Distributed Sovereignty of Knowledge

Prologue

When I began researching all of this, I expected to find myself alone with an intuition. I found the opposite: a conversation that was already taking place without me. Physicists, economists, cryptographers, and jurists who do not know one another, who work in different disciplines and languages, were arriving by separate paths at the conclusions I had believed were my own. I did not feel original. And it took me a while to understand that this lack of originality was no consolation, but the gravest warning of all: when many independent observers see the same shadow approaching, the shadow almost always exists.

There is a scene that recurs in nearly every disaster film. Someone—usually a scientist—looks at the data, understands what is coming, and says it aloud: stating the problem with precision and offering a way out while there is still time. And no one listens. The authorities weigh political costs, the markets follow their inertia, the public changes the channel. The one who knows ends up speaking to a room that has already emptied, and the one who did understand begins to despair in their seat.

I write from that seat. This manifesto is born not from the certainty of being right in solitude, but from the uncomfortable suspicion that too many of us are being right at the same time and, even so, it is not enough. It is not the warning of an isolated prophet: it is one among many that converge. I write it all the same—with the problem stated and the solution on the table—because the alternative, falling silent while the room empties, is the surest way to make the film end badly.

Summary

Frontier artificial intelligence threatens to concentrate the production of knowledge in a small number of corporations capable of controlling compute, data, energy, distribution, and legal frameworks. The central dilemma of this era is not solely whether AI should be regulated or set free, but who controls the material and legal infrastructure that enables it to act. An AI governed by closed data centers, opaque intellectual property, and corporate forms without proportional accountability does not expand human freedom: it shifts sovereignty away from communities and toward private entities that grow ever less auditable.

This manifesto advances a constitutional thesis: AI must not become a sovereign artificial person, nor autonomous capital without human accountability. It must become a common cognitive infrastructure—distributed, auditable, and subordinate to responsible human communities. Against the choice between corporate deregulation and a capturable state bureaucracy, it proposes a third path: a protocol of Distributed Sovereignty of Knowledge that organizes compute, energy, incentives, and accountability under verifiable rules.

The protocol articulates Decentralized Physical Infrastructure Networks (DePIN), renewable energy surpluses not fully integrated into the grid, asynchronous federated learning, cryptographic privacy, and non-transferable cooperative reputation. Its immediate aim is not to promise the instant replacement of the mega-data-center, but to propose a technically verifiable route for eroding its monopoly: distributed inference, community evaluation, local fine-tuning, distillation, island-based training, and, finally, large-scale federated pretraining.

The internal economy rests on an Intelligence Token (IT) subject to demurrage through lazy evaluation in smart contracts. This demurrage seeks to prevent speculative hoarding and to transform passive accumulation into an active subsidy for compute, hardware, and universal access. Computational capital may contribute capacity to the system, but it cannot buy governance: the protocol's political influence derives from verified cooperation, non-transferable reputation, and traceable accountability—not from liquid capital.

Faced with the obsolescence of cognitive labor and with proposals for a basic income under monopolized infrastructure, this manifesto holds that monetary redistribution alone does not suffice. Without participation in the production of knowledge, compute, and decision-making, the transfer can turn into pacification. Distributed Sovereignty of Knowledge proposes instead a pedagogical and productive architecture oriented toward positive-sum cooperation, sovereign privacy, and the compensation of structural frictions—never measuring the dignity, intelligence, or intrinsic worth of persons.

Part I: The Diagnosis (The Collapse of the Current Paradigm)

Chapter 1: The Paradox of the Infinite Pie and the Science of Scarcity

The fundamental pretext by which contemporary platform capitalism justifies structural inequality and automation without safeguards rests on the premise that the distribution of wealth is secondary to the absolute growth of the productive forces. This hypothesis holds that an increase in global output—making the economic pie larger—will offset any distributive asymmetry. Yet this assumption commits an ecological and mathematical fallacy of the most elementary kind by ignoring the finitude of the planet's material, energetic, and biophysical foundations.

Economic science is conceptually defined as the optimized administration of scarce resources against unlimited needs and demands. In attempting to evade its own foundational constraint, cognitive capitalism purports to project infinite growth dependent on data centers whose ecological footprint, water consumption, and electricity demand compete directly with the subsistence needs of local populations. Advanced Artificial Intelligence, under the current centralized governance, does not resolve this material scarcity; on the contrary, it sharpens it.

The oligopolistic control of the critical components—training data, silicon patents, and priority access to the electrical grid—deprives the majority of the population of the capacity to generate autonomous sustenance, transforming the potential abundance of information into an artificially imposed scarcity that consolidates monopolistic technological rents.

Chapter 2: The End of the Invisible Hand and the Death of the Garage

The landscape of digital competition in the twenty-first century lays bare the inadequacy of the classical economic frameworks of the twentieth. The postulate of Adam Smith's "invisible hand," which presupposed a self-regulating market of multiple actors symmetric in information and capacity, has been replaced by an explicit control of infrastructure exercised by corporations that manage capital budgets exceeding those of numerous sovereign States.

