← Back

Research

AI under PROUT: how the system should work

A working model: ownership, labour, governance, knowledge, scale, and verification

[Draft] · not part of the knowledge graph

Abstract

This work answers one question: how should an artificial intelligence system be organised if it is built according to PROUT. The answer is not a market survey or a philosophy of consciousness but a working model of six rules. Ownership: small systems private or cooperative, medium cooperative, large public. Labour: automation shortens working time, not people. Governance: a moral council, technical audit, community voice, and a human in the loop of irreversible decisions. Knowledge: an answer must lead to a source and distinguish quotation, synthesis, and hypothesis. Scale: data and applications at block level, standards at inter-block level, chips, climate, and weapons at world level. Metrics: minimum requirements, purchasing capacity, free time, ownership concentration, energy, language, and verifiability. The rules yield an operating model with roles, three phases of implementation, and 12 tests. The evidence base is 18 corpus texts and 16 external works; every claim is marked as quotation, synthesis, or hypothesis. The work does not predict what AI will become. It fixes how AI should work so that it serves society rather than rent.

1. Question and answer

One question: how should an AI system work if it is built according to PROUT? Not “what is intelligence,” not “when will superintelligence arrive,” not “who will win the market.” We are interested in the arrangement: who owns models, compute, and data; what happens to labour; who makes decisions; how answers are verified; where the competence of a block ends and the world begins.

The short answer: AI should be a public productive force, not a private rent. This is not a slogan but a construction. It consists of six rules: goal, ownership, labour, governance, knowledge, metrics. Each rule can be tested on a concrete system, and each has a cost: remove it, and the gain goes to the owner while the costs go to society.

The work is deliberately narrow. It contains no history of AI, no model survey, no philosophy of consciousness, and no argument about whether a machine will ever be a subject. There is one premise without which the model does not work: AI is an instrument; responsibility belongs to people. It rests on the description of the body as a machine used by the mind, and on the remark that even machines with minds will be weaker than their creators.1213 It does not follow that AI is safe; it follows only that subjecthood cannot be used as an excuse.

The method is simple. First, state the PROUT rule. Then translate it into a question for the system: who owns, who pays, who controls, what counts as success. Then formulate a test that can be applied without a laboratory. If the test fails, the system does not conform to PROUT regardless of accuracy and convenience.

What the reader gets: an ownership map, a labour scheme, four locks of governance, knowledge requirements, a distribution of scales, a set of metrics, a role list, three phases of implementation, and 12 tests. That is enough to discuss a concrete project rather than “AI in general.”

2. Rule one: ownership

PROUT distinguishes three levels of industry: small business is left to individuals, large industry to society, and the middle to cooperatives. “(1) small businesses should be left to individuals; (2) big industries should be owned by the immediate government; and (3) the industries in between the big and small industries should be run on a cooperative basis.”5 For AI this gives a ready map.

Small — applications, local models, tools for a community. Private initiative is permissible here: low entry cost, high local specificity. Medium — sectoral models for health, agriculture, education, and language. Cooperative ownership is required: the model uses shared data and serves common infrastructure. Large — compute, energy, foundation models, and the data on which they are trained. This is a key industry; a private monopoly turns infrastructure into rent.

Why does compute fall into “large”? Because it concentrates. Epoch AI estimates that five hyperscalers held 71% of global AI compute in the fourth quarter of 2025; NVIDIA controls roughly 80–90% of the accelerator market.21 This is not a market with a low entry threshold; it is infrastructure on which everyone else depends. In PROUT terms it is a key industry.

Data are a separate case. They cannot simply be nationalised: they are produced by people and often contain the personal and the cultural. Common patrimony does not mean free capture. It requires consent, provenance, compensation, the right to deletion, and differentiated access. The practical criterion is simple: access to data is delimited by consent and provenance, not by the fact of owning a copy. This is not a universal solution, but it is an example of the fact that “common” and “uncontrolled” are not synonyms.

Test of the rule: name the owner of each layer. If compute and the foundation model belong to one private company, and the data were collected without consent, the system fails the rule regardless of its licence.

3. Rule two: labour

PROUT distinguishes mechanisation under capitalism from mechanisation in a collective economy. Under capitalism, “when the productive capacity of machinery is doubled, the required human labour is decreased by half, so the capitalists retrench large numbers of workers.”2 In a collective economy, “mechanization will lead to less labour and more prosperity. With the double increase in the productivity of machines, the working hours of labourers will be reduced by half.”2 The same technology, different institutions, different results.

