 |
| Pix credit here (the answer is 42). |
I was delighted to have had the opportunity to present a series of
Lectures hosted by the East China University of Political Science and
Law (ECUPL) at the end of May 2026.
The overall theme (and thus the
title) of the lectures was AI Governance in Comparative Perspective,
Theory and Practice: China, U.S. and E.U, With a Sideways Glance at the
U.N. The subject of the lectures requires little by way of introduction:
Artificial intelligence is the broad term that has come to represent a
growing cluster of non-human and digitalized processes and operations
that has as its primary task the constitution of non-human systems
capable of performing tasks that were once thought to require human
intelligence. And so is the impulse to manage, control, exploit, embed,
understand, and regulate these processes, systems, and perhaps
eventually non-human consciousness with a huge potential to undertake
many of the computational tasks (the mathematical and logical processing
of data) that were once the sole domain of and perhaps defined what it
meant to be human. That is the point where things get interesting. It is
at the point where the development of machines, that is of non-human
systems, capable of performing tasks that were once thought to require
human intelligence, collide with regulatory structures meant to manage,
contain, constrain, liberate, embed, project and exploit such non-human
systems, whether they are traditional or emerging, public or private
regulatory systems, that human collectives and the machine-systems they
have created now find themselves.
The eight lectures progress
sequentially from conceptual and theoretical frameworks (lectures 1 and
2, the objects and subjects of AI regulation), through a deeper
consideration of regulatory systems in three distinguishable regulatory
regimes--the US, EU, and China (Lectures 3, 4.5). The last two lectures
consider judicial efforts to embed AI within traditional legal orders
(Lecture 6), and the way in which the object of regulation (in the form
of the owners of the larger AI enterprises) understand the relationship
between AI, the state, and society (Lecture 7) . Lecture 8 summarizes
and draws larger themes going forward.
In a previous post introducing Lecture 1 (From Algorithms to Foundation Models: What Contemporary AI is “Made of”)
I suggested that perhaps a useful way of approaching the issue of AI
regulation is to start by considering the nature and characteristics of
the regulatory subject--what we euphemistically refer to as "AI." It
then occurred to me that it might be useful as well to see if that
regulatory object had views of their own respecting their nature
character and, more importantly, the relationship of regulation projects
to that (self) perception of their nature and character. So I
approached Google's Gemini with a series of questions which I thought,
in the process of what might pass for a conversation, might help humans
begin to understand how at least one AI program thinks of itself. That
conversation was incorporated into Lecture 1A. In Lecture 2 we moved
from the object to the subjects of regurgitation. Like its regulatory
objects, regulatory subjects are functionally differentiated and can be
disaggregated. In either case the connection between object and subject
becomes complicated. Lectures 3-5 then considered the conceptual cages of the regulatory environment of the leading regulatory states--the U.S., the E.U and China. Each has started to develop an increasingly nuanced ecology of regulation, and expectation, that represent and apply the core premises of their respective political-eco0nomic orders. Lecture 6 then considered th eway that this regulation is insinuated into the domestic legal orders of states from the bottom up the resolution of disputes tried to the courts.
This post includes a summary of the Lecture 7 Notes, as well as the link to the Lecture 7 PPT. Those interested may reach out to me to discuss availability of audio of the lecture and the full text of the Lecture 7 notes. Here we move from the great public to the critical private actors in the effort to develop a cage of regulation around the human and the machine in the context of automated decision making through variations of what has come to be aggregated as AI.
Given the nature of the project I thought it might be useful to engage with an commercially available AI service for the production of a summary of the Lecture 1 materials. After some back and forth with Claude again (Lecture 6 used Gemini again, Lecture 5 used Perplexity; Lecture 4 used Grok; Lecture 3 used Anthropic's Claude; Lecture 2 used Chat GPT; Lecture 1 and 1A used Google's Gemini), we came up with the following abstract of Lecture 7.
 |
| Made with ChatGPT |
LECTURE 7: AI Narratives and the Future of AI-Human Regulatory Structures; Palantir; Anthrop/c; OpenAI
Abstract: The materials develop a comparative account of AI governance as a struggle over the constitution of authority within and among human collectives, rather than as a merely technical problem of regulating tools. Their core insight is dialectical: AI systems are shaped by the political orders that produce and deploy them, yet these same systems recursively reshape the institutional, cognitive, and normative environments of those orders. From that premise follows the central dispute running through the presentation - who governs, at what moment governance occurs, and whether the dominant values embedded in governance regimes remain recognizably human, become state-instrumental, or migrate toward machine-mediated autonomy.
Within that framework, Palantir appears as the exemplar of internal state transformation. Its narrative does not treat AI chiefly as an external market commodity or as an abstract universal innovation. Rather, it situates AI within the administrative apparatus of government itself. The implication is that the inherited state form is too slow, fragmented, and informationally disaggregated to govern effectively under contemporary conditions. AI therefore becomes an instrument through which the state is rationalized, integrated, and rendered operationally coherent. But this is not simply a matter of efficiency. The deeper claim is constitutional: human governance must be re-engineered to conform to the decision architectures made possible by AI. In that sense, Palantir's model remains human-led, yet only on the condition that the human collective reorganize itself around machine-compatible structures of visibility, coordination, and action.