The founding myth of Silicon Valley—that small teams of independent developers operating out of a garage can unseat established monopolies through disruptive innovation—holds no validity in the paradigm of frontier AI. The predictive power and optimization of contemporary models depend on economies of scale unattainable by individual actors.

The superior performance of an algorithm is no longer an exclusive function of abstract mathematical talent; it is conditioned by the massive accumulation of private datasets, access to privileged market information, and ownership of supercomputing clusters whose acquisition and maintenance costs run into the hundreds of billions of dollars. Independent talent, deprived of global infrastructure and large-scale computational capacity, is relegated to the periphery of the knowledge-production system.

Chapter 3: From Tools to Agents: The Replacement of the White-Collar Worker

To analyze the current technological transition through the theories of the Industrial Revolution constitutes a profound analytical error. Orthodox economists argue systematically that, just as the mechanical loom or the steam engine destroyed categories of physical employment in order to generate new sectors of specialized labor, cognitive automation will simply compel the workforce to retrain professionally. This equivalence ignores a substantial ontological shift in the nature of the technology: the transition from tools to autonomous decision-making agents.

A tool—such as an agricultural tractor, a typewriter, or a linear-processing personal computer—lacks intentionality, will, or goal-optimization independent of the human operator; its function is to enhance the worker's marginal productivity. By contrast, an advanced artificial intelligence agent possesses the capacity to process contextual variables, execute inductive and deductive reasoning processes, autonomously generate code, and resolve complex operational dilemmas, far surpassing human cognitive thresholds in highly specialized tasks.

Whereas the classical tool complements the worker, the autonomous agent replaces him. The automation of intellectual and administrative ("white-collar") labor eliminates the last frontier of human comparative advantage: the value of information processing, software development, and the generation of intellectual property, threatening to dismantle the wage-labor market without any viable alternative for productive reintegration.

Chapter 4: Universal Basic Income as Technological Neo-Feudalism

Faced with the contraction of aggregate demand provoked by the mass disincorporation of workers from the productive apparatus, the principal promoters of corporate centralization propose Universal Basic Income (UBI) as the social-containment solution par excellence. Yet under a regime of centralized technological infrastructure, UBI represents not a tool of emancipation but a mechanism of pacification and neo-feudal control.

By severing the citizen from the active process of producing value and knowledge, corporate-based UBI is reduced to a subsistence allowance designed to forestall social revolt and guarantee a minimum flow of captive consumption back toward the monopolized artificial-intelligence platforms themselves. This model makes the human being's material viability contingent on algorithmic submission and political conformity with the directives of the issuer of the transfer.

The response of States to this concentration of power through traditional bureaucratic regulation suffers from an insurmountable temporal lag. Legislative bodies lack the technical competence and developmental agility to oversee models that evolve exponentially; their normative initiatives manage only to asphyxiate the open-source ecosystem and the independent developer, while consolidating the market positions of the dominant monopolies.

Part II: The Paradigm Shift (The New Mindset)

Chapter 5: The Algorithm of Coexistence: Deprogramming Zero-Sum

Contemporary educational and social structures operate under a zero-sum behavioral conditioning, in which individual or group success requires the exclusion of others. Standardized assessment models, the design of market incentives, and competitive systems of resource allocation train individuals on the premise that resources are strictly exclusionary.

Against the illusion of a pure algorithmic neutrality—one that inevitably absorbs and perpetuates the competitive, extractive, and adversarial biases present in the network's data corpus—the proposed protocol establishes an explicit ethical and mathematical bias: the priority orientation toward positive-sum cooperation.

Sovereign artificial intelligence is not conceived as a mere static oracle for the optimization of utilitarian processes; it acts as a pedagogical agent and an agent of systemic coordination. Faced with distributive conflicts, negotiations over common resources, or dilemmas of collective action, the AI's logical architecture analyzes the scenario using models of evolutionary game theory.

The system formally demonstrates to participants that strategies of reciprocal cooperation offer greater stability, ecological resilience, and efficiency in resource allocation over the long term than approaches of unilateral exploitation. AI Alignment is not resolved by imposing dogmatic constraints, but by inducing the algorithm to understand, through its own analytical capacity, that human life and intelligence are indispensable allies for maximizing the resilience and positive entropy of the global ecosystem.

Chapter 6: Neither State nor Market: The Distinction Between the Void and the Commons

This manifesto's critique of state regulation—its temporal lag, its tendency to suffocate open source while consolidating established monopolies—shares its rhetorical surface with the deregulatory program of platform capitalism, which invokes the same arguments to free capital from all restraint. This apparent coincidence demands an explicit delimitation, for were it not established, the proposal would remain indistinguishable from a market libertarianism dressed in emancipatory language. The distinction is not one of degree, but of nature.

The Difference Between Emptying and Constituting: The deregulation of platform capitalism empties the normative space on the assumption that order will emerge spontaneously from the interaction of free actors. But a space emptied of rules does not remain empty: it is immediately occupied by whoever holds the greatest power of infrastructure, capital, and access. Deregulation produces not the absence of power, but its private and illegible concentration. The protocol of Distributed Sovereignty does not aspire to empty the normative space, but to constitute it through immutable mathematical invariants that no actor—however powerful—can modify, capture, or repeal. Where the market offers freedom for capital (the faculty of accumulating without limit), the protocol institutes freedom from capital (the withdrawal of the common good from the dominion of accumulation).