This is not utopia but a question of ownership. A capitalist is not obliged to preserve employment; the goal is profit. A collective can shorten hours because its goal is welfare. The question “will AI create unemployment” is therefore badly posed. The right question is: who appropriates the gain from productivity, and who bears the costs of transition.

The scale of the transition is known. The IMF estimates that almost 40% of global employment is exposed to AI, and about 60% in advanced economies; roughly half of exposed jobs may benefit from complementarity, half from automation.19 The ILO is more cautious: automation potential is about 2.3% of global employment, augmentation potential about 13%; in high-income countries automation reaches 5.5%.20 These are exposure estimates, not verdicts. They say that the transition will be uneven and that “just wait” is not a policy.

The rule requires concrete mechanisms, not a promise. Guaranteed minimum; rising purchasing capacity; shorter working time; retraining and public works; cooperatives and public funds; a transition fund financed by those who appropriate the gain. Without these mechanisms, “shorter hours” remains a slogan for those who still have a job.

Test of the rule: show what happens to the people the system replaces. If the answer is “they will find another job” without a timeline, training, and income, the rule is not met.

4. Rule three: governance

The most direct argument in the corpus concerns control. “When a machine works, there is an intellectual entity controlling its operations. If a machine is permitted to work without any control, it is bound to cause catastrophe.”14 This is not about AI, but it applies to it more precisely than many contemporary texts. A machine without control is dangerous not because it is evil, but because it has no judgement of its own about the goal.

The second argument concerns oversight. “If science remains completely in the hands of static people, the outcome will be disappointing.”3 The requirement here is technical, not metaphysical: the development of technology must be under external control, not in the hands of one narrow circle. In current practice this means independent audit, public criteria, and a right of veto, as in the EU AI Act and the NIST AI RMF.3334

AI governance is not reduced to a single body; it is distributed across four locks, each of which can refuse.

A moral council checks ends and admissibility. A technical audit checks behaviour and safety: evaluations, red teams, bias, injection, incidents. Community voice and veto checks consent and local consequences. A human operator holds the final switch. No single lock is sufficient: a council without technique is a declaration, an audit without morality is optimisation of a bad goal, a community without resources is a focus group, an operator without procedures is a scapegoat.

Red lines are not a matter of “regulation”: autonomous weapons, mass surveillance, social scoring, manipulative companions, systems that make irreversible medical or judicial decisions without a human. They do not need to be “improved”; they should not be deployed.

Test of the rule: name who can stop the system and by what procedure. If the answer is “nobody, it is autonomous,” the rule is not met.

5. Rule four: knowledge

AI sells not only computation but authority. People consult a model as they once consulted a book or a specialist. The corpus warns: “Perception, inference and authority are the valid sources of knowledge … oftentimes they tend to give more weight to authority than to perception.”14 A machine that speaks confidently without a source amplifies exactly this error.

The rule is simple: an answer must be verifiable. Not “smart,” not “natural,” but such that the reader can reach the ground. Six requirements follow: a citation leads to the primary text; the answer distinguishes “the corpus says,” “it follows from what is said,” and “this is a hypothesis”; uncertainty is shown; refusal is permitted; the system logs what the model sees; local language and local data are not replaced by translation.

A verifiable architecture works like this: one retrieval path, access control that defaults to denial (no access, no result), citations with document and fragment identifiers, and the model sees only what is logged. This does not prove truth, but it makes the answer checkable: the reader can open the source and see where a statement came from. This is the opposite of an oracle.

Test of the rule: take a random answer and try to reach the source. If it takes more than a minute, or the source does not support the claim, the rule is not met.

6. Rule five: scale

PROUT thinks locally: the economic unit is a block, small enough to understand the problems and large enough to provide employment and purchasing capacity. “Block-Level Planning” sets the goals: guarantee the minimum, eliminate unemployment, raise purchasing capacity, make the unit self-sufficient.7 But some problems cross block boundaries: communications, key industries, energy. For these there is “inter-block planning.”7

AI belongs exactly here. Data, compute, chips, energy, and standards are global, while consequences are local. What is needed is neither one world model nor a hundred isolated ones, but a multi-level system: local data and applications, inter-block standards and shared models, world red lines.

The principle is subsidiarity: a decision is made at the lowest level competent to make it. Local data, language, applications, and working conditions at block level. Standards, shared models, migration of skills at inter-block level. Chip supply chains, climate, weapons, treaties at world level. This is not a hierarchy of power but a distribution of competence.

Test of the rule: name the level at which the decision is made. If everything is decided in one centre, the system fails; if everything breaks into local islands, it also fails.