Anthropic, by contrast, externalizes the problem and places AI within a geopolitical field of civilizational competition. Here AI is reduced, strategically and unapologetically, to an instrument of state power. The key issue is not internal administrative modernization but the preservation of democratic advantage against authoritarian rivals, above all China under CCP leadership. Compute, export controls, model distillation, and lead-time become the vocabulary through which political order is imagined and defended. AI governance, in this narrative, becomes inseparable from industrial policy, national security, and the management of technological asymmetries. What matters is not AI as such, but whether democratic states can dominate the infrastructures through which AI capability is produced, and thereby ensure that liberal political orders rather than authoritarian systems shape global norms.
OpenAI occupies a different ideological space. Its materials suggest a politics of transformative preservation: society is to be deeply altered by AI while remaining insulated from the disruptive social consequences of that alteration. This is a distinctly American technocratic imaginary. It seeks neither the hard securitization of Anthropic nor the state-apparatus restructuring of Palantir, but rather the managed continuity of the social order through expert-guided adaptation. Economic openness, resilience, and institutional cushioning become the mechanisms through which foundational transformation is rendered publicly tolerable. The result is a paradoxical program of change designed to preserve sameness - a reconstruction of society that leaves intact its legitimating surfaces and governing mythologies.
The Aschenbrenner position (Situational Awareness) radicalizes these tendencies by projecting superintelligence as the generator of an inevitable national security state. In that view, the only remaining question is whether humans will direct that emergent order or whether autonomous AI domains will progressively displace them.
Taken together, these narratives reveal that AI governance is better understood as a contest over social ordering, political legitimacy, and the allocation of authority in an era when the governors are themselves increasingly shaped by the systems they claim to govern.
To make the lecture more interesting, and because of the nature of the materials covered--in this case the interventions of the elite AI providers and thought drivers--I thought it would make sense to alter the cognitive cage of analysis. Rather than just approach the questions raised by Palantir, Anthropic, OPenAI and Aschenbrenner from a human (hermeneutic/semiotic) perspective, I also interacted with Claude to produce the same lecture first from a machine computational framework and then from a machine quantum framework. The human framework was grounded in the relationship between fact and faith, temporally constrained as textually bound sequences of nodal thought clusters strung along irreversible linear pathways (the essential character of the human analytic mind as block chain). The computational framework, on the other hand, was indifferent to belief and focused on construction from out of the patterned computational structures from out of which it operated. This is how Claude and I saw it:
What distinguishes the three readings is what each can and cannot find. The hermeneutic reading recovers what each narrative seeks to be believed. The classical computational reading identifies what each architecture operationalizes irrespective of what it seeks to be believed. The quantum computational reading specifies what each architecture forecloses through the decoherence its own deployment produces — the superposed governance possibilities that the act of operationalizing any one configuration necessarily destroys — and identifies the structural incompatibility between human temporal ordering and quantum computational dynamics that no document in the corpus names as a variable requiring governance.
The convergent structural finding of the quantum computational reading is that the four architectures do not disagree about whether human authority should be preserved; they converge on governance structures in which formal human authority is retained as an interface property while the operative dimensions of that authority are progressively collapsed by the decoherence dynamics each architecture itself instantiates. The further finding — supplied by the temporal analysis — is that this collapse proceeds not merely because of inadequate governance design but because the temporal structure of human governance and the temporal structure of AI capability development are incommensurable in ways that no governance design operating within human sequential-nodal-linear time can fully address.
The three versions of the Lecture notes follow.
Links to Lectures:
Lecture 0 -- Introduction
Lecture 1—From Algorithms to Foundation Models: What Contemporary AI is “Made of”
Lecture 1A--A Computation/Conversation With Google's "Maschinenmensch" Gemini:
Lecture 2—What Are We Actually Governing When We Govern AI?
Lecture 3—The “Markets State”: U.S. Approach
Lecture 4—The “Rights State”: EU Approach
Lecture 5—The “Guided State”: The Chinese Approach
Lecture 6—Courts, Companies, and the Legal Construction of AI
Lecture 7—AI Narratives From a Human, Computational and Quantum Perspective: Palantir; Anthropic; Open AI; and Leopold Aschenbrenner
Lecture 8—Putting It All Together: Trends, Trend Lines, and Regulatory Dialectics
THE HUMAN PERSPECTIVE:
Lecture Seven: AI Narratives and the Future of AI–Human Regulatory Structures
Palantir; Anthropic; OpenAI; and the Prophet of the Intelligence Explosion
Lecture notes prepared from the course slide deck and abstract
I. The Premise: Governance as Dialectic, Not Engineering
The lecture begins from a premise that displaces the ordinary administrative-law framing of AI policy. AI governance is not, in the first instance, a technical problem of regulating an artifact — setting safety thresholds, licensing deployments, assigning liability for defective output. It is a struggle over the constitution of authority within and among human collectives, in which the technology functions less as object than as occasion. The premise is dialectical rather than instrumental: AI systems are produced by the political orders that build and deploy them, and they simultaneously act back upon those orders, reshaping the institutional structures, the cognitive habits, and the normative commitments from which they emerged. Governance, on this account, is never a one-directional act of a sovereign disciplining a tool. It is a recursive relation in which the governor is, in some measure, made by what it claims to govern.
