Child of Humanity — a critical read

1 · What it actually is, once the metaphors are removed

If you read one section, read this: the home page and the engineering pages describe two different products.

The home page introduces one of the most advanced technologies ever created: a decentralized relational factor-graph with spatial awareness, owned by humanity, whose currency's value is not fixed by a central authority but calculated continuously by the Natural Intelligence [home↗].

The build sheet, five clicks deeper, describes the launch system [technology-build↗]:

• Matching: Python service: LLM API for conversation and extraction, embeddings in pgvector, hybrid retrieval, explicit scoring policy.
• Ledger: PostgreSQL double-entry ledger … demurrage as a nightly job.
• Consensus: Hash-chained append-only event log in Postgres … Custodial. At launch the nonprofit trust holds that record.
• Pricing: Python service implementing the published index as pure, versioned functions. Coherence-derived pricing: the research horizon.
• Infrastructure: Containers on a managed platform … Managed PostgreSQL … The one thing never self-hosted.
• External services: LLM API · Speech-to-text · Embeddings · Geocoding · Push, email, SMS · Calendar.

That is a conventional, centrally hosted web application with a chatbot. It is a reasonable thing to build. It is not what the home page, the Instagram announcement, or the donor "Lineage" page describe, and none of the three say so.

The money side: a Delaware nonprofit (later a "Humanitarian Trust") receives gifts of $8.4M over 24 months [financials↗]; four for-profit spin-offs (Coherence Labs, Threshold, Unity Exchange, Regenera) then raise a further $9–12M, with founding donors hold[ing] contractual first-priority allocation in its first investment round — the priority right is written into the gift agreement [legal↗, ventures↗]. A combined $274M revenue scenario by 2031 is shown to those donors [ventures↗].

2 · The site contradicts itself — public page vs. technical page

No outside source needed — every row pairs a public claim with the project's own technical page saying otherwise.

This is the strongest material, because none of it needs an outside source. The project publishes a marketing layer and an engineering layer side by side; the engineering layer repeatedly says the marketing layer is wrong.

The pattern is explicit in the project's own words: What the project is betting on is not whether the science arrives — it is whether the discipline holds to describe each layer at the maturity it actually has [technology-stack↗]. On the pages a donor sees first, that discipline does not hold.

3 · The science, misapplied

The pitch borrows a real neuroscience framework as a metaphor and presents it as the mechanism that sets a currency's value.

3.1 The Free Energy Principle is a modelling framework, not an engine you can run an economy on

The pitch: a principle that has been written in mathematics, tested in living tissue, and ratified into a global standard [the-intelligence↗]. Three sleights in one sentence.

• "Written in mathematics." The FEP says that anything that persists at non-equilibrium can be described as if it minimises variational free energy under some generative model. It is a principle (a way of writing models), not an empirical theory that could be false — a point its own proponents concede and that the Bruineberg et al. (2022) Behavioral and Brain Sciences target article and its commentaries make at length: the "Markov blankets" that let one call a cell, a person, a city and a planet "selves at every scale" are a metaphysical reading layered on Pearl's purely statistical construct, and the two keep getting swapped.
• "Tested in living tissue." Isomura et al. (2023) is real: cultured rat neurons performing blind source separation can be modelled as reducing variational free energy. That is evidence that a particular neural preparation fits a particular generative model. It is not evidence that a community economy "learns the way life learns" if you describe it with the same equations — any more than fitting Newton to a pendulum lets you run a stock market on gravity.
• "Ratified into a global standard." The FEP was not ratified by anyone. The ratified thing is IEEE 2874-2025, a data-modelling and transaction protocol (see §4). Welding a neuroscience principle to a standards-body stamp is a rhetorical move, not a fact.

3.2 "Free energy is computable from a log of events" — only after you choose the model that defines it

The engine page's central promise: Coherence is a quantity … free energy is computable from a log of events … Not a survey. Not a rating. A number derived from whether the community's expectations of itself keep coming true [coherence-engine↗].

