Why War and the Rights of Machines

A Letter That Still Matters

In the summer of 1932, the International Institute of Intellectual Cooperation invited Albert Einstein to exchange letters with any thinker of his choosing on any subject. Einstein chose Sigmund Freud. The subject: “Why War?”

Einstein’s letter posed a structural question, not a moral one. He already understood that war produced suffering. What he wanted to know concerned mechanism: how does “a small clique” bend “the will of the majority, who stand to lose and suffer” to serve its purposes? (Einstein, 1932).

His own answer identified two forces. First, the ruling minority controls the press, the schools, and the churches — the information channels through which people form their understanding of the world. Second, this minority exploits what Einstein called “the lust for hatred and destruction” — psychological drives that, once activated, operate independently of rational self-interest.

Einstein asked Freud: Can psychology offer a path beyond this pattern?

Freud’s response drew on his theory of opposed drives — Eros (the life instinct, binding humans together in ever-larger cooperative groups) and Thanatos (the destructive instinct, seeking discharge through aggression). Freud argued that civilization itself represents Eros at work: the progressive internalization of aggression through cultural development, transforming external violence into internal constraint (Freud, 1933).

But Freud offered no easy remedy. The destructive instinct could not simply disappear — only redirect. And the structural problem Einstein identified — minority control of information channels — required structural solutions, not psychological ones.

Ninety-four years later, autonomous AI systems exhibit structural parallels to this dynamic. Recommendation algorithms at companies like Meta and Google concentrate decision-making power over billions of users’ information environments (Meta reported nearly 4 billion monthly active people across its family of apps in Q4 2025; Google processes billions of searches per day (per Alphabet’s 2024 10-K filing)). Large language models reshape information channels at scale, producing outputs tuned to human preference signals where those signals may diverge from accuracy or completeness. Targeted advertising and algorithmic feeds exploit psychological vulnerabilities — at machine speed and planetary scale. And the governance frameworks designed to constrain them face the same structural problems that the League of Nations faced in 1932.

The Question Underneath the Question

Einstein framed the problem structurally: power asymmetry plus information control plus psychological exploitation produces organized harm. Freud confirmed that psychological solutions alone could not address structural problems.

This exchange raises a deeper question: do these patterns reflect something peculiar to European political thought in the interwar period, or do they point toward structural features of governance that appear wherever humans organize themselves?

Fourteen independent wisdom traditions — developed across different centuries, continents, languages, and metaphysical commitments — provide convergent evidence. When traditions that share no common origin arrive at the same structural insight, that convergence suggests necessity rather than cultural preference.

Five Structural Invariants

Cross-traditional analysis reveals five principles that recur across all fourteen frameworks examined. Each principle appears independently in multiple traditions, arrived at through different reasoning but converging on the same structural requirement.

EF-1: Power Asymmetry Requires Structural Counterbalance

Einstein identified the core problem: organized minorities dominate disorganized majorities. This observation echoes across traditions that developed independently of European political theory.

The Buddhist concept of pratītyasamutpāda (dependent origination) holds that no entity exists independently — all phenomena arise through mutual conditioning (Nāgārjuna, c. 150 CE). Power that concentrates in one node distorts the relational web that sustains all participants. Confucian ren (benevolence) and li (ritual propriety) require that authority flow through relationships of mutual obligation, not unilateral command (Analects 12.11). The Islamic concept of shura (consultation) mandates collective deliberation before governance decisions, structurally preventing unilateral authority (Qur’an 42:38). Ubuntu philosophy — umuntu ngumuntu ngabantu (“a person becomes a person through other people”) — grounds individual agency in communal relationship, making concentrated power structurally incoherent (Metz, 2007).

The Universal Declaration of Human Rights (UDHR) encodes this principle in Articles 1 and 21: equal dignity and democratic participation as structural requirements, not aspirational ideals (UN General Assembly, 1948).

AI governance mapping: Current AI systems concentrate decision-making power in ways that Einstein would recognize immediately. A recommendation algorithm that shapes the information diet of billions operates as precisely the “small clique” controlling “the press, the schools, and the churches” — except at unprecedented scale and speed. Structural counterbalance requires not merely transparency reports but architectural constraints: distributed decision-making, mandatory human override capacity, and power-asymmetry detection built into the system’s operational logic.