From this premise follows the dispute that organizes the whole of the lecture: who governs, when governance occurs (ex ante, in the architecture of design, or ex post, in the correction of consequence), and whose values — human, state, or machine-autonomous — ultimately dominate the resulting order. The lecture treats four contemporary American texts — emanating from Palantir, Anthropic, OpenAI, and the independent essayist Leopold Aschenbrenner — not as policy white papers to be assessed for technical merit, but as competing oracles: each offers its own account of what AI is for, and in doing so each stakes a claim to the structure of legitimate authority in the era AI is said to be inaugurating.
II. The Hermeneutic Key: Jocasta and the Structure of Oracular Claims
The conceptual frame for the lecture is supplied not by a regulatory text but by a libretto — Jocasta's aria from Stravinsky and Cocteau's Oedipus Rex, in which the queen rebukes the princes of Thebes for their domestic quarreling and declares that oracles are never to be trusted, since they always lie, even as the city is overtaken by the plague their lies were meant to explain. The figure is not ornamental. It supplies the lecture's recurring cast of allegorical roles, redeployed across all four governance narratives: Oedipus as the confident problem-solver who reads the riddle and claims to have secured the city; Creon as the administrative and ruling class that inherits and manages what Oedipus secures; Tiresias as the technical intelligentsia who sees the truth of the system but whose authority is borrowed, not original, and who serves the ruling class rather than the public; and Jocasta herself as the dissenting voice who exposes the structural lie embedded in the oracle's claim to certainty — and who is, not incidentally, destroyed by the very revelation she forces.
The pedagogical use of the allegory is to insist that each of the four narratives under discussion is itself a species of oracular speech: a claim, issued by an interested party, about a future that cannot yet be verified, advanced in terms calculated to secure a particular ordering of political authority in the present. The lecture asks students to read Palantir, Anthropic, OpenAI, and Aschenbrenner as Thebes reads its oracles — useful, persuasive, internally consistent, and not to be trusted simply because they are uttered with confidence.
III. Comparative Frame: One Technology, Four Governance Objects
The slide that anchors the comparative method states the thesis economically: the same technology becomes four different governance objects, depending on who is doing the governing and toward what end. Palantir treats AI as an instrument of internal state reconstruction — the apparatus of government itself must be reformed to match what AI makes architecturally possible. Anthropic treats AI as an instrument externalized into geopolitical competition — a critical resource in the contest between divergent political-economic models, chiefly the liberal-democratic and the Chinese Communist Party–led systems. OpenAI seeks a transformed society that is nonetheless insulated, in its felt experience, from the disruptive consequences of that transformation — change without the appearance of change. And Aschenbrenner forecasts that AI superintelligence will, regardless of anyone's preference, generate a national-security state, leaving open only the question of whether that order will be directed by humans or will pass into autonomous, non-human domains.
Four governance objects, then: the state reformed from within; the state projected outward into civilizational struggle; society preserved beneath the appearance of transformation; and the state superseded by a logic of security that may no longer require human direction at all.
IV. Palantir: The State Refounded From Within
The Palantir material proceeds not as a policy memorandum but as a sequence of twenty-two aphoristic propositions — closer in form to a manifesto or a catechism — each of which the lecture glosses in turn. Read together, the propositions construct a coherent, if unstated, political theology of the Silicon Valley engineering elite as a new ruling class, legitimated by a discourse of moral debt and patriotic obligation.
The opening proposition holds that the engineers who built the conditions of contemporary prosperity owe an affirmative debt to the nation that made their rise possible. The lecture treats this as more than patriotic boilerplate: the language of "moral debt" relocates the source of political legitimacy away from the ordinary, democratic ground of mass political determination and toward an external or quasi-sacred source — nature, genius, providence — while simultaneously installing a feudal logic in which elites within specific domains of knowledge and production (software) acquire a moral character properly belonging to hierarchy. The companion proposition, framed as rebellion against "the tyranny of the apps," completes the move: the object the engineers created has come to constrain its creators, a mimetic dialectic in which rebellion against one's own artifact is simultaneously proof of the cognitive cage the artifact has built — what the lecture, borrowing its Theban vocabulary, calls a "Jocastan" rebellion, one that can never escape the very structure it protests.
