Strategy note: access-as-product and AI-with-strings museum signals

ROADMAP.md

@@ -1,6 +1,6 @@
 # ARCHAI Roadmap
 
-Last updated: 2026-06-07
+Last updated: 2026-07-11
 
 This roadmap turns the current progress logs, audits, and research notes into a practical build sequence. ARCHAI is the main app and source of truth. The public website is a demo window into that system.
 
@@ -104,6 +104,33 @@ Goal: make the project understandable to GLAM partners, RMIT, funders, and colla
 
 Status: improving.
 
+## Field Signals → Product Language
+
+Goal: turn what the museum field is asking for out loud into how ARCHAI
+describes itself. These are positioning/framing items that sharpen features we
+largely already have, not new engineering milestones.
+
+Prompted by two 2026-07 field signals (see
+`docs/STRATEGY_NOTE_ACCESS_AS_PRODUCT_2026-07-11.md`): the "museums are showing
+what they used to hide / access as a product" conversation, and the "AI gives
+museums a lifeline — with strings attached" data-homogenisation debate.
+
+- Make the **evidence boundary** a visible, demonstrable product feature: what
+  the system knows, what it doesn't, and where it got it — the direct answer to
+  the "AI slop / cultural homogenisation" fear.
+- Frame object detail as **"behind the record"**: the provenance / rights /
+  source-link layer is ARCHAI's equivalent of open storage and the conservation
+  window — value is in showing the working, not hiding it.
+- Carry a **data-sovereignty talking point** into partner/funder decks
+  (RMIT/pre-seed, outreach): position ARCHAI against both extremes surfaced by
+  the IWM/V&A licensing-marketplace story — giving heritage data away for free,
+  or renting it back from a platform.
+- Guardrail: do not pitch ARCHAI as "monetise access with an AI bot." Our
+  contribution is the trustworthy record beneath human access, keeping curators
+  and their labour visible.
+
+Status: captured, feeding positioning and outreach materials.
+
 ## Near-Term Build Queue
 
 1. Main app result browse/search parity with website demo.
@@ -113,3 +140,4 @@ Status: improving.
 5. Whisper/Piper prototype as separate optional path.
 6. Rights audit command before every public demo push.
 7. Generated AUXIO manifest sync for true page editing.
+8. Make the evidence boundary a visible product feature (Field Signals theme).

docs/RESEARCH_INDEX.md

@@ -15,6 +15,11 @@ The goal is to keep the conceptual, technical, and PhD-facing work easy to find.
 - [APP_FUNCTIONAL_AUDIT_2026-06-03.md](/Users/robgraham/Desktop/APPS/ARCHAI%20APP/docs/APP_FUNCTIONAL_AUDIT_2026-06-03.md)
   - functional review of the app and major workflow gaps
 
+- [STRATEGY_NOTE_ACCESS_AS_PRODUCT_2026-07-11.md](STRATEGY_NOTE_ACCESS_AS_PRODUCT_2026-07-11.md)
+  - field-signal capture: "access as a product" + "AI with strings attached"
+  - distilled positioning against AI-slop/homogenisation and free data give-away
+  - roadmap items (evidence boundary, "behind the record", data sovereignty) and a public response draft
+
 ## Historical Design
 
 - [HISTORICAL_DESIGN_BRIEF.md](/Users/robgraham/Desktop/APPS/ARCHAI%20APP/docs/historical-design/HISTORICAL_DESIGN_BRIEF.md)