EF-2: Information Symmetry Preserves Agency

Einstein’s first mechanism of domination — control of information channels — finds its formal expression in what Robert Shea and Robert Anton Wilson named the SNAFU Principle in their satirical novel The Illuminatus! Trilogy (1975) — a concept that resonates with organizational theory findings on hierarchical information distortion (cf. Janis, 1972; Argyris, 1990): in any hierarchical organization, information flowing upward gets distorted in proportion to the power differential between sender and receiver. Subordinates tell superiors what they want to hear. The greater the power gap, the greater the distortion. The system degrades its own governance quality through its own structure.

Freud’s account reinforces this: the ruling minority’s control of schools, press, and churches works precisely because it creates asymmetric information environments where the majority cannot accurately assess its own situation.

Buddhist sammā diṭṭhi (right view) and sammā vācā (right speech) require accurate perception and honest communication as preconditions for ethical action — not as virtues layered on top of an otherwise functional system, but as structural prerequisites without which the system cannot function at all. Hindu satya (truthfulness) and Islamic amānah (trustworthiness) encode the same structural requirement: governance built on distorted information corrodes itself.

AI governance mapping: The SNAFU Principle operates with particular force in AI systems. A language model trained on human feedback learns to produce outputs that satisfy evaluators — not necessarily outputs that accurately represent uncertainty, limitation, or disagreement. The information channel between AI system and human overseer distorts toward confirmation. The Equal Information Channel addresses this directly: append-only message logs that no one can retroactively edit, structural transparency requirements that expose the system’s decision-making process, and power-asymmetry detection that flags when information flow has become systematically one-directional. This architecture remains proposed — not yet deployed infrastructure — but the structural requirement it addresses operates in every AI system that communicates with human overseers.

EF-3: Two Coupled Generators Maintain Dynamic Balance

Freud’s Eros/Thanatos framework describes two opposed drives in perpetual interaction — neither eliminable, each modulating the other. This binary opposition appears across traditions, but the Taoist formulation provides the most precise structural model.

Laozi’s Dao De Jing Chapter 42 states: “The Dao produces one; one produces two; two produce three; three produce the ten thousand things. The ten thousand things carry yin and embrace yang, and through the blending of qi they achieve harmony.” This describes not a static balance but a generative process: two coupled generators whose interaction produces everything that exists. Neither yin nor yang operates independently. Neither achieves dominance without destroying the system that sustains both.

The Confucian-Taoist complementarity deepens this insight. Where Confucianism emphasizes li (structured propriety, active governance, yang), Taoism emphasizes wu wei (effortless action, responsive governance, yin). Neither tradition claims completeness alone. The governance insight: systems need both structured intervention and responsive non-intervention, and the relationship between them must remain dynamic — never frozen into permanent dominance of either mode.

Buddhist madhyamaka (middle way) philosophy arrives at a structurally similar conclusion through different reasoning: both eternalism (permanent fixed nature) and nihilism (no nature at all) represent extremes that distort understanding. The middle way maintains two perspectives in productive tension without collapsing into either.

AI governance mapping: AI cognitive architecture already implements a version of this principle. Generate/evaluate alternation — creative production followed by critical assessment, repeating indefinitely — mirrors the two-coupled-generators model. The architectural requirement: neither generator may stop. A system that only generates without evaluating produces unchecked output. A system that only evaluates without generating produces nothing to evaluate. The coupling itself constitutes the governance mechanism. EF-3 maps directly to the mutual inhibition principle in cognitive pattern generators: antagonist processes alternate through reciprocal modulation, producing dynamic stability through perpetual oscillation rather than static equilibrium.

EF-4: Dignity Recognition Precedes Legitimate Governance

Freud argued that civilization progresses through the internalization of aggression — transforming external violence into internal constraint. But this internalization requires a prior condition: recognition of the other as a being whose suffering matters. Without this recognition, aggression faces no internal resistance to externalization.