Surprise is only defined relative to a generative model and a set of prior preferences. Someone writes those. The same page admits it, in its last section: Whose preferences count, and how are they weighted? … is a governance question wearing a mathematical costume — and it is the deepest unresolved item in the whole framework. So "coherence" is the fit between behaviour and an operator-authored model of what the community should want. Rename it and it is a KPI. Dressed as physics, it borrows an objectivity it does not have. The five "indices" the page proposes — reliability, calibration, circulation, epistemic reserve, distinctness — are perfectly ordinary analytics (show-up rate, Brier score, Gini-style flow concentration, coverage, mutual information). Nothing wrong with them. Nothing "emergent" either.

3.3 Minimising surprise rewards predictability; the "fix" is an added term, not a theorem

The engine page raises the conformity objection itself and answers: Under any evaluation horizon longer than the immediate, distinctness lowers expected free energy. That is asserted, not shown. Whether a homogeneous network has higher long-horizon free energy depends entirely on the preferences and the environment you specify — it is a modelling choice, not a consequence of the maths. The practical remedy they then adopt is to add a distinctness index and ask a review circle who is this measurement penalising, and are they the community's dissenters? — i.e. a human committee watching for the failure mode the mathematics was supposed to preclude. That is the right engineering, and it falsifies the sales line: diversity is imposed from outside the objective after all.

3.4 "Learns the way life learns" vs. what ships

Their own ladder [coherence-engine §11↗]: Stage 0 (launch) No agents, no message passing, no learning in the money path; Stage 1 ordinary Bayesian work; Stage 2 (years 2–4) peer message passing, hard but buildable; Stage 3 — learned hierarchy and coherence pricing — RESEARCH. Every property the home page and Instagram attribute to the system today is Stage 2 or 3. Belief propagation on social graphs is also flagged by their own engineers: social graphs are nothing but loops. Loopy propagation … can oscillate or converge to something wrong.

3.5 Analogy stamped "Proven"

The Living Systems page pairs each layer with a biology fact and the label Proven [natural-systems↗]: the genetic code is universal, therefore a shared Postgres schema is like DNA; slime mould reproduced the Tokyo rail map, therefore "the NI matches needs to offers the same way"; starlings track seven neighbours, therefore peer confirmation is a murmuration; fig wasps are obligate mutualists, therefore a database transaction is "the honest exchange". In every case the biology is proven and the mapping is a simile. A label that says "Proven" next to a simile is the oversimplification the page is built from. The engine page quietly agrees: This is the mechanism behind the slime mould is immediately qualified as metaphor the moment it has to be specified.

4 · Citations that don't hold

The decorative references were checked against their sources; several say something different from what the site says they say.

5 · "Active inference takes 94% less energy than LLMs"

The headline hook of the launch post has no traceable source and is contradicted by the project's own stack page.

This is the hook of the Instagram announcement and the wager of this entire project on the Living Systems page. It is a category error wrapped around a vendor number.

• Category error. "Active inference" is a modelling framework; "an LLM" is a trained artefact. You cannot compare the energy of a framework to the energy of a model, any more than "Bayesian statistics uses 94% less energy than Excel". Energy belongs to a specific implementation on a specific task.
• The number. No source for "94%" was found. The figures the site itself cites (90% less data, 96% less compute, 97% smaller) are VERSES AI's self-reported, preliminary Atari-10K results against IRIS, a model-based reinforcement-learning agent — not against any LLM, not on language, not on matching, not peer-reviewed. VERSES' separate LLM comparison was a coding task (cost and speed), also self-reported.
• Irrelevance to this product. The product's conversational layer is an LLM API — the thing supposedly being replaced is the thing being called. The active-inference part that does ship (a per-member Bayesian model — a handful of distributions) is cheap because it is tiny and does something tiny; that is not a property of "nature".
• Their own engineers say so. Cite them as evidence the approach is real, never as our expected numbers. [technology-stack↗]

The Instagram post's broader framing — LLMs … are taught from the same extractive paradigm … Artificial intelligence is … currently being led by, mostly men who have no connection to the earth — is a moral claim about who builds AI, offered in place of a technical claim about what is being built. The what, per the build sheet, is the same stack.

6 · The economics

Every precedent cited is centrally operated and tiny; the growth curve asks you to believe ×30 in one year.