docs/STRATEGY_NOTE_ACCESS_AS_PRODUCT_2026-07-11.md

@@ -0,0 +1,215 @@
+# Strategy Note: Access As Product, And AI With Strings Attached
+
+Date: 2026-07-11
+Status: field-signal capture + positioning note
+Source: two LinkedIn signals seen in-feed (screenshots archived), read against ARCHAI's current roadmap.
+
+This note captures two related conversations moving through the museum field
+right now, pulls out what is vital enough to steer the roadmap, and drafts a
+public response. It is a working note, not a public post. The response draft at
+the end is the thing meant for outside eyes.
+
+---
+
+## Signal 1 — "Museums are showing you what they used to hide"
+
+Regina Connell (Relevance, Resonance), 1d, edited. Distilled:
+
+- Museums are exposing the work they used to keep behind closed doors, and it is
+  irresistible to audiences.
+  - Rijksmuseum let the public watch the *Night Watch* restored **live, behind glass**.
+  - The V&A opened a quarter-million objects to **open storage**.
+  - The Smithsonian gave its conservation lab **floor-to-ceiling windows**.
+- Claudio Buda named this first as a **curatorial shift**. Connell argues it is
+  more than that.
+- The deeper claim: *we no longer trust the finished, polished version of
+  anything.* We trust the seams, the mess, the hand at work. We crave going
+  behind the curtain.
+  - Luxury already monetises this: Hermès has run a school, a traveling
+    exhibition, and a documentary series around **watching craftspeople work**
+    for ~20 years. Cartier does a private version for its best clients.
+- The provocation: **museums have the same "access luxury" that luxury brands
+  do — and have not turned it into something.**
+  - "Watch the conservator work" is today a perk buried in a membership tier,
+    "somewhere between the gala invite and free coat check."
+  - It *could — should — be a product.* With a name. A price. A deeper way to
+    engage with the museum.
+  - "What would you pay to watch a 400-year-old painting get its varnish
+    stripped, layer by layer? That's pretty juicy."
+
+## Signal 2 — "AI gives museums a lifeline — with strings attached"
+
+The Art Newspaper / FT-style feature by Naomi Rea (shared by Vikas Lohchab;
+illustration by Pete Reynolds). Distilled:
+
+- Facing dwindling visitor numbers and tight budgets, museums are adopting AI
+  characters and chatbots to revive exhibitions and engagement.
+  - Leighton House ran a voiced AI model of Frederic, Lord Leighton via QR codes.
+  - The Met built a chatbot mimicking a 20th-century socialite; the Dalí Museum
+    reanimated the painter by phone; an AI Theodore Roosevelt appeared at a
+    Presidential Library.
+  - Cultural strategist **András Szántó** frames conversational tech as the
+    inevitable *"language of our time, our future."*
+- The catch, per **Nataliya Kosmyna (MIT Media Lab)**: unreliable tools risk
+  **"cultural homogenisation."** Generative AI has been shown to weaken memory
+  pathways and problem-solving, and to homogenise thinking. To stop heritage
+  being diluted into "AI art slop," cash-strapped institutions must manage their
+  data **with absolute care.**
+- Ask Mona's Marion Carré: bespoke models trained on an institution's own
+  unvetted-but-owned material are "qualitatively different from popular LLMs
+  trained on vast tracts of internet text… garbage in, garbage out."
+- The Natural History Museum's Doug Gurr: museums sit on **a trove of specialist
+  knowledge — data — that is a font of untapped revenue.** London's Imperial War
+  Museum and the V&A are in a UK-government-backed pilot marketplace licensing
+  high-quality data to AI developers. Gurr's caution: public institutions should
+  stop giving database access to commercial users for free, and any partnership
+  must be "hard-wired" so cultural and ethical considerations, and the
+  institution's mission, survive the commercial deal.
+
+## Why these two signals belong in the same note
+
+They are the same argument from two directions:
+
+- Connell says the museum's scarcest asset is **authentic access** — the seam,
+  the hand, the process — and it is under-monetised.
+- The AI feature says the museum's scarcest asset is **specialist, trustworthy
+  data** — and it is being either given away for free or diluted into slop.
+
+Both land on the same imperative: **the institution should own, name, and
+govern the thing that makes it trustworthy — and not rent it back from a
+platform, or leak it away for free.** That is ARCHAI's whole thesis.
+
+---
+
+## Where this maps onto ARCHAI
+
+ARCHAI already argues for sovereign, source-grounded, rights-aware heritage
+infrastructure. These signals sharpen three things we can say more sharply:
+
+1. **"Behind the curtain" is a native ARCHAI surface, not a bolt-on.**
+   Our object pages already carry provenance, rights/licence status, source
+   URL, and interpretation as first-class, visible metadata. "Watch the process"
+   is the same instinct: expose the working layer, not just the polished label.
+   The conservation-camera version is theirs; the *documented-record* version is
+   ours. We show the seams of the record: where it came from, what is verified,
+   what the machine does and does not know.
+
+2. **"Absolute care with data" is our default, not their afterthought.**
+   Kosmyna's homogenisation warning and Carré's "garbage in, garbage out" are
+   exactly why ARCHAI preserves raw records, normalises into a canonical layer,
+   keeps rights/provenance/language/protocol visible, and treats speech,
+   translation, and AI interpretation as **optional, documented, replaceable
+   layers**. Our answer to "AI slop" is source-grounding and a visible evidence
+   boundary — the model says when it does not know.
+