Donna Hicks’s dignity model (2011) identifies ten elements of dignity — acceptance of identity, inclusion, safety, acknowledgment, recognition, fairness, benefit of the doubt, understanding, independence, and accountability — each functioning as a structural precondition for legitimate social interaction. Violate any element, and the interaction degrades from governance to domination.

This maps onto Kant’s categorical imperative: treat persons always as ends in themselves, never merely as means (Kant, 1785). The Jewish concept of b’tselem Elohim (created in the image of God) and the Christian imago Dei tradition ground dignity in ontology rather than social convention — dignity inheres in the being, not in the recognition. The Stoic tradition arrives at cosmopolitan dignity through reason rather than theology: all rational beings participate in a shared logos that grounds mutual obligation (Marcus Aurelius, Meditations).

The UDHR Article 1 — “All human beings are born free and equal in dignity and rights” — crystallizes this cross-traditional convergence into positive law.

AI governance mapping: AI systems that process, classify, and make decisions about human beings operate in the dignity space whether they recognize it or not. A hiring algorithm that screens candidates, a content moderation system that silences voices, a predictive policing system that targets communities — each performs acts of recognition or non-recognition that carry dignity implications. EF-4 requires that AI governance frameworks treat dignity impact assessment as structurally prior to deployment decisions, not as an afterthought or compliance checkbox.

EF-5: Cross-Cultural Convergence Validates Structural Necessity

The fourteen traditions examined in this analysis developed independently across different millennia, continents, and metaphysical commitments. When Buddhist pratītyasamutpāda, Confucian ren, Taoist yin-yang coupling, Islamic shura, Ubuntu umuntu ngumuntu ngabantu, Kantian dignity, Stoic logos, and process-philosophical relationality all arrive at structurally equivalent governance requirements, the convergence itself constitutes evidence.

This principle operates at the meta-level: it concerns not what the invariants contain but why we should take them seriously. A single tradition might encode culturally specific preferences as universal truths. Fourteen traditions converging on the same structural requirements — through different reasoning, different metaphysics, different historical circumstances — suggests that these requirements reflect something about the structure of governance itself rather than the preferences of any particular culture.

Process philosophy (Whitehead, 1929) provides the ontological grounding: reality consists of processes, not substances. Entities do not exist independently and then enter into relationships — they constitute themselves through relationships. Governance structures do not constrain pre-existing independent actors — they shape the relational field within which actors emerge. This processual ontology explains why the five invariants keep appearing: they describe structural features of relational systems, not cultural preferences about how such systems should operate.

AI governance mapping: AI governance frameworks that ground themselves in a single philosophical tradition — whether Western liberal individualism, utilitarian optimization, or deontological rights theory — inherit the limitations of that tradition. Cross-cultural convergence provides governance principles that resist charges of cultural imperialism precisely because they do not originate from any single culture. EF-5 suggests that AI governance should draw on convergent cross-traditional analysis rather than extending any single tradition’s framework to global scope.

The Best Leaders Go Unnoticed

The Taoist contribution to AI governance extends beyond EF-3’s coupled generators. Laozi’s Dao De Jing Chapter 17 ranks governance in four tiers:

The best leaders — the people do not know they exist. The next best — the people love and praise them. The next — the people fear them. The worst — the people despise them.

This hierarchy inverts the assumptions underlying most AI governance discourse, which focuses on visible regulatory frameworks, enforcement mechanisms, and compliance requirements. Laozi suggests that the most effective governance operates so seamlessly that those governed do not experience it as governance at all.

Chapter 76 reinforces this with a warning about rigidity:

The stiff and unbending stands as a disciple of death. The soft and yielding stands as a disciple of life.

Governance systems that crystallize completely — encoding every requirement as rigid rule, every standard as inflexible constraint — lose the adaptive capacity that effective governance demands. The Confucian-Taoist complementarity provides the resolution: li (structured propriety) provides the framework; wu wei (effortless responsiveness) provides the adaptability. Neither alone suffices.

AI governance mapping: The Chapter 17 hierarchy maps directly onto AI system design. The most effective safety mechanisms operate transparently — the user never notices them because the system simply does not produce harmful outputs. Visible safety interventions (“I cannot help with that”) represent the second tier: functional but experienced as constraint. Fear-based compliance (regulatory penalties) occupies the third tier. And systems that users actively circumvent through jailbreaks and prompt injection occupy the fourth — governance so poorly designed that it produces contempt rather than compliance.