7. Rule six: metrics

If a system cannot be measured, it cannot be verified. But the metric of PROUT is not model accuracy or profit. It is pramā: dynamic balance among physical, psychic, and spiritual development.15 For AI this becomes a set of indicators counted at block level and aggregated at inter-block level.

Test of the rule: show a quarterly report. If it contains only accuracy and revenue, the system fails the rule even if the model is good.

8. Operating model: who does what

Rules become a system only when each actor has a role, a right, and a duty. Below is the minimum model. It does not describe a particular jurisdiction; it sets the separation without which the rules remain declarations.

Money in this model does not flow from the user to the model owner but from the gain to a common fund. The fund finances compute and foundation models, cooperatives build sectoral applications, and the community receives the minimum and time. This is not a “tax on AI” in general but the return of common value: the model was trained on common data and runs on common infrastructure.

Test of the model: for every decision there must be one role that can say no and one that must disclose the ground. If only the owner can say no, this is not PROUT but private power with public rhetoric.

9. Implementation: three phases

The model is not introduced by decree. It is introduced through pilots that can be stopped. The three phases below are a minimum schedule; the timing depends on the block, but the sequence matters: rights and audit first, scale later.

At every stage the fifth principle applies: the method of utilisation changes with time, place, and person.1 The rules are revised; the goals — minimum, distribution, oversight — remain.

Test of implementation: a pilot has a stop date, a transition budget, and an independent report. If a pilot cannot be stopped, it is not a pilot but a capture.

10. Twelve tests

The tests can be applied to any system: from a local chatbot to a state platform. They are not quotations; they are a synthesis, and each states the ground on which it rests. Their value is that they make the argument concrete: who owns, who pays, who controls, who can refuse.

1. Instrument, not subject. The system is not granted legal personality, authorship, or moral responsibility. Responsibility remains with people and institutions. Ground: the ontology of machine and witness.12

2. Common patrimony. Foundation models, data, and compute created with collective resources belong to society; private appropriation is limited. Ground: common patrimony and the first principle.1

3. Use, not rent. Deployment is judged by needs met and time freed, not by profit. Ground: economic democracy.6

4. Minimum and purchasing capacity. No deployment may reduce the guaranteed minimum or purchasing capacity. Ground: “The Cosmic Brotherhood” and “Economic Democracy.”16

5. Three tiers of ownership. Small — private or cooperative; medium — cooperative; large — public or under oversight. Ground: three-tier industry.5

6. Local deployment. Decisions at block level; global infrastructure by inter-block plan. Ground: block-level planning.7

7. Moral oversight. Independent councils with public criteria, rotation, and veto. Ground: external oversight and the right of veto (EU AI Act; NIST AI RMF).3334

8. Human in the loop. Irreversible decisions — weapons, medicine, courts, infrastructure — remain with a human. Ground: a machine without control is dangerous.14

9. Transparent grounds. Sources, uncertainty, and limits are visible; no oracle. Ground: perception, inference, and authority.14

10. Neohumanist metric. Evaluation includes effects on people, animals, plants, and ecosystems, not only accuracy or GDP. Ground: Neohumanism.16

11. Linguistic autonomy. Models and data in local languages; the community’s right to its memory. Ground: synthesis of decentralisation and common patrimony.6

12. Pramā. Balance between automation and employment, centralisation and autonomy, innovation and stability; revision by time, place, and person. Ground: pramā and the fifth principle.15

11. Limitations

The corpus has no doctrine of AI. It has concepts that we transplant; the reader may reject the transplant. The work does not present synthesis as doctrine and tries to mark where the text ends and construction begins.

The metaphysical argument is not a safety proof. The phrase “machines with minds will be weaker than their creators” may reassure, but it does not explain why a system will not cause harm. A powerful machine need not be conscious to destroy infrastructure, manipulate people, or deepen inequality.

Scale is not solved locally. PROUT describes a local economy, while AI is global infrastructure. Block planning does not solve the problem of chips, lithography, and rare earths; inter-block planning is a useful frame, but not a guarantee of justice among blocks.

Data are not solved by one principle. Common patrimony does not resolve the conflict between common knowledge and personal privacy. It requires procedures of consent, compensation, and deletion that the corpus does not contain. Access control is an example, not a solution.

The empirical figures are dated. Numbers on employment and compute are orientation, not forecast; the literature on AI changes faster than this work. The reader should check currency rather than take numbers as truth.

12. Conclusion

AI under PROUT is not a “smart machine” to which society can be entrusted. It is a public productive force organised by six rules: the goal is the minimum and free time; ownership has three tiers; labour means hours, not people; governance has four locks; knowledge requires source and status; metrics follow pramā. The rules yield roles, phases, and 12 tests.