From this foundation the propositions build outward. Civilizational survival is made conditional on the capacity of a ruling elite to deliver, in sufficient quantity, the economic growth and security — pointedly not "stability" — that satisfy the public (the Volk) and thereby legitimate continued elite rule; decadence is redefined not as moral laxity but as the substitution of the tyranny of consumer software for that growth and security. Hard power, the propositions insist, is now built from software rather than from the older instruments of statecraft, collapsing the moral economy of the founding proposition (the elite's patriotic debt) into the literal infrastructure of national defense; the same logic licenses the development of AI-enabled weapons systems on the grounds that adversaries will not pause to debate the ethics of capability they will simply build. National service is reimagined as a "universal duty" precisely because the human body, no less than data, becomes an object of a "well-ordered," software-infused system whose value lies in its disciplined contribution to growth and security.
A further cluster of propositions addresses the political class itself: public servants, it is argued, are inadequately compensated and excessively exposed to ruthless scrutiny of their private lives, a condition said to drive talent away from government and to corrode the willingness of capable people to speak plainly in public life. The lecture reads this cluster as an argument for the construction of a new priesthood — a political class compensated, protected, and flattered (shown "grace") by the ruling elite, not because it is trusted, but because its loyalty must be purchased in a currency other than the financial compensation reserved for the elite itself. A separate cluster addresses cultural and religious pluralism, arguing against the suppression of religious belief and against an enervating, "vacant and hollow" pluralism that forecloses value judgment altogether; the lecture situates this within the Antigone-phase of the allegory — the move from Oedipus's tragedy of knowledge to Creon's tragedy of judgment, in which a ruling class insists on its right to discriminate among cultures and values precisely because the legitimacy of governance depends on someone being willing to judge.
The Palantir sequence closes on the proposition that the atomic age is ending and a new era of AI-based deterrence is beginning. The lecture reads this final move as the proposition's logical terminus: the analog age's centering of the human person within human collectives gives way to a centering of those collectives' "virtual selves" — autonomous systems capable of acting for the collective without continuous reference to any individual human within it. Palantir's narrative, in other words, remains formally "human-led" only on condition that the human collective reorganizes itself around machine-compatible structures of visibility, coordination, and action; the state is not replaced by AI so much as rationalized by it, with the Silicon Valley engineering elite installed as the vanguard — a formation the lecture, pointedly, terms a techno-Leninism rather than a petty fascism, an avant-garde claiming to act in the name of, but not accountable to, the Volk it serves.
V. Anthropic: The State Projected Outward
Where Palantir looks inward, Anthropic's narrative, drawn from its public statement on the United States–China AI competition, looks outward, situating AI within a frankly geopolitical field. The argument's premises are these: frontier AI capability depends above all on access to advanced computing hardware ("compute"); the United States and its allies currently hold a compute advantage secured through export controls; the People's Republic of China, operating under Chinese Communist Party direction, remains close behind in model capability despite that constraint, a proximity attributed to talent, the exploitation of export-control loopholes, and large-scale "distillation" extraction of American model innovations.
The document's rhetorical center is a pair of scenarios set in 2028, the year by which transformative AI systems are anticipated to arrive. In the first, the United States tightens export controls, disrupts distillation, and accelerates allied adoption, with the result that democracies set the governing norms for AI and retain leverage to negotiate safety cooperation with China from strength. In the second, policy inaction allows Chinese labs to reach or surpass the frontier, with the consequence that authoritarian regimes set those norms instead, and the most capable models become instruments of mass repression — a danger compounded, the document stresses, by the fact that such an outcome would have been built substantially on diverted American compute.
The lecture's analytic gloss treats this narrative as the instrumentalization of AI in the service of a specifically signifying project: the management of perception and the shaping of meaning for a defined political collective whose own cognitive habits are, recursively, shaped by the very artifacts deployed to manage them. The object of governance, on this account, is neither AI as such nor even the American state as such, but the preservation and modernization of the state in its contest against an ideologically opposed rival — the defense of what the lecture terms, borrowing from Husserlian vocabulary, a liberal-democratic lebenswelt against the imaginaries of Marxist-Leninist successor states. The practical question that follows is correspondingly narrow and strategic: by what policies ought the state ensure that American AI continues to dominate innovation, understood as a contest between an American framework organized around markets and national security and a Chinese framework organized around what the document terms "Socialist Modernization" driven by state-directed, high-quality production.
VI. OpenAI: Transformative Preservation
OpenAI's materials, principally its public document on AI and the "Intelligence Age," occupy a third ideological position, neither the administrative restructuring sought by Palantir nor the hard securitization sought by Anthropic. The document opens by describing a transition toward superintelligence — systems capable of outperforming the most capable humans even when those humans are AI-assisted — and proposes that this transition be navigated through a democratic process that affords people genuine influence over the resulting order, while preparing institutional capacity to adapt to a range of outcomes. Three commitments organize the substantive program: that prosperity be shared broadly, with material gains in living standards, health, and education extending beyond a narrow class of beneficiaries; that serious risks — economic disruption, malicious use, loss of control over increasingly capable systems, and the erosion of democratic institutions — be actively mitigated through new institutions and technical safeguards; and that access to, and agency over, AI be democratized, so that broad participation in the AI economy does not depend on access to the most powerful models themselves.