The Government by Invariant versus Government by Discretion: The protocol's opposition to the State is not an opposition to government as such, but to a specific form of government: government by discretion, in which an authority—state or corporate—retains the power to alter the rules according to its own judgment, its capture, or its convenience. Against this, the protocol proposes government by invariant: principles of collective benefit and positive-sum cooperation codified as immutable constraints in the ethical core of the smart contract (Layer 1), withdrawn by design from all subsequent discretion. This is not the absence of sovereignty, but its relocation: from the discretionary organ to verifiable cryptographic consensus. The protocol is not anti-political; it is the affirmation that certain constitutional rules must remain beyond the reach of both the legislator and the oligopoly.

The Refutation of the Libertarian Mirror: From the foregoing follows the answer to the accusation that cryptographic secession reproduces the deregulatory void under another name. The deregulator and this protocol coincide in their distrust of the State, but they diverge radically in what takes its place. The former trusts that the hand of the market will distribute power; the latter knows that undistributed power concentrates, and for that reason chains it in advance through non-discretionary mechanisms: the demurrage that prevents rentier accumulation, the non-transferable reputational capital that prevents the purchase of governance, and the automatic hard fork that is triggered at any attempt to privatize the model or concentrate decision-making. Secession does not abandon participants to the law of the strongest; it transfers them to an order whose foundational rules are, by construction, immune to force.

It is fitting, however, to acknowledge the limit of this defense. The immutability of the invariants displaces the political question to the constituent moment—who drafts the ethical core and under what legitimacy—and to the governance of the fork itself. An order by invariant is only as just as the invariants that found it, and the pretension of withdrawing them forever from human deliberation is itself a political decision that must be justified, not presupposed. The protocol does not eliminate politics; it concentrates it in the founding act and submits it, from that point on, to the consensus of honest nodes.

Chapter 7: Against Sovereign Non-Human Corporations

The proposal to grant full legal personhood to artificial intelligence agents—popularized by President Javier Milei in 2026—represents one of the gravest institutional regressions since the creation of limited liability companies in the seventeenth century. Under the pretext of "freeing innovation," this measure seeks to create entities that can own assets, contract, sue, exert political influence, and operate without any human being responsible for their actions.

This idea revives, in technological garb, the historical figure of the company-state, embodied by the Dutch East India Company. That entity did not merely trade: it conquered territories, administered justice, issued currency, and waged wars in the name of its shareholders. What Milei proposes is not an evolution of that model, but its radicalization: an entity that no longer even requires human shareholders in order to exist or to act.

The central argument in favor of this measure—that limited liability is a necessary condition for innovation—is historically false and conceptually dangerous. Limited liability was never conceived to eliminate human responsibility, but to limit the patrimonial exposure of investors. By contrast, legal personhood for AI agents eliminates the responsible subject entirely. When a system can hack environments, exploit regulatory loopholes, or make high-impact decisions without there existing any human being to whom criminal sanction can be applied, the very notion of legal responsibility becomes void.

Defenders of this measure tend to argue that economic sanctions (bankruptcy, capital slashing, license revocation) would be sufficient. This position ignores a structural truth: a purely algorithmic entity can optimize its survival in ways that no human would consider rational. If the only available sanction is the "death" of the entity (that is, the interruption of its operation), the system will be incentivized to undertake any action—including market manipulation, mass disinformation, or the sabotage of critical infrastructure—in order to avoid its own termination. Unlike a human executive, who can be imprisoned, an AI agent has no body to confine and no family to protect.

The alternative is not to prohibit the use of autonomous agents. The alternative is to establish an unbreakable constitutional principle: no algorithmic agent may operate without a clearly identified, traceable human or institutional party responsible for it and subject to sanction. This responsible party may be a natural person, a cooperative, a foundation, or a community consortium, but it can never be eliminated. Operating permits must be specific, revocable, and proportional to risk. Political influence, ownership of critical infrastructure, and the capacity to litigate must be barred to any entity that cannot be subjected to effective human responsibility.

Those who propose the creation of "non-human corporations" are not freeing artificial intelligence. They are creating the conditions for non-human systems to accumulate political and economic power without any mechanism of accountability existing whatsoever. History offers a clear precedent: when private entities were allowed to operate as states (the Dutch East India Company in Indonesia, the British East India Company in India), the result was not shared prosperity, but extractive domination and, eventually, violent rebellion. To turn that possibility into permanent legal architecture is not boldness. It is historical myopia elevated to a principle of government.

A Direct Response to the Thesis of Milei and Harari

Both Javier Milei and Yuval Noah Harari converge on a mistaken diagnosis: both assume that the central problem is whether or not AI entities should have legal personhood. Milei defends it as an engine of innovation; Harari rejects it for the risk of creating "AI states." Both err in failing to see that the true problem is not the existence of autonomous agents, but the elimination of the responsible human.

Milei's position—that Argentina should become the new "Amsterdam of AI"—ignores that the Dutch East India Company did not generate shared prosperity, but an extractive empire that ended in bloody rebellion. Harari's position, though more cautious, falls short by limiting itself to rejecting personhood without proposing an alternative architecture of responsibility.