+3. **The data-marketplace moment is a governance opportunity, not just a
+   revenue one.** IWM/V&A licensing into a government-backed marketplace is the
+   headline; Gurr's "hard-wired" ethics and mission clause is the real story.
+   ARCHAI's position: sovereignty means the institution keeps the hardware,
+   model, data, logs, and decisions — so it can enter (or refuse) a data deal
+   from a position of control rather than dependency. Own the machine, own the
+   model, own the conversation.
+
+### What to be careful about (so we don't become the thing we critique)
+
+- Do **not** pitch ARCHAI as "monetise access with AI." The access-as-product
+  idea is Connell's, and it is a *human-access* argument (real conservators,
+  real process). Our contribution is the trustworthy record underneath it, not
+  a paywall bot.
+- Do **not** overclaim "no hallucination." Keep our honest language:
+  source-grounded, bounded, designed to reduce confabulation, says when it does
+  not know.
+- Keep curators in frame. The seam-and-hand story is a story about **human
+  labour made visible.** ARCHAI should make that labour more legible, never
+  replace or obscure it.
+
+---
+
+## Vital items promoted to the roadmap
+
+Folded into `ROADMAP.md` as a new theme ("Field Signals → Product Language").
+Summary of what was promoted:
+
+- **Evidence-boundary as a visible product feature.** Make "what the system
+  knows, what it doesn't, and where it got it" a named, demonstrable surface —
+  the direct answer to the "AI slop / homogenisation" fear.
+- **"Behind the record" framing for object detail.** Lean into the provenance /
+  rights / source-link layer as the ARCHAI equivalent of open storage and the
+  conservation window: the value is in showing the working, not hiding it.
+- **Data-sovereignty talking point for partner/funder conversations.** Position
+  ARCHAI against the free-give-away and rent-it-back extremes surfaced by the
+  IWM/V&A marketplace story, for the RMIT/pre-seed and outreach decks.
+
+These are *positioning and framing* items, not new engineering milestones — they
+sharpen how we describe features we largely already have.
+
+---
+
+## Response draft (for public reply)
+
+Context: a reply to Regina Connell's post, usable trimmed as a LinkedIn comment
+or expanded into a short post of its own. Voice per
+`docs/SUBSTACK_STYLE_ANALYSIS.md`: human first, clear argument, moral spine, no
+hype, a little open-source refusal.
+
+### Short version (comment on Connell's post)
+
+> Love this — and I'd push it one layer deeper. The reason "watch the
+> conservator" is irresistible is the same reason the AI-in-museums story (Naomi
+> Rea's piece this week) has everyone nervous: trust now lives in the seam, not
+> the polish. Audiences want the process *and* the receipts.
+>
+> So if access becomes a product, the thing you're really selling isn't the
+> spectacle of the varnish coming off — it's the museum's authority to say "this
+> is real, here's how we know." That authority is a data asset. Right now most
+> institutions are either giving it away free to AI developers or watching it get
+> flattened into slop.
+>
+> The behind-the-curtain move and the data-governance move are the same move:
+> own the record, show the working, don't rent your own memory back from a
+> platform. That's exactly what I'm building ARCHAI around — source-grounded,
+> rights-visible, runs on the institution's own machine. Happy to show anyone
+> who wants to poke at it.
+
+### Longer version (standalone post)
+
+> **Museums are showing the seams. The harder question is who owns them.**
+>
+> Two things crossed my feed this week. Regina Connell on museums turning
+> "behind the curtain" — the *Night Watch* restored behind glass, V&A open
+> storage, the Smithsonian's glass-walled lab — into something you could name
+> and price, the way Hermès sells you the sight of a craftsperson at work. And
+> Naomi Rea on museums reaching for AI to survive shrinking budgets, with MIT's
+> Nataliya Kosmyna warning it could homogenise heritage into "AI slop" unless
+> institutions manage their data with absolute care.
+>
+> They're the same story. Both are about the one thing a museum has that a
+> platform doesn't: **authentic, accountable access to the real thing.** The
+> conservator's hand. The provenance. The record of how we know what we claim.
+>
+> Connell is right that this is under-monetised. But the asset underneath the
+> experience is *data* — specialist, hard-won, trustworthy. And right now most
+> institutions are doing one of two risky things with it: giving it to AI
+> developers for free, or letting a generic model blur it into something that
+> sounds confident and knows nothing.
+>
+> There's a third option. Show the working *and* keep the keys. Put the
+> provenance, the rights, the source, and the limits of what the machine knows
+> on the surface where visitors can see them. Run it on hardware the institution
+> owns. Let the model say "I don't know" out loud. Make the seam visible, and
+> make it yours.
+>
+> That's what I've been building — an open, source-grounded, rights-aware
+> heritage system that runs local-first, so a museum can open the curtain without
+> renting its own memory back from a platform. The criticism is the thing I made.
+> If you work in a collection and want to break it, come find me.
+
+---
+
+## Provenance of this note
+
+- Two screenshots each of the Connell post and the Rea feature were shared in a
+  working session and read into this note. The screenshots themselves are not
+  committed (third-party feed content); this note is the durable record.
+- Attributions above are transcribed from those screenshots and should be
+  double-checked against the original posts/article before any public quotation.