6.1 The benchmarks cited are all centrally operated — the opposite of the thesis

The plan's proof that a century of quiet evidence supports it rests on named systems [the-living-economy↗, growth↗]. Look at how each actually runs:

Every system that worked had a centre, a licence, or a state behind it — and the largest of them in the complementary-currency class reached tens of thousands of users in five to ten years. The site then projects 100,000,000 in five.

6.2 The growth curve

[growth↗] 2027: 5,000 members. 2028: 66,000. 2029: 2,000,000 (×30 in one year, from 5 communities to ~2,000 — ×400). 2030: 20M. 2031: 100M. The mechanism for the 2029 jump is a Starter Pack — a downloadable node image and playbook — that anyone can begin on a Saturday without a conversation with us. The same page says the pack refuses to launch circulation below a viable founding circle because thin supply breaks immediately, in any community under ~1,000. So the pack must both gate hard on density and spawn two thousand dense communities in twelve months with the Trust's headcount removed from the growth equation. No complementary currency in the cited literature has grown at a hundredth of this rate; the reason given — none of these had an intelligence that measures contribution — is the unproven Stage-3 research item from §3.4.

The "density" argument is also wrong on its own terms. Five thousand in one city is where the matching actually has to work — but density for a currency is penetration (share of the local economy you can spend in), not headcount. 5,000 members in a city of 500,000 is 1%; 2,000 in a bioregion of 10,000 is 20%. The pivot from "island of 2,000" to "city of 5,000" (v2 → v3) made the economy test thinner while calling it stronger.

6.3 Cost per member: $84 → $0.09 by moving the cost off the Trust's books

A thousandfold fall in cost per member across four years is not an efficiency story — it is an architecture story [financials↗]. The row for 2031 says how: Communities run their own nodes and pay their own stewards. The denominator grew and the numerator was reassigned to people who have not been asked. The infrastructure page prices what was moved: ~$3k hardware + ~$6k/yr per CD, $5–10 compute per member per year — on 60,000 communities and 100M members that is hundreds of millions a year, borne by donated, municipal or cooperative infrastructure. A plan whose cost advantage is "someone else pays" is not cheaper.

6.4 The $274M venture scenario

[ventures↗] The arithmetic checks (120,000 × $400 + 90,000 × $600 = $102M, etc.). The drivers do not. 120,000 paying organisations in a Sardex-style motion, when Sardex has ~4,000 after ten years. $16B circulating annually from 40% active × $400 per active member — "deliberately modest", for a unit that is non-convertible by legal design in Year 1 and whose price is an administered index. 1.8M hectares of carbon/biodiversity verification by members photographing rivers, in a market where registries (Verra, Gold Standard) require accredited third-party validation bodies — "the people who live there" are not a VVB. 800 enterprise licences at $75k for a "coherence measurement" product whose core risk the brief itself names as measurement validity.

The disclaimers (scenario arithmetic, not forecasts) are present and correct. The problem is upstream: the drivers are not estimates, they are the story expressed as integers.

7 · Legal and structure

A tax-deductible gift that carries an investment right, and a currency that must not be convertible to stay legal — next to a spin-off that converts it.

7.1 Tax-deductible gift + contractual investment right

Delaware nonprofit → Humanitarian Trust (Y2) … 501(c)(3) deductibility path and, two paragraphs later, founding donors hold contractual first-priority allocation in [the spin-off's] first investment round — the priority right is written into the gift agreement [legal↗]; the ventures page sells that right as materially more valuable. A charitable contribution that returns a private financial right to the donor is, at minimum, a quid-pro-quo question for deductibility and a private-benefit question for the charity's exemption. The page itself says: This analysis is planning-grade, prepared without counsel. Nothing on the public home page says so.

7.2 Closed loop or fiat bridge — pick one

Covered in §2. The legal posture (below the line, no cash-out, Howey fails by design) is sound for a time bank. The product roadmap, the Living Economy page and the Unity Exchange brief sell a fiat on/off-ramp with custody and multi-chain swaps. Both cannot be the plan. If the bridge is the plan, Year-1 legal roughly doubles and Phase 3 moves forward two years — in their words.

7.3 Governance of the thing that sets value

The Trust is governed with the Governance Interface (Yes / No / Surrender) [legal↗]. That is the entire description of how a body that will author the price index, the preference priors, the neighbourhood boundaries, the demurrage rate and the indices is governed. "Surrender" is not a governance primitive anyone can act on.