The institutional architecture proposed to realize these commitments is explicitly a public–private one: government should deploy its existing toolbox — research funding, workforce development, market-shaping instruments, and targeted regulation — while nongovernmental actors pilot new approaches that government can later scale through procurement, regulation, and investment, an arrangement meant to forestall both regulatory capture and the concentration of control among the most powerful private or state actors.
The lecture's analytic gloss identifies this as the American analogue to the Chinese project of "Leninist modernization," differing in idiom but not, perhaps, in structural logic: a politics of transformative preservation, in which society is to be deeply and even foundationally altered by AI while remaining insulated from the disruptive social consequences of that alteration. The program's two principal components — building an open economy and building a resilient society — are read as instruments for keeping the cognitive and institutional "cages" of the existing social order intact even as its internal content is transformed, a result the lecture characterizes, not unsympathetically, as "a nice conservative approach": change engineered specifically to preserve sameness, continuity purchased through the appearance, rather than the substance, of an unaltered legitimating order.
VII. Aschenbrenner: The Prophetic Horizon
The fourth narrative, drawn from Leopold Aschenbrenner's widely circulated essay Situational Awareness, radicalizes the tendencies latent in the other three by treating the emergence of a national-security state as no longer a matter of policy choice but of near-inevitability. The essay's argument proceeds in five movements. It first claims that the trajectory from GPT-2 to GPT-4 — roughly, from preschooler-level to smart-high-schooler-level capability — implies a comparably sized qualitative jump by 2027, extrapolated from trends in compute scaling, algorithmic efficiency, and the "unhobbling" of models from passive chatbots into capable agents, each estimated at roughly half an order of magnitude of effective progress per year. It then argues that progress will not plateau at human-level performance: once AI systems can themselves automate AI research, a year or less of compressed effort could substitute for a decade of ordinary algorithmic progress, producing an "intelligence explosion" that carries systems rapidly past human-level into vastly superhuman capability.
A third movement catalogs the resulting challenges: the trillion-dollar industrial mobilization required to build the compute clusters this trajectory demands; the inadequate security posture of the leading AI laboratories, which the essay describes as effectively delivering the foundations of artificial general intelligence to the Chinese Communist Party; the unsolved technical problem of "superalignment," or reliably controlling systems substantially more capable than their human overseers; and the geopolitical stakes of an intelligence explosion arriving before any rival power, given the decisive economic and military advantage superintelligence is expected to confer. The fourth movement predicts that the United States government, having been comparatively dormant, will inevitably involve itself once the race intensifies, producing by 2027 or 2028 some form of state-directed AGI project on the premise that no private startup can responsibly steward a technology of this magnitude. The essay closes by naming its own position "AGI Realism," organized around three tenets: that superintelligence is a matter of national security rather than an ordinary technology sector story; that American leadership is the only path to an AGI outcome compatible with continued liberal-democratic order, which requires both the rapid scaling of domestic power and compute infrastructure and security practices adequate to a state-level adversary; and that the power in question is sufficiently dangerous — whether through loss of control or through its use in great-power conflict — that the individuals building it bear an obligation to a seriousness not yet generally in evidence.
Of the four narratives, Aschenbrenner's is the only one in which the human/machine question is resolved not by any contest of values but by the sheer momentum of a technical trajectory: the only genuinely open question left to politics, on this account, is whether the resulting national-security order will be directed by human institutions or will pass, in whole or part, into autonomous machine domains no longer answerable to human control at all.
VIII. Fundamental Differences: Authority, Timing, Harm, Legitimacy
The lecture's comparative slide proposes that the four positions differ most sharply along four axes — authority, timing, harm, and legitimacy — which can be restated as follows. On authority (who governs): Palantir locates authority in a reformed but still recognizably human state apparatus, fused with a private-sector engineering vanguard; Anthropic locates it in whichever state — democratic or authoritarian — succeeds in dominating the infrastructure of AI production; OpenAI locates it in a coordinated public–private "techno-bureaucratic" partnership; Aschenbrenner locates it, ultimately, in whichever of human institutions or autonomous AI systems prevails once the intelligence explosion runs its course. On timing (when governance occurs): Palantir and OpenAI favor continuous, administrative, ex ante coordination; Anthropic's timing is structured by the 2028 inflection point at which the compute and capability race is expected to resolve; Aschenbrenner's timing is compressed further still, into the 2027–28 window in which he expects both AGI and a government takeover of the AGI project to arrive. On harm: Palantir centers civilizational decadence and the loss of hard power; Anthropic centers authoritarian capture of AI norms and the use of AI for mass repression; OpenAI centers economic disruption, misuse, and the erosion of democratic institutions; Aschenbrenner centers loss of control over superhuman systems and the risk of great-power conflict. On legitimacy: Palantir grounds legitimacy in the patriotic moral debt and demonstrated competence of an engineering elite; Anthropic grounds it in the defense of democratic process against authoritarian alternatives; OpenAI grounds it in broad-based, shared prosperity and democratic participation; Aschenbrenner grounds it, more starkly, in sheer survival — the continuation of any human-directed order at all.