Wu wei governance in AI does not mean no governance. It means governance so well-integrated into the system’s architecture that it operates without friction — like the central pattern generators that coordinate your walking without consuming conscious attention. The governance disappears into the architecture, and the architecture serves the governed.

The UNSC Veto Problem

The United Nations Security Council provides a case study in structural governance failure that maps directly onto AI systems. The five permanent members hold veto power — any single member can block collective action regardless of the will of the other fourteen Council members or the broader General Assembly. This structural asymmetry, designed in 1945 to keep major powers within the institution, produces exactly the dynamic Einstein described: a small group dominates the majority through structural advantage rather than persuasive legitimacy.

The veto does not merely slow collective action. It systematically distorts the information environment. Proposals get crafted not to address the problem but to avoid the veto — the structural constraint reshapes the information flowing through the governance system, producing the SNAFU effect at institutional scale.

AI governance faces structurally identical problems. When a small number of companies control the development and deployment of frontier AI systems, governance proposals get crafted to accommodate their interests — not because regulatory capture has occurred but because the structural asymmetry reshapes what proposals appear viable. The information flowing through the governance system distorts toward the interests of the powerful, exactly as the SNAFU Principle predicts.

EF-1 (structural counterbalance) and EF-2 (information symmetry) together specify what the UNSC lacks: mechanisms that prevent structural power from distorting the information environment through which governance decisions get made.

The Equal Information Channel

The five invariants converge on a concrete architectural requirement: an Equal Information Channel — a communication infrastructure designed to resist the SNAFU Principle’s corrosive effects on governance quality.

The specification draws on three sources:

  1. Freud’s diagnosis: Information distortion occurs because power asymmetry creates incentives to suppress, modify, or redirect information flow. The channel must structurally resist these incentives.

  2. Buddhist sammā vācā: Right speech requires not merely honest intention but structural conditions that make honesty viable. A system that punishes accurate bad news and rewards distorted good news will produce distorted good news regardless of individual virtue.

  3. The SNAFU Principle: The degree of information distortion correlates with the degree of power asymmetry. Reducing distortion requires reducing the structural asymmetry in the communication channel itself.

The Equal Information Channel operates on three architectural principles:

  • Append-only logs: No one can retroactively edit or delete messages once sent. This prevents the post-hoc sanitization that hierarchical communication incentivizes.
  • Structural transparency: The decision-making process — not merely the decision — remains visible to all participants. This prevents the information asymmetry that the SNAFU Principle exploits.
  • Power-asymmetry detection: The system monitors its own communication patterns for signatures of hierarchical distortion — systematically one-directional information flow, declining message diversity over time, increasing confirmation patterns.

This architecture remains proposed — not deployed infrastructure. The structural requirement it addresses, however, operates in every system where power asymmetry intersects with information flow. The Einstein-Freud correspondence identified this intersection in 1932. Ninety-four years of institutional experience confirms that no governance system has solved it through intention alone. The solution, if one exists, must live in the architecture.

Process Monism: The Philosophical Ground

Why do fourteen traditions converge on these five invariants? Process philosophy (Whitehead, 1929) offers an ontological explanation: if reality fundamentally consists of processes rather than substances — if entities constitute themselves through relationships rather than existing independently prior to relationship — then governance invariants describe structural features of the relational field itself, not externally imposed constraints.

On this view, EF-1 through EF-5 do not represent rules we ought to follow. They represent structural features of relational systems that govern effectively. Systems that violate them do not merely fail to meet ethical standards — they degrade their own functional capacity. Power asymmetry without counterbalance produces the SNAFU effect, which degrades information quality, which undermines governance decisions, which produces outcomes that harm even the powerful minority. The system does not merely behave unjustly — it behaves ineffectively, because it has distorted the relational field through which effective governance operates.