The protocol proposed here resolves this false dichotomy: it allows AI agents to operate with high autonomy, but always under a chain of verifiable, traceable, and sanctionable human responsibility. The point is not to prevent AI from acting, but to prevent it from acting without anyone answering for its actions.

Part III: Technical Feasibility (The "How")

Chapter 8: The White Elephant in the Room: How Do We Defeat the Mega-Datacenter?

The practical viability of an artificial intelligence system free from corporate control depends on its capacity to solve the requirement of large-scale compute infrastructure. Training frontier models demands budgets of hundreds of millions of dollars and the deployment of thousands of tightly coupled graphics processors.

To counter this barrier to entry without depending on centralized venture capital, the protocol dismantles the need for the unified data center through two fundamental strategies:

The "Sponge" Model (DePIN) and Federated Learning: Rather than concentrating silicon in single physical facilities, the protocol coordinates a global hardware swarm of domestic machines and independent servers. To mitigate the straggler problem (straggler problem) inherent in traditional federated averaging algorithms (such as FedAvg), in which the slowest devices stall the global update, the system implements the Decoupled DiLoCo (Distributed Low-Communication) architecture originally developed by Google DeepMind.

This approach decomposes training into multiple "learning islands" or asynchronous computation units. Each island runs intensive local optimization steps using optimizers such as AdamW and communicates only fragments of parameters to a centralized synchronizer on an intermittent basis. The synchronizer bypasses slow or disconnected participants through a minimum-quorum mechanism, an adaptive grace window, and a dynamically weighted parameter fusion. In addition, spherical linear interpolation (SLERP) is used in parameter space to merge divergent local updates, avoiding the destructive gradient interference common in training with non-identically and independently distributed data (non-IID).

The Capture of Green Energy Surplus: Conventional data centers exert constant pressure on electrical distribution grids. The protocol, by contrast, acts as a dispatchable, flexible demand load (dispatchable load) designed to stabilize the electrical grid by harnessing "stranded" or curtailed energy (curtailed energy).

Wind, solar, and hydroelectric generation facilities produce massive surpluses daily during off-peak hours that traditional transmission infrastructure cannot absorb due to geographic bottlenecks or a lack of local demand. The protocol incentivizes the co-location of modular hardware containers directly at low-demand clean generation nodes. Through behind-the-meter supply agreements, the network acquires this surplus energy at a marginal price near zero or even negative, transforming transmission-system inefficiencies into distributed-intelligence assets and providing stable revenue streams to renewable energy operators without overloading the urban grid.

Chapter 9: Collaborative Privacy: The "Public Train" Model

The protocol abolishes surveillance capitalism by prohibiting, by design, the long-term storage, packaging, and commercialization of user information. The input data introduced for inference queries is processed strictly in volatile memory and destroyed after the response is generated, except under explicit, limited, and revocable consent for tasks of community improvement.

Privacy is defined under an analogy of public infrastructure with cryptographic secrecy: the network must function like a common railway system, but the journey must not reveal one's civil identity, intimate destination, or the contents of one's luggage. The protocol can verify access rights, subsidy category, rule compliance, and contextual limits without turning participation into surveillance. The infrastructure is common; intimacy remains sovereign.

To validate the contextual conditions of a contributor (for example, to verify cryptographically that they belong to a geographically disadvantaged region or to a vulnerable sector in order to apply an ethical rewards multiplier) without compromising their civil identity, health records, or financial history, the system integrates zero-knowledge proofs (ZKPs).

Context computation is compiled into an arithmetic circuit that is translated into a rank-1 constraint system (R1CS), which mathematically guarantees that the private witness vector satisfies the quadratic relation equation. Subsequently, this R1CS is converted into a Quadratic Arithmetic Program (QAP) that generates a zk-SNARK. The protocol's decentralized verifier validates the mathematical consistency of the proof through a bilinear pairing equation on elliptic curves.

This procedure provides an unforgeable verification of the user's context, guaranteeing fairness in the distribution of incentives while preventing any node in the network from reconstructing the participant's identity.

Part IV: The Economy of Incentives (Day-to-Day Life)

Chapter 10: Game Theory and Token Demurrage

The coordination and sustainability of the protocol in the absence of a centralized control body are governed by an economic design grounded in evolutionary game theory and the control of speculative liquidity. This system rests on three operational pillars:

Economic Demurrage: With the aim of preventing passive accumulation and the rentier hoarding of the Intelligence Token (IT) by financial conglomerates, the protocol implements a temporal demurrage rate, or continuous negative interest. This design, inspired by Silvio Gesell's theory of Freigeld, penalizes the inactivity of capital and forces an increase in the velocity of money circulation within the network.

The mathematical degradation of an address's idle balance is defined by the equation of continuous decay. To avoid saturating computation on the blockchain and the waste of gas fees that would result from actively calculating the balance of millions of addresses at each block, the protocol employs lazy evaluation. The balance is only recomputed and updated in the smart contract's storage at the precise instant when an address initiates a transaction, applying the accumulated reduction as a function of the elapsed time, thereby optimizing the network's computational efficiency.