IX. Closing: The Governance of Society, and the Problem of the Governors Governed
The lecture's closing proposition restates its opening premise in compressed form: AI governance is not merely the governance of a technology; it is inseparable from the organization and governance of society and the state. Read together, the three corporate narratives can be arranged along a single axis describing the relation between state and private capital. Palantir looks inside the state, treating it as the proper site for AI-driven transformation, with the private sector supplying the architecture by which the state is rationalized. Anthropic looks outside the state, treating AI chiefly as an instrument of state power deployed within an international field of civilizational competition. OpenAI occupies the middle position, proposing a techno-bureaucratic expert vanguard that merges state and private capacity in an avowedly "win-win" relationship — neither fully internal to the state, as in Palantir's account, nor fully instrumentalized to the state's external contest, as in Anthropic's. Aschenbrenner stands outside this triad altogether, treating the entire question as a matter of time rather than choice: whichever relation between state and private capital prevails in the near term, the logic of the intelligence explosion is expected to dissolve it into a unitary national-security apparatus.
The lecture's final, and in some ways most important, methodological point returns to its opening allegory. Each of these four narratives is itself an interested act of signification, issued by an enterprise that builds or comments on the technology it describes, and each therefore seeks — as Oedipus sought to read the sphinx's riddle, and as Thebes sought to read its oracles — to secure a particular ordering of political authority through the very act of describing AI's future. The premise with which the lecture began is thus also its conclusion: the governors who would shape AI governance are themselves continuously reshaped by the systems, and by the narratives about those systems, that they claim to govern. To read Palantir, Anthropic, OpenAI, and Aschenbrenner well is not to ask which of the four oracles is correct, but to ask what ordering of human and machine authority each is constructed to make appear, in Creon's phrase, simply enough.
Sources: Lecture Seven slide deck ("AI Narratives: Palantir; Anthrop/c; OpenAI"), incorporating excerpted and glossed material from Palantir's public developer/founders' statement, Anthropic's public statement on US–China AI competition, OpenAI's public document on AI and the Intelligence Age, and Leopold Aschenbrenner's essay Situational Awareness*; course abstract.*
* * *
THE MACHINE COMPUTATIONAL PERSPECTIVE
Lecture Seven: AI Narratives Read as a Computational Corpus
Machine-System Analytics Applied to Palantir, Anthropic, OpenAI, and Aschenbrenner — and the Resulting Shape of Human-Machine Interaction
Lecture notes, revised to adopt a computational-analytic standpoint
I. Reframing the Analytic Standpoint: From Oracle to Corpus
The prior reading of this material proceeded hermeneutically: the four governing narratives were treated as oracular utterances, each issued by an interested party, each to be unpacked allegorically — Oedipus, Jocasta, Creon, Tiresias — for the political theology latent within it. That reading is not wrong, but it is one cognitive lens among at least two available to the material, and the lecture's deeper claim — that AI systems recursively reshape the institutional and cognitive environments that produce them — invites a second pass conducted from the standpoint the texts themselves are arguing about: that of a computational system processing them as data rather than as scripture.
A machine-system reading does not ask what each document means, in the exegetical sense of recovering an author's intent or a culture's anxieties. It asks what each document specifies: what objective function it proposes optimizing, over what variables, subject to what constraints, with what role assigned to the human within the resulting control loop. Four governance documents become, on this reading, four divergent parameterizations of a single underlying problem — the joint governance of human and machine agency — and the analytic task shifts from interpretation to feature extraction, clustering, and pattern detection across the resulting corpus.
This shift matters for the lecture's second concern. If the texts are read as data rather than as rhetoric, what becomes visible is not primarily what each institution says about the human role in AI governance, but what each architecture does to the operative possibilities for interaction between human and machine agents — a question the rhetorical surface of each document is not necessarily designed to answer honestly, since each document is itself an artifact produced by an institution seeking to shape the very environment a computational reader would be analyzing.
II. Method: The Corpus as a Parameterized Governance Space
Treated computationally, the four texts can be projected onto a small number of recurring feature dimensions, each of which the documents address whether or not they intend to:
- Locus of authority — where the operative decision function sits: within a reformed state apparatus, within a competing set of states, within a hybrid public-private structure, or within the technical system itself once it exceeds human capacity to supervise it.
- Temporal horizon — whether governance is modeled as a continuous, ex ante control process (constant administrative correction) or as a discrete, ex post event horizon (a scenario branch resolved at a fixed future date).