The alternative is known: a private oracle trained on common knowledge and selling access; automation that privatises the gain and socialises the cost; a machine that works without intellectual control because control seems too slow. The corpus warns precisely against this: “If a machine is permitted to work without any control, it is bound to cause catastrophe.”14

A machine may think faster than a human. The question is not speed. The question is who owns the thought, who pays for its costs, and who can say no. PROUT does not give the machine a soul; it reminds us that the soul belongs to the society that decides which machine to serve.

Sources and literature

Links to the shastras open the reader at the relevant chapter. Collections require sign-in. Quotations are abridged and kept in the original; the English text is analysis, not a substitute for the source.

Sources

  1. “The Cosmic Brotherhood,” 5 June 1959. In Prout in a Nutshell, part 5; Idea and Ideology. reader
  2. “Problems of the Day.” In Prout in a Nutshell, part 4. reader
  3. “Synthesis and Analysis,” 27 July 1958. In Prout in a Nutshell, part 4. reader
  4. “Discourses on Prout,” 17–18 October 1959. In Prout in a Nutshell, part 5. reader
  5. “Economic Democracy,” June 1986. In Prout in a Nutshell, part 21. reader
  6. “Block-Level Planning”; “Socio-Economic Groupifications.” In Prout in a Nutshell, part 12. reader
  7. “Capitalism in Three Spheres.” In Prout in a Nutshell, part 13. reader
  8. “Talks on Prout.” In Prout in a Nutshell, part 15. reader
  9. “Minimum Requirements and Maximum Amenities.” In Prout in a Nutshell, part 17/4; Proutist Economics. reader
  10. “The Biological Machine,” 20 July 1990. In Discourses on Kṛṣṇa and the Gītā. reader
  11. Light Comes, “Abhimat III, 3.” reader
  12. Namaḥ Śivāya Śantāya, Shivokti 2. reader
  13. “Pramā – 1 / Dynamic Equilibrium and Equipoise.” In Prout in a Nutshell, part 9; Proutist Economics. reader
  14. The Liberation of Intellect: Neohumanism, discourse 11. reader
  15. A Guide to Human Conduct: Yama and Niyama. reader
  16. “The Transformation of Psychic Pabula into Psycho-spiritual Pabulum.” In Prout in a Nutshell, part 12. reader

Literature

  1. Cazzaniga, M., Jaumotte, F., Li, L., Melina, G., Panton, A. J., Pizzinelli, C., Rockall, E., & Tavares, M. M. (2024). Gen-AI: Artificial Intelligence and the Future of Work. IMF Staff Discussion Note SDN/2024/001. pdf
  2. Gmyrek, P., Berg, J., & Bescond, D. (2023). Generative AI and Jobs: A Global Analysis of Potential Effects on Job Quantity and Quality. ILO Working Paper 96. ilo
  3. Epoch AI (2026). Five Hyperscalers Now Own Over Two-Thirds of Global AI Compute. epoch
  4. Crawford, K. (2021). Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence. Yale University Press.
  5. Zuboff, S. (2019). The Age of Surveillance Capitalism. PublicAffairs.
  6. Gray, M. L., & Suri, S. (2019). Ghost Work: How to Stop Silicon Valley from Building a New Global Underclass. Houghton Mifflin Harcourt.
  7. Acemoglu, D., & Restrepo, P. (2019). Automation and New Tasks: How Technology Displaces and Reinstates Labor. Journal of Economic Perspectives, 33(2), 3–30. doi
  8. Russell, S. (2019). Human Compatible: Artificial Intelligence and the Problem of Control. Viking.
  9. Christian, B. (2020). The Alignment Problem: Machine Learning and Human Values. W. W. Norton.
  10. Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the Dangers of Stochastic Parrots. FAccT ’21, 610–623. doi
  11. Sastry, G., Heim, L., Belfield, H., et al. (2024). Computing Power and the Governance of Artificial Intelligence. arXiv:2402.08797. arxiv
  12. Couldry, N., & Mejias, U. A. (2019). The Costs of Connection: How Data Is Colonizing Human Life and Appropriating It for Capitalism. Stanford University Press.
  13. Whittaker, M. (2021). The Steep Cost of Capture. Interactions, 28(6), 50–55. doi
  14. European Union (2024). Regulation (EU) 2024/1689 (Artificial Intelligence Act). eur-lex
  15. NIST (2023). Artificial Intelligence Risk Management Framework (AI RMF 1.0). doi

18 November 2026