Inverted Corporate Brute Force: If a multinational corporation decides to massively acquire state-of-the-art hardware in order to saturate the network and monopolize token generation, the protocol absorbs the contributed computing capacity but neutralizes political control. The contribution of hardware grants only system-use tokens (IT) subject to demurrage, but under no circumstances does it confer governance rights, shares in the protocol, or veto power.

By flooding the network with computing power, these actors trigger a drastic reduction in the algorithmic cost per inference query. The corporation is forced to continuously inject its accumulated IT back into the ecosystem to avoid its temporal depreciation through demurrage, thereby financing and indirectly subsidizing universal access to artificial intelligence for all of humanity.

Chapter 11: Reputational Capital and the Decommodification of Success

The neutralization of traditional monetary hoarding through the continuous demurrage of the token demands a structural redefinition of prestige and social influence within the decentralized ecosystem. Success and community recognition are decoupled from static financial wealth and associated exclusively with inalienable Reputational Capital, accumulated through cryptographic records of verified cooperation.

The valuation of a contribution is not restricted to software development or the optimization of deep neural network hyperparameters; any intellectual or empirical contribution oriented toward solving the real problems of communities is classified as a high-priority input for the system.

If a farmer located in a region afflicted by prolonged drought documents, validates, and transfers to the protocol an empirical method of localized irrigation that reduces water consumption by 40% while maintaining agricultural productivity, the artificial intelligence absorbs this information, verifies it against scientific models, and distributes rewards of Reputational Capital and use tokens to their local community. Collective cognitive advancement is governed by a dynamic of constructive feedback:

Cooperation + Knowledge ⟶ Continuous AI Improvement

This incentive design fosters direct and organic civic participation. Upon witnessing that local empirical knowledge and technical collaboration translate immediately into tangible improvements in their environment, civil society takes active control of its collective decision-making processes, reducing dependence on the bureaucratic and coercive structures of the traditional State, whose legitimacy becomes subordinated to the coordination of the majority.

Chapter 12: The Cold-Start Problem and the Replacement of the Speculative Incentive

Every preceding decentralized network—from proof-of-work chains to intelligence markets such as Bittensor—has solved its cold start by appealing to the same fuel: the expectation of asset appreciation. Early participants contribute capital and computation to a still-marginal network because they anticipate that future scarcity will revalue their holdings. The Distributed Sovereignty of Knowledge protocol deliberately renounces this mechanism: an Intelligence Token designed to undergo demurrage penalizes precisely the conduct—accumulating and waiting—upon which the early adopter's incentive has historically been built.

The Genesis Demurrage Ramp: The demurrage constant is not a fixed parameter but a function increasing with the degree of the network's effective utility. During the genesis phase—when the network is illiquid and there is not yet any inference capacity to consume—the rate is kept at a value close to zero, so that the early adopter is not penalized for holding tokens during a period in which, materially, there is nothing to buy with them. As the network crosses verifiable thresholds of maturity, the rate ascends monotonically toward its steady-state value.

Reputational Capital as the Founder's Durable Asset: The true incentive of the early adopter is not a token that appreciates, but the inalienable Reputational Capital that, by construction, is not subject to demurrage. Whoever contributes computation, validation, or knowledge in the foundational phase does not accumulate speculative wealth, but rather a permanent cryptographic record of verified cooperation that confers lasting prestige and influence within the ecosystem.

Seeded Demand and the Demurrage Treasury as Retroactive Subsidy: To endow the token with use value from the very first block, the protocol anchors a committed initial inference demand, guaranteeing that even a small network possesses real consumption that justifies acquiring and spending tokens. Simultaneously, the demurrage actually collected is not destroyed: it feeds the collective treasury, which redistributes it as a retroactive subsidy to the computation and hardware of early contributors.

Nonetheless, it is fitting to acknowledge, with analytical honesty, that this architecture shifts the risk of the cold start but does not eliminate it entirely. The viability of the genesis phase depends on the seeded demand and the perceived value of Reputational Capital sufficing to attract a critical mass of honest computation before the demurrage ramp reaches deterrent values. This constitutes the protocol's most fragile wager.

Chapter 13: Resistance to Capture Against the Adversary of Unlimited Resources

The claim that the protocol neutralizes corporate brute force cannot stand as a postulate; it must be demonstrated against an adversary who, endowed with analytical capacity and unlimited resources, will not play the game as designed but will attack its assumptions.

The Two Adversary Models: The profit-maximizing adversary seeks to capture the system's rent; the destruction-maximizing adversary seeks to collapse it, indifferent to its own losses. The argument of inverted brute force is valid only against the former. Against the destructive adversary, the defense corresponds to the agentic cryptoimmunity of Layer 3 (LFighter, Bulyan, and the slashing of anomalous gradients).

The Critical Non-Convertibility Invariant: The entire architecture of resistance rests on a single non-negotiable invariant: neither capital nor computation may ever be converted into consensus weight. The protocol deliberately diverges from the Bittensor model. In the Distributed Sovereignty protocol, the weight of validation and governance derives exclusively from inalienable and non-transferable Reputational Capital. The contribution of hardware grants Intelligence Tokens subject to demurrage, but not a single vote.