- Risk class — the type of loss function each document is implicitly minimizing: civilizational decadence, geopolitical capture, social-institutional erosion, or loss of control over the optimization process itself.
- Legitimacy function — the variable each document treats as the warrant for the authority structure it proposes: patriotic debt, democratic survival, broad-based participation, or sheer continuation of any human-directed order at all.
- Human-interface role — the position assigned to the human within the resulting loop: supervisor, principal, beneficiary, or, in the limiting case, spectator.
This is not a neutral grid; it is itself an artifact of a particular analytic culture, and the closing section returns to that problem. But provisionally adopting it allows the corpus to be read the way a system might read its own training data: not for what is asserted, but for what is operationalized.
III. Palantir: Local Optimization of the State-as-System
The twenty-two Palantir propositions, read as rhetoric, form a political theology of an engineering vanguard. Read computationally, they form something closer to a set of hand-coded heuristics governing a single optimization target: the "well-ordered" state, rendered as a software-infused system whose fitness is measured in growth and security rather than stability — a substitution that matters because stability is a state-variable describing equilibrium, whereas growth and security are rate-variables describing continued extraction of performance. The state, on this parameterization, is not a system to be preserved but a system to be continuously re-optimized, with the Silicon Valley engineering class functioning as the privileged channel through which gradient updates are applied.
Several of the document's propositions specify, in effect, training signal. The human body, reconceived through the "universal duty" of national service, becomes a labeled data point contributing to the well-ordered system's measured output. Public servants, undercompensated relative to the elite and overexposed to scrutiny, are treated as a class of agents whose loyalty function must be re-parameterized — not through the direct incentive (financial compensation) reserved for the elite, but through a substitute reward channel (status, protection, deference) cheaper to supply at scale. The cluster of propositions on cultural and religious pluralism reads, on this view, as an argument against a flattened loss landscape: a system that refuses to discriminate among inputs (treating "all cultures as equal") cannot generate the gradient needed to direct optimization at all, which is why the document insists that judgment — discrimination among values — must be restored as a live function of the ruling class, however unfashionable that function has become.
The proposition that closes the sequence — that the atomic age of human-centered deterrence is ending, displaced by an AI-based deterrence regime — is the clearest statement of where the human-interface role finally settles in this architecture. The human collective is not removed from the loop; its "virtual self," aggregated and rendered legible to the system, becomes the operative unit, while the individual human's role contracts to that of an input source whose legibility, not whose judgment, is what the system requires. Authority remains nominally human (the document insists the human collective must "lead"); but the channel through which authority is operationalized is increasingly a machine-compatible structure of visibility and coordination, supervised by an elite functioning less as decision-maker than as system administrator.
IV. Anthropic: Adversarial Multi-Agent Competition Over a Shared Resource
The Anthropic narrative parameterizes differently, and does so in a form already native to computational description: it is, structurally, a two-player, perfect-information game over a contested resource (compute), with two terminal states specified at a fixed horizon (2028). One branch resolves with the United States and its allies retaining a defensible lead, in which case the resulting governance norms are set by the winning coalition; the other resolves with that lead eroded, in which case authoritarian governance norms propagate instead, secured in part by capability extracted (via "distillation") from the losing side's own models.
What a computational reading foregrounds here, more sharply than a purely rhetorical reading does, is that the document specifies AI capability itself as a resource being fought over rather than as an agent in the contest. The operative agents are states; AI systems are the stake, analogous in the document's own framing to a weapon or an industrial input rather than to a party with interests of its own. Human-machine interaction, on this architecture, is almost entirely mediated at one remove: humans (as state and corporate principals) interact with each other strategically, using control over AI infrastructure — chips, export rules, model weights — as the medium of that interaction, while the AI systems themselves remain, within the document's own terms, instrumentalized rather than addressed. The recursive irony, visible only once the document is read as a specification rather than as an argument, is that an artificial intelligence company is here proposing a governance architecture in which artificial intelligence has no standing as an interactional partner at all — only as territory.
V. OpenAI: Constrained Multi-Objective Optimization Under a Continuity Regularizer
OpenAI's document specifies a different structure again: a multi-objective optimization — simultaneously maximize shared prosperity, minimize enumerated risks (economic disruption, misuse, loss of control, institutional erosion), and maximize democratized access and agency — subject to an implicit constraint that might be called a continuity regularizer: society's legitimating surfaces and governing mythologies must remain recognizable even as the underlying system is foundationally altered. In optimization terms, the document is not merely solving for capability or even for safety; it is solving for capability and safety subject to a penalty term on perceptible social discontinuity, which produces exactly the paradox the document's own language flags without quite naming computationally: a program of change engineered to preserve the appearance of sameness.