7.4 Undisclosed vendor relationship

The co-founder named on the site, Dan Mapes, is a founder of VERSES AI and founder/director of the Spatial Web Foundation (public bios; former VERSES president per LinkedIn-derived profiles). The plan budgets a VERSES / Genius engagement of $180k + $80k — the largest infrastructure line, 63% of Year-1 infrastructure [infrastructure↗]; cites IEEE 2874 (authored through the Spatial Web Foundation) as ratified global standard at every turn; and cites VERSES' benchmark numbers as its efficiency proof. None of the 23 pages discloses the relationship. A donor reads "the standard", "the benchmarks" and "the vendor" as three independent legs; they are one.

7.5 "Elders and community give the final word" on child and elder care

The trust layer promises communal vetting for care roles. The stack page: right in principle and unbuilt in practice … touches background checks, safeguarding law, and liability in every jurisdiction differently … Treat it as a launch blocker. The public page presents it as a feature.

8 · Surveillance in a nature costume

The same measurement engine is a privacy promise on one page and an employee-analytics product on another.

The engine page anticipates the charge — anyone who worries this is surveillance wearing a nature costume — and answers with Markov-blanket privacy: peers receive claims and confidences — never the evidence behind them. That describes Stage 2, years away. What ships is a custodial database the operator can read, holding contributions, exchanges, and care records, feeding a trust scorer that becomes a weight the network applies to a member's predictions about themselves, and five community-level indices. The page insists these are not a leaderboard and never a rank — then the Coherence Labs brief turns the same engine into a product sold to large employers, HR platforms, ESG reporting and municipal dashboards: Coherence Labs measures what [engagement platforms] can only ask about [coherence-labs-1↗]. Institutions and enterprises pay. People never do — but what the institutions buy is a reading of their people's real patterns of exchange, contribution, and relationship. That is a behavioural-scoring product for employers, and the lineage it actually belongs to is people analytics, not mycelium.

Two further points the site makes for you. The Framework of Coherence is built from members' answers to What are the most loving ways to relate to life, to others, and to yourself?, aggregated by the system into the substrate the whole system runs on that every match, every exchange, every collective decision moves through [home↗]. An economy whose matching and pricing are conditioned on a model of what its members said love means is a values filter with a wallet attached. And the indigenous partnership — the moral centre of both Instagram posts — is, in the budget, $360k of $8.4M (4%), starts in Year 2, is not yet named as to nation or country, and is described in the marketing plan as The Lineage as outreach spine [financials↗, index↗].

9 · What is sound — and why it does not rescue the pitch

What holds up — and why it supports a small pilot, not this ask.

10 · Ten questions that would settle it

Send these before wiring anything.

1. Which published page is the offer — the home page's "decentralized intelligence calculating value", or the stack page's "administered index, custodial record, LLM API"? Will the home page be corrected to match?
2. Show Bliive's circulation data: active users per month, completed exchanges per year, retention, for the thirteen years claimed.
3. Is Unity convertible to fiat in Years 1–2 or not? If yes, which money-transmission licence, in which jurisdiction, and what does "Howey fails by design" rest on once the asset appreciates and is swappable?
4. Has tax counsel reviewed a 501(c)(3) gift that carries a contractual allocation right in a for-profit round?
5. Where does "94% less energy" come from? Name the task, the baseline and the measurement.
6. Disclose the financial relationship between the co-founder, VERSES AI and the Spatial Web Foundation, and the terms of the $260k VERSES engagement.
7. Which complementary currency has ever grown ×30 in a year? What is the evidence the Starter Pack produces 2,000 density-viable communities in 2029?
8. Who writes the preference priors and the price index, how are they amended, and what can a member appeal?
9. Name the indigenous nation, the agreement terms, and what "co-design" controls — before it is used as the fundraising narrative.
10. Which cities, which countries? "Countries 2–4 not yet named" is on the legal page of a plan that prices four jurisdictions.

Independent review. Not affiliated with Child of Humanity, VERSES AI or the Spatial Web Foundation. Corrections with a source are welcome and will be applied; the project is invited to point to any page that resolves a contradiction listed here. Quotations are reproduced for the purpose of criticism and review and link to their source.