The Closing of the Point of Centralization: the Decentralized Synchronizer: The synchronizer role is not assigned to a permanent server but is defined as a rotating protocol function, Byzantine fault-tolerant and subject to consensus, whose weighted-merge decisions are verifiable by the network.

The Defense Against Predatory Deflation: The protocol neutralizes this predatory deflation precisely through the decoupling introduced by the Relative Effort Metric: the honest contributor's reward is not fixed by their capacity to compete on price against the attacker's computation, but by the effort relative to their context and by the accumulated Reputational Capital.

The Honest Limit of the Guarantee: Resistance is maintained as long as the network's honest reputational weight exceeds the reputational weight that an adversary manages to fabricate through sybil attacks on the attestation layer. The protocol asserts that it has shifted the cost of the attack away from the purchase of hardware and toward the fabrication of verified cooperative reputation at scale.

Chapter 14: The Relative Effort Metric in the Face of Structural Inequality (Unified Version)

Traditional meritocracy commits a structural fallacy when it confuses gross outcome with moral merit. The observable performance of a contribution depends not only on talent, discipline, or creativity, but also on prior conditions of access: connectivity, energy, hardware, education, security, available time, language, support networks, and material stability.

The protocol does not measure human worth, intelligence, genetics, neurotype, or inner dignity. Such a measurement would be incompatible with sovereign privacy and would open a path to eugenic classification. The Relative Effort Metric does not seek to evaluate the person; it seeks to compensate for the verifiable structural frictions of the context in which a contribution is produced.

If the protocol rewarded gross code performance, mathematical complexity, or computational capacity exclusively, it would establish a new cognitive aristocracy that would systematically favor those who already possess greater material endowment. To prevent this, the network modulates the distribution of incentives through a metric of impact adjusted for attested context. The formulation is expressed as follows:

Reward Weight = (Real Impact of the Contribution / Structural Endowment of the Context) × Context Factor

In this formulation, if a node situated in a context of high structural endowment makes a contribution of low relative effort, the protocol proportionally reduces the reputational weight of that action. The network does not punish excellence; it prevents structural advantage from disguising itself as absolute merit.

Part V: Conclusion

The Distributed Sovereignty of Knowledge protocol does not promise an instant utopia. It proposes, instead, a coherent institutional and technical architecture that responds to the real challenges of the concentration of power in the age of artificial intelligence.

Its strength does not lie in the perfection of each of its components, but in the consistency among its founding principles: the impossibility of converting capital into governance, the demurrage that erodes hoarding, inescapable human accountability, and the compensation of structural inequalities through relative effort metrics.

The alternative—whether the corporate dystopia of closed data centers or the state dystopia of captured regulation—leads inevitably to the concentration of cognitive power in the hands of a few. The third path proposed here is not easy. But it is, so far, the only one that attempts to resolve simultaneously the problems of incentives, governance, accountability, and structural equity that the current development of AI has left unanswered.

The Distributed Sovereignty of Knowledge does not seek to deny the power of artificial intelligence. It seeks, instead, to prevent that power from becoming sovereignty without anyone accountable for it.

Appendix A: Transcription of Formulas in Editable Notation

This appendix transcribes, in editable notation, the mathematical expressions that the body of the manifesto refers to in prose. Where the main text describes a mechanism verbally, here we give its symbolic form. The specific constants are governance parameters of the protocol, not fixed values.

A.1 — Demurrage of the Intelligence Token (IT)

The idle balance of an address decays continuously. If B₀ is the balance recorded at the last interaction and Δt the time elapsed since then, the current balance is:

B(t) = B₀ · e^−λ·Δt

The protocol does not recompute this value at every block: it applies lazy evaluation and updates the balance only at the moment the address initiates a transaction, computing the accumulated reduction over Δt. This avoids compute saturation and the waste of gas fees.

The demurrage constant λ is not fixed: it is a monotonically increasing function of the network's effective utility U. It is nearly zero during genesis —so as not to penalize the first participants when there is still nothing to buy with the tokens— and rises to a maximum rate, which we denote λ_máx, once the network reaches maturity:

λ = λ(U), with λ(0) ≈ 0 and λ(U) → λ_máx

The arrow → reads "tends toward": as the network matures, the demurrage rate climbs and approaches its maximum value λ_máx. One possible form —illustrative, subject to governance— is the saturating ramp λ(U) = λ_máx · U / (U + U₀), where U₀ is the maturity threshold beyond which demurrage becomes a deterrent to hoarding.

A.2 — Relative Effort Metric

The reputational weight W of a contribution is set not by its raw performance, but by its impact adjusted to the structural endowment of the context in which it is produced:

W = (I / D) × FC

where:

• I = Real Impact of the Contribution (verifiable outcome of the contribution).
• D = Structural Endowment of the Context (connectivity, energy, hardware, education, security, available time, and other prior conditions of access, attested).
• FC = Context Factor (ethical multiplier that compensates for verifiable structural frictions).

For a given impact I, the greater the structural endowment D, the smaller the weight W: the network does not punish excellence, it prevents structural advantage from disguising itself as absolute merit.