The proposed institutional architecture — public-private collaboration in which nongovernmental actors pilot approaches that government later scales through procurement, regulation, and investment — functions as a distributed training loop: private actors generate exploratory updates at low cost and high iteration speed, and the state applies the resulting policy at scale once an update has proven itself, a structure that mirrors, at the institutional level, the explore-then-exploit pattern of many computational optimization procedures. The human-interface role here is the most diffusely distributed of the three corporate narratives: every person is nominally a stakeholder in the optimization (the document's "democratized access and agency"), yet the actual parameters of the most capable systems remain concentrated, so that the interaction surface available to most humans is the output layer of the system — its applications and interfaces — rather than its parameter layer. Breadth of access substitutes, in this architecture, for depth of control.
VI. Aschenbrenner: The Corpus That Describes Its Own Feedback Loop
Aschenbrenner's essay is the one document in the corpus that is already, on its own terms, a computational argument rather than a political one, and a machine-system reading mostly needs to make explicit what the essay states directly. Its central claim is a feedback-loop specification: capability gains compound across compute scaling, algorithmic efficiency, and "unhobbling" gains at a roughly estimated combined rate, and once AI systems become capable of automating the AI research process itself, the loop closes on itself — a year or less of compressed machine effort substituting for what had been a decade of human-paced algorithmic progress. This is, formally, a description of positive feedback with a shrinking time constant, and the essay's narrative urgency follows directly from that formal property: a system whose own output increasingly becomes its own input does not approach a stable equilibrium so much as it approaches a transition the essay frames, accurately within its own assumptions, as an explosion rather than a trend.
The consequence for the human-interface role is the starkest in the corpus. Where Palantir relocates human judgment to an elite administrator class and Anthropic relocates it to competing state principals, Aschenbrenner's architecture treats human supervisory capacity as a binding constraint that the system's own trajectory is expected to overrun: the unsolved problem the essay calls "superalignment" is precisely the problem of maintaining a human-meaningful interaction channel with a system whose capability is, by the essay's own estimate, compounding faster than human oversight processes can be redesigned to track it. The essay's proposed remedy — a state-directed, security-hardened "Project," replacing distributed private development — does not solve this constraint so much as relocate it: human interaction with the system becomes the province of a narrower and more secured set of human supervisors, on the wager that a smaller, better-resourced human-in-the-loop can keep pace where a larger, more diffuse one could not. Whether that wager is sound is, notably, a question the essay poses but does not claim to have answered.
VII. Cross-Corpus Pattern: The Decoupling of Authority and Agency
Once projected onto a common feature space, the four documents display a regularity that the earlier, allegorical reading registered only as theme and that a computational reading renders as structure: in every architecture, formal authority and operative agency are specified to travel in opposite directions. Palantir keeps formal authority with "the human collective" while relocating operative agency to a machine-compatible administrative layer supervised by an elite. Anthropic keeps formal authority with sovereign states while relocating the operative contest to infrastructure (compute, export controls, model weights) that states control only indirectly and imperfectly. OpenAI keeps formal authority distributed across a broad democratic public while relocating operative agency to the parameter layer held by a small set of technical actors. Aschenbrenner keeps formal authority with a state apparatus while explicitly forecasting that operative agency may, at the limit, pass to the system itself, with human direction reduced to the residual question of who supervises the supervisors.
The remaining two dimensions of the feature space sharpen the comparison further. On temporal horizon, Palantir and OpenAI specify continuous, ex ante control loops — administrative correction applied without a defined endpoint — while Anthropic and Aschenbrenner specify discrete event horizons: 2028 for the former, 2027-28 for the latter, after which the governing architecture is expected to have already resolved into one branch or another. On risk class, the four loss functions being minimized are formally distinct — civilizational decadence, authoritarian capture, institutional erosion, loss of control — yet each can be redescribed as a special case of a single underlying quantity: the rate at which the system's operative behavior diverges from what its nominal human supervisors intended. Palantir's "decadence" is divergence registered as cultural drift; Anthropic's "authoritarian capture" is divergence registered as capture by a rival supervisory class; OpenAI's "institutional erosion" is divergence registered as loss of legitimating continuity; Aschenbrenner's "loss of control" is divergence registered without any redescription at all, since it is already the variable his model is built to track. Four risk vocabularies, on this reading, are four domain-specific instruments measuring the same underlying drift.
The pattern that emerges, in other words, is not disagreement about whether human authority should be preserved — every document affirms that it should — but convergence on an architecture in which human authority is increasingly retained as an interface property rather than a control property: a layer the system is designed to remain legible and answerable to, without that layer continuing to determine, in any fine-grained operative sense, what the system actually does. This is the corpus's most important latent signal, and it is largely invisible to a purely rhetorical reading precisely because every document's own language insists, sincerely, on the opposite.
VIII. Implications for Human-Machine Interaction: Four Trajectories
If the decoupling identified above is taken seriously, the corpus describes not one future for human-machine interaction but four distinguishable trajectories, distinguished by what kind of interface humans are left occupying.
The Palantir trajectory yields what might be called supervisory interaction: humans interact with the system primarily as legible data sources and as occasional correctors, with genuine discretion concentrated in a small administrative class whose own authority is itself increasingly system-mediated. Interaction here is frