XX26005 Fungible and Non-fungible Entities. V01 080226

 A fungible thing is something where each unit is interchangeable with any other unit of the same kind, with no loss of value or meaning.

If you swap one unit for another, nothing changes.

Simple definition

Fungible = interchangeable

One unit can replace another perfectly.

Clear examples of fungible things

• Money

One $10 bill is equivalent to any other $10 bill.

• Commodities

Oil, wheat, gold (of the same grade). One barrel of Brent crude is the same as another.

• Shares of stock

One share of Apple stock = any other Apple share.

• Electricity

One kilowatt-hour is the same no matter where it comes from.

Non-fungible things (the contrast)

These cannot be freely swapped without changing value.

• A specific house

• An original painting

• A signed first edition book

• A particular editor’s judgment or taste

• A person’s experience or skills

This is why NFTs are called non-fungible tokens — each one is meant to be unique.

Why this matters in Hidalgo’s thinking

Hidalgo’s big move is arguing that knowledge is non-fungible.

Knowing how to:

• edit a literary novel

• market a romance series

• negotiate foreign rights

These are distinct capabilities, not interchangeable units.

You can’t say:

“We have lots of knowledge, so we’ll just reallocate it.”

You either have that knowledge, or you don’t.

Quick test

Ask:

“If I swap this unit for another of the same type, does anything important change?”

• If no → fungible

• If yes → non-fungible

If you want, I can explain how this applies to jobs, AI, data, or creative work next.

Great choice — AI + data is where fungibility gets really subtle (and where a lot of people get misled).

Is AI data fungible?

Short answer: Some of it is, most of the valuable parts aren’t.

Let’s break it down cleanly.

1. Fungible AI data (commodity data)

This is data where one unit can replace another with little loss of value.

Examples

• Millions of generic images scraped from the web

• Large volumes of public text (Wikipedia, news, forums)

• Sensor readings at scale (temperature, clicks, logs)

• Random speech samples in major languages

Why fungible:

If you swap one batch for another of similar size and distribution, the model barely notices.

Economic effect:

• Value trends toward zero over time

• Advantage goes to whoever can collect at massive scale

• This is why open datasets and scraping dominate early AI waves

2. Non-fungible AI data (capability data)

This is where real power lives.

Examples

• Expert-annotated datasets (doctors, lawyers, editors)

• Proprietary workflows (how a newsroom edits stories)

• Longitudinal data (the same users or systems over time)

• Feedback loops showing why decisions were made

• Edge cases and failure modes

Why non-fungible:

You can’t replace:

• a cardiologist’s labels

• a senior editor’s revisions

• a decade of consistent user behavior

with “more generic data.”

Each dataset encodes specific know-how, not just information.

3. Hidalgo’s lens: data ≠ knowledge

Hidalgo would say:

• Data can be fungible

• Knowledge almost never is

AI systems don’t just need data — they need structured experience.

A million random examples ≠ one deeply informed one in the right context.

4. Why “more data” often stops helping

Teams hit diminishing returns because:

• they already covered the easy cases

• remaining performance gaps are qualitative, not quantitative

• the missing piece is judgment, not volume

This is why:

• small, high-quality datasets can outperform massive noisy ones

• fine-tuning beats retraining from scratch

• human-in-the-loop systems scale better than pure automation

5. Publishing + AI (your earlier thread)

In publishing, the most valuable AI data is non-fungible:

• annotated edits showing why a sentence was changed

• acquisition decisions with context

• cover tests linked to audience response

• long-term author development notes

A scraped corpus of books helps models speak English.

A curated editorial dataset teaches them taste.

6. Strategic mistake to avoid

“We’ll just get more data.”

That works only when data is fungible.

The real question should be:

“What capability does this data encode?”

If the answer is “none in particular,” you’re buying a commodity.

Rule of thumb

• Scale wins with fungible data

• Depth wins with non-fungible data

The strongest AI systems combine both.

If you want, I can:

• map this to training vs fine-tuning

• explain why synthetic data only partly works

• or help you identify what non-fungible data your organization already has

In publishing, the most valuable things are non-fungible — they cannot be swapped, copied, or scaled without losing what makes them valuable. Here’s a clear, practical map of what that actually is.

Core non-fungible assets in publishing

1. Editorial judgment (taste)

Two editors with the same résumé are not interchangeable.

What’s non-fungible:

• the ability to spot a manuscript’s latent potential

• knowing what to cut vs protect

• sensing timing and cultural readiness

• understanding when not to publish a book

This judgment is accumulated through years of lived experience and failure. You cannot replace it with:

• more data

• AI predictions

• another editor “at the same level”

2. Author–editor relationships

You cannot swap relationships.

What’s non-fungible:

• trust built over multiple books

• shared language for feedback

• emotional safety during revisions

• credibility when pushing hard edits

This is why authors often follow editors across houses — the knowledge lives between them.

3. Genre-specific know-how

Knowing publishing ≠ knowing romance publishing ≠ knowing academic monographs.

Non-fungible elements:

• tropes and reader expectations

• pacing instincts

• cover semiotics

• pricing and release rhythms

• influencer and bookseller ecosystems

You can’t redeploy a literary editor into romantasy without friction — the alphabets are different.

4. Institutional memory

This is the “why we don’t do that anymore” knowledge.

Non-fungible because it includes:

• failed experiments

• historical reader backlash

• agent politics

• market scars

When senior staff leave, this memory disappears — and mistakes get repeated.

5. Reputation and trust

You can’t buy credibility.

Examples:

• an imprint booksellers trust

• a house known for editing quality

• a children’s publisher parents trust

This reputation compounds slowly and collapses quickly. No amount of marketing can instantly substitute for it.

6. Backlist stewardship knowledge

Not the books — how to keep them alive.

Non-fungible know-how includes:

• when to relaunch

• how to repackage

• which formats unlock new readers

• how to reposition without alienation

Two publishers owning the same rights will get different outcomes.

7. Sales intuition and human persuasion

A great sales rep’s pitch is not replicable.

Non-fungible:

• knowing which buyer will respond to which angle

• reading hesitation in a meeting

• adjusting narrative in real time

CRM data records what happened, not why it worked.

8. Curatorial identity

An imprint’s “voice” is not the sum of its books.

It includes:

• what it refuses to publish

• pacing of releases

• aesthetic coherence

• moral and cultural stance

This identity cannot be reverse-engineered quickly and is damaged by incoherent expansion.

What looks fungible but isn’t

• Manuscripts (until edited)

• Marketing plans

• Metadata templates

• Sales scripts

• AI models trained on generic corpora

These are scaffolding. The value emerges only when non-fungible human knowledge activates them.

Why this matters right now

AI and consolidation push publishers to treat work as interchangeable.

Hidalgo’s warning applies perfectly:

If you manage non-fungible knowledge as if it were fungible, you destroy it.

One-sentence test

Ask:

“If we replace this person/process with another ‘equivalent’ one, do we lose something invisible but important?”

If yes — it’s non-fungible.

If you want, I can:

• map this to AI adoption without hollowing out value

• design a non-fungible asset audit for a publishing house

• or apply it to authors’ careers