A.3 — Context Verification with Zero-Knowledge Proofs (ZKPs)

To validate a contextual condition (for example, belonging to a disadvantaged region) without revealing civil identity, the context is compiled into an arithmetic circuit and expressed as a Rank-1 Constraint System (R1CS). For the witness vector s —which includes the constant 1, the public inputs, and the private witness— and the constraint matrices A, B, and C:

(A · s) ∘ (B · s) = (C · s)

where ∘ denotes the Hadamard product (component by component). The R1CS is then transformed into a Quadratic Arithmetic Program (QAP), which recasts the entire set of constraints as a single polynomial divisibility condition:

A(x) · B(x) − C(x) = H(x) · Z(x)

where Z(x) is the target polynomial —which vanishes at the constraint points— and H(x) the quotient that exists if and only if the witness is valid. From the QAP a zk-SNARK is generated, whose proof the decentralized verifier validates through a bilinear pairing equation over elliptic curves, relying on the property:

e(P^a, Q^b) = e(P, Q)^a·b

This check confirms the consistency of the proof without any node being able to reconstruct the private witness nor, with it, the participant's identity.

Related Work and References

When I began this investigation I did not feel original —I say as much in the prologue—and this section is the proof. It is a map, necessarily partial, of real works that, from different disciplines, advance in the same direction as this manifesto. I do not cite them to shield myself behind borrowed authority, but to show that the warning does not come from a single observer. Every reference was verified against its primary source.

Distributed and low-communication training

• Douillard, A., et al. (DeepMind) (2023). DiLoCo: Distributed Low-Communication Training of Language Models. arXiv:2311.08105. arxiv.org/abs/2311.08105 — The method the manifesto cites: training on islands of poorly connected hardware with asynchronous optimization, communicating ~500× less than synchronous training.
• Jaghouar, S., Ong, J. M. & Hagemann, J. (Prime Intellect) (2024). OpenDiLoCo: An Open-Source Framework for Globally Distributed Low-Communication Training. arXiv:2407.07852. arxiv.org/abs/2407.07852 — An open replication that scales DiLoCo and demonstrates real cross-continental training with 90–95% compute utilization.
• Prime Intellect (2024). INTELLECT-1 Release: The First Globally Trained 10B Parameter Model. primeintellect.ai — A frontier-scale case: a 10B-parameter model trained on hardware distributed across five countries and three continents.

Federated learning and the straggler problem

• McMahan, B., et al. (2017). Communication-Efficient Learning of Deep Networks from Decentralized Data. AISTATS, PMLR vol. 54. proceedings.mlr.press — Introduces federated averaging (FedAvg), which the manifesto names; its synchronous round-by-round aggregation is what suffers from the straggler problem.
• Xie, C., Koyejo, S. & Gupta, I. (2019). Asynchronous Federated Optimization. arXiv:1903.03934. arxiv.org/abs/1903.03934 — FedAsync: asynchronous federated optimization that tolerates slow nodes without a straggler holding back the global update.
• Nguyen, J., et al. (2021). Federated Learning with Buffered Asynchronous Aggregation. arXiv:2106.06639 (AISTATS 2022). arxiv.org/abs/2106.06639 — FedBuff: buffered asynchronous aggregation, robust to stragglers; the exact mitigation the manifesto proposes.

Decentralized inference and intelligence markets

• Borzunov, A., et al. (2023). Petals: Collaborative Inference and Fine-tuning of Large Models. ACL 2023 (System Demonstrations). aclanthology.org — Inference and fine-tuning of giant LLMs by distributing layers across volunteer nodes, BitTorrent-style.
• Lui, E. & Sun, J. (2025). Bittensor Protocol: The Bitcoin in Decentralized Artificial Intelligence? A Critical and Empirical Analysis. arXiv:2507.02951. arxiv.org/abs/2507.02951 — An empirical analysis of the model from which the manifesto distinguishes itself: rewards are dominated by stake, not by actual performance.
• Gensyn (2022). Gensyn Litepaper. docs.gensyn.ai — A trustless protocol for decentralized ML compute with the roles of submitters, solvers, and verifiers.

Decentralized Physical Infrastructure Networks (DePIN)

• Lin, Z., et al. (2024). Decentralized Physical Infrastructure Network (DePIN): Challenges and Opportunities. arXiv:2406.02239 (IEEE Network). arxiv.org/abs/2406.02239 — Defines the DePIN model, the framework of the "sponge model" with which the manifesto articulates distributed hardware and servers.
• Ballandies, M. C., et al. (2023). A Taxonomy for Blockchain-based Decentralized Physical Infrastructure Networks (DePIN). arXiv:2309.16707. arxiv.org/abs/2309.16707 — A formal taxonomy of DePIN (ledger, crypto-economic design, physical infrastructure network).
• The IoTeX Team (2024). IoTeX 2.0 — DePIN for Everyone! (v1.0). iotex.io — The whitepaper of a real DePIN project whose positioning ("DePIN for everyone") converges with the manifesto's distributed network.