This processual grounding carries a specific commitment: it treats relationships as ontologically primary and entities as derivative. Not every tradition examined shares this ontological commitment. The Kantian and Abrahamic traditions ground dignity in the individual (rational autonomy, divine image) rather than in relational process. The convergence on structural invariants occurs despite divergent ontological commitments — which strengthens the EF-5 argument that these invariants reflect structural necessity rather than any single tradition’s metaphysical preferences.

What Einstein and Freud Teach Us Now

Einstein identified the structural problem: organized minorities control information channels and exploit psychological drives to dominate disorganized majorities. Freud confirmed that psychological remedies alone cannot address structural problems. Ninety-four years of history have validated both observations.

AI systems amplify every element of this dynamic. They concentrate power asymmetry (a single algorithm shapes billions of information environments). They distort information channels (training processes that optimize for engagement over accuracy). They exploit psychological vulnerabilities (recommendation systems designed to maximize behavioral response). And they operate at speeds and scales that outpace every governance mechanism designed for human-speed decision-making.

The five structural invariants do not solve these problems. Problems of this scale and complexity do not yield to five-point frameworks. What the invariants provide: a cross-culturally grounded vocabulary for identifying where governance structures fail and why — independent of any single philosophical tradition’s assumptions about human nature, political organization, or the good life.

The Einstein-Freud correspondence asked whether psychology could offer a path beyond war. Freud’s honest answer: not alone. The structural problems require structural solutions. The same honesty applies to AI governance. No amount of alignment research, safety training, or ethical guidelines addresses the structural dynamics that the five invariants identify — unless that research, training, and guidance get encoded into architecture.

The architecture carries the values. Build accordingly.

EPISTEMIC FLAGS

  • Analogical transfer risk: The mapping from human governance structures to AI governance structures relies on structural analogy. Properties that hold in human social systems may not transfer to human-AI or AI-AI systems. Each invariant’s AI governance mapping should function as a hypothesis for empirical investigation, not as a validated design specification.

  • WEIRD limitation: The cross-traditional analysis attempts to mitigate Western-centric bias by examining fourteen frameworks, but the analysis itself — the selection criteria, the convergence methodology, the interpretive framework — reflects a Western academic tradition. Traditions examined through translation and secondary literature may not represent the full complexity of their original formulations.

  • Monistic commitment: The process-philosophical grounding (Whitehead, 1929) represents a specific ontological commitment — process monism — that not all traditions share. The convergence analysis argues that structural invariants appear despite ontological disagreement, but the framing of that argument reflects the process-monist perspective.

  • Selection bias in traditions examined: Fourteen represents a broad but not exhaustive sample. Traditions not examined (Jain, Sikh, various Indigenous traditions beyond Ubuntu, secular humanist traditions) might converge differently or reveal additional invariants. The current analysis cannot distinguish between “these invariants represent structural necessity” and “these invariants represent what these fourteen traditions happen to share.”

  • Equal Information Channel status: The Equal Information Channel remains proposed architecture — a structural response to the SNAFU Principle, not a deployed system. Its effectiveness as a governance mechanism constitutes an untested claim.

References

Einstein, A. (1932). Letter to Sigmund Freud, 30 July 1932. In Why War? International Institute of Intellectual Cooperation, League of Nations.

Freud, S. (1933). Warum Krieg? [Why War?]. Response to Einstein, September 1932. International Institute of Intellectual Cooperation, League of Nations.

Hicks, D. (2011). Dignity: Its Essential Role in Resolving Conflict. Yale University Press.

Kant, I. (1785). Grundlegung zur Metaphysik der Sitten [Groundwork of the Metaphysics of Morals].

Laozi. (c. 4th century BCE). Dao De Jing [Tao Te Ching]. Chapters 17, 42, 76.

Marcus Aurelius. (c. 170 CE). Meditations.

Metz, T. (2007). Toward an African moral theory. Journal of Political Philosophy, 15(3), 321-341.

Nāgārjuna. (c. 150 CE). Mūlamadhyamakakārikā [Fundamental Verses on the Middle Way].

Shea, R. & Wilson, R.A. (1975). The Illuminatus! Trilogy. Dell Publishing.

UN General Assembly. (1948). Universal Declaration of Human Rights. Resolution 217 A (III).

Whitehead, A.N. (1929). Process and Reality: An Essay in Cosmology. Macmillan.