction. Its successor was explicitly named Hal after the computer in the 1968 film.

  5. Read context

    Smart House. The comparison is thematic; no causal relationship is claimed.

  6. Read context

    A Philips Research chapter and the EU ISTAG scenario report made ambient intelligence an explicit object of design and policy discussion. Neither source establishes invention priority or proves that its scenarios would arrive.

  7. Read context

    The 2002 film Minority Report is included as a public-space surveillance boundary comparison, not as a domestic example or evidence of technical influence. Project-created, materially AI-assisted diagrams replace film imagery.

  8. Read assessment

    Evidence includes the Matter 1.6 feature release, certified Matter 1.6 software components, the latest listed Matter 1.6.x version at the cutoff, first-party on-device model documentation, and generally available home-platform APIs. These establish standards and vendor-reported capabilities—not seamless integration, verified product availability, reliability, adoption, or household consent. Read the narrower Matter 1.6 technology profile.

  9. Read the research horizon

    No forecast is issued; this is a check-in. The review looks for released features, certification records, product support, enabled experiences, and named markets. Any later forecast requires a scope, baseline, threshold, and resolution rule set in advance.

  10. Read the research horizon

    This is a research check-in, not a forecast. Local execution and some local automations already exist. What remains to be documented is where interpretation happens, what actions are allowed, who grants permission, what fails, and when the cloud takes over.

  11. Read the research horizon

    No arrival date is being promised. This review looks for evidence that interoperability, long-lived support, multi-person authorization, accessible non-voice controls, readable history, and enforceable privacy and safety work together—not merely appear as isolated features.

  12. Read the research horizon

    No date or probability is assigned. The open review tracks authorization, memory, liability, and the right to refuse across several overlapping paths. Those paths are not exhaustive, and none has a claimed arrival date.

07 / Comparison Shutter

One presence. Many systems.

Use the buttons to line the two readings up or pull their differences apart. The full comparison stays readable as text.

The rows align categories for inspection. Alignment does not make the imagined and observed systems equivalent.

Imagined / Editorial composite under review

One coherent presence

FunctionContinuous assistant and mediator
CapabilityConversation, memory, planning, action
MechanismPresented as a unified intelligence
OversightOften implicit or centralized
Social contextEveryday dependence is normalized

Observed / Editorial composite under review

Many partial systems

FunctionSeparate assistants, sensors, routines, and devices
CapabilityStrong in narrow tasks; continuity varies
MechanismFragmented models, services, and control layers
OversightUser-configured, vendor-governed, inconsistently legible
Social contextTrust, privacy, reliability, and household negotiation remain unresolved

08 / Horizon Window

What would change the view?

These dates are for returning to the evidence, not predicting an arrival. Each horizon names what would strengthen the case and what would weaken it. Any future forecast would go through a separate documented review.

Current posture

The uncertainty belongs in the exhibit.

Drivers and signposts

  • Smaller local models with practical latency.
  • Cheaper, lower-power sensing.
  • Interoperable device standards.
  • User-readable memory and permission controls.

Blockers and invalidators

  • Reliability remains too inconsistent for household routines.
  • Fragmented ecosystems prevent useful continuity.
  • Regulation or public resistance restricts ambient sensing.
  • Control remains opaque enough to erode household trust.

09 / Source Room

Provenance before polish.

This register gathers the sources the exhibition depends on most. We checked them against current primary or authoritative records; independent fact-checking and final editorial review are still required before publication.

  1. S-0012001: A Space OdysseyAFI catalog · Film record and 1968 release
  2. S-002Star Trek: The Next GenerationOfficial series record · 1987–1994
  3. S-002aThe Sound of Their VoicesOfficial Star Trek account · Computer voice and representative episodes
  4. S-003Smart HouseDisney catalog · Film record and official synopsis
  5. S-004Minority ReportAmblin official page · 2002 film; surveillance boundary case
  6. H-001The Computer for the 21st CenturyMark Weiser · Ubiquitous-computing vision
  7. H-003Design Principles for Intelligent EnvironmentsMIT AI Laboratory / AAAI · Primary prototype paper
  8. H-007Ambient IntelligencePhilips Research authors · TU Eindhoven bibliographic record
  9. H-006Scenarios for Ambient Intelligence in 2010EU ISTAG · Historical scenario exercise, not prediction
  10. E-001Matter 1.6 release notesConnectivity Standards Alliance · Released standard, not product adoption
  11. E-001aSmartThings Hub Platform certificationCSA · Matter 1.6 software-component certification; not availability proof
  12. E-001bMatter specification indexCSA · Matter 1.6.1 listed as latest 1.6.x version at cutoff
  13. E-002Third-generation Apple Foundation ModelsFirst-party on-device and PCC server-model architecture and evaluations
  14. E-002aPrivate Cloud Compute core requirementsVendor-stated request-data no-retention and no-logging boundary; not an independent audit
  15. E-003Gemini for Home platform reportVendor-reported capabilities; execution location, network dependency, and fallback unstated
  16. E-003aGoogle Home APIs Android release notesVendor-reported GA milestone and supported Matter versions
  17. E-003bGoogle Home APIs iOS release notesSome automations may execute locally; not evidence of local Gemini inference or complete offline behavior
  18. E-004Research Report: User Perceptions of Smart Home Privacy and SecurityNIST IR 8330 · Interviews about smart-home data understanding
  19. E-005Smart-home security and privacy perceptionsNIST SP 1343 · Survey of 401 U.S. users
  20. E-006Security and Privacy in Multi-User Smart HomesUSENIX Security · Month-long seven-household study
  21. E-007Enhancing smart home services among people with disabilitiesAssistive Technology · Qualitative study with 25 participants
  22. E-009Foundational Cybersecurity Activities for IoT Product ManufacturersNIST IR 8259 Revision 1 · Lifecycle and end-of-life guidance

What we still do not know

  • Whose context wins?How much product language and account design still assumes a single owner?
  • What survives a vendor?A room’s continuity is fragile if its memory cannot move, expire, or be independently audited.
  • When is restraint measurable?What language can describe a system that correctly declines to act?

10 / Reading Room

Rooms That Notice.

The Intelligent Room is one exhibition and one editorial issue. Each story enters the same system from a different human question.

01 / Feature · 12 min When the room begins to notice

A field guide to hidden seams, domestic memory, and the boundary between help and observation.

Read now ↗
02 / Visual essay Seven verbs in an empty room

How sensing becomes judgment—and judgment becomes architecture.

In development
03 / Household study A house with more than one person

Guests, children, workers, and the fiction of shared permission.

In development
04 / Exit manual How to leave an intelligent home

Deletion, deactivation, resale, portability, and the right to pull the plug.

In development