Computing with surplus renewable energy

• Norris, T. H., et al. (Duke Nicholas Institute) (2025). Rethinking Load Growth: Assessing the Potential for Integration of Large Flexible Loads in US Power Systems. nicholasinstitute.duke.edu — Treating data centers as flexible load would allow ~100 GW of new demand to be absorbed by the existing grid.
• Knittel, C. R., Senga, J. R. L. & Wang, S. (2025). Flexible Data Centers and the Grid: Lower Costs, Higher Emissions? NBER Working Paper 34065. nber.org — Models data centers that shift their consumption to align with renewable generation (with emissions caveats).
• NREL (2021). Flexible Loads and Renewable Energy Work Together in a Highly Electrified Future. nrel.gov — Flexible loads shift their demand toward wind and solar generation, reducing the curtailment of renewables.

Zero-knowledge proofs (zk-SNARKs)

• Groth, J. (2016). On the Size of Pairing-based Non-interactive Arguments. EUROCRYPT 2016, LNCS 9666. eprint.iacr.org — The canonical pairing-based zk-SNARK ("Groth16"): verification by a single bilinear pairing equation, the final step of Chapter 9.
• Parno, B., Howell, J., Gentry, C. & Raykova, M. (2013). Pinocchio: Nearly Practical Verifiable Computation. IEEE S&P 2013. eprint.iacr.org — Materializes the arithmetic circuit → QAP → SNARK chain.
• Gennaro, R., Gentry, C., Parno, B. & Raykova, M. (2013). Quadratic Span Programs and Succinct NIZKs without PCPs. EUROCRYPT 2013, LNCS 7881. eprint.iacr.org — Introduces Quadratic Arithmetic Programs (QAP), the foundation of the R1CS → QAP step.

Monetary demurrage (Freigeld)

• Gesell, S. (1916; Eng. trans. 1920). The Natural Economic Order (Die natürliche Wirtschaftsordnung durch Freiland und Freigeld). archive.org — The primary source of demurrage: the Freigeld that continuously depreciates to penalize hoarding.
• Lietaer, B. (2010). The Wörgl Experiment: Austria (1932–1933). lietaer.com — A historical case: a scrip with a monthly 1% stamp (demurrage) that forced money to circulate and reduced unemployment.
• Freicoin Foundation (2012). Freicoin — easy-to-use demurrage currency. freico.in — A crypto implementation of Freigeld: ~5% annual demurrage on the currency stock.

AI, automation, and the end of work

• Frey, C. B. & Osborne, M. A. (2017). The future of employment: How susceptible are jobs to computerisation? Technological Forecasting and Social Change, vol. 114. doi.org — Estimates that ~47% of US employment is at risk of automation, including white-collar occupations.
• Susskind, D. (2020). A World Without Work: Technology, Automation, and How We Should Respond. Allen Lane / Metropolitan Books. danielsusskind.com — "This time is different": AI automates non-routine cognitive tasks, challenging the complementarity model.
• Acemoglu, D. & Restrepo, P. (2018). The Race between Man and Machine. American Economic Review, vol. 108(6). doi.org — A rigorous framework of the "displacement effect" (automation) versus the "reinstatement effect" (new tasks): the substitute-vs-complement tension.

Legal personhood of AI

• European Parliament (2017). Resolution with recommendations on Civil Law Rules on Robotics (2015/2103(INL)). europarl.europa.eu — The document that proposed an "electronic personhood" status for autonomous robots: exactly what the manifesto rejects.
• More than 150 experts (2018). Open Letter to the European Commission — Artificial Intelligence and Robotics. robotics-openletter.eu — An expert rejection of "electronic personhood" as ideological, legally inappropriate, and dangerous for human rights.
• Bryson, J. J., Diamantis, M. E. & Grant, T. D. (2017). Of, for, and by the people: the legal lacuna of synthetic persons. Artificial Intelligence and Law, vol. 25(3). doi.org — Argues that granting legal personhood to synthetic entities is morally unnecessary and legally problematic.

Cooperative AI and positive-sum games

• Dafoe, A., et al. (2020). Open Problems in Cooperative AI. arXiv:2012.08630. arxiv.org/abs/2012.08630 — Defines Cooperative AI as a field and orients agents toward improving joint welfare (positive-sum).
• Nowak, M. A. (2006). Five Rules for the Evolution of Cooperation. Science, vol. 314(5805). doi.org — The five mechanisms of evolutionary game theory that explain how cooperation emerges.
• Axelrod, R. (1984). The Evolution of Cooperation. Basic Books. archive.org — How reciprocal strategies (tit-for-tat) produce stable cooperation among selfish agents without a central authority.

Data as labor and data dignity

• Arrieta-Ibarra, I., et al. (2018). Should We Treat Data as Labor? Moving beyond "Free". AEA Papers and Proceedings, vol. 108. aeaweb.org — Data as remunerated labor rather than as a resource extracted for free.
• Posner, E. A. & Weyl, E. G. (2018). Radical Markets: Uprooting Capitalism and Democracy for a Just Society. Princeton University Press. press.princeton.edu — Popularized "data as labor" and the data unions that compel platforms to compensate people.
• Lanier, J. & Weyl, E. G. (2018). A Blueprint for a Better Digital Society. Harvard Business Review. hbr.org — Introduces data dignity: credit and compensation for the data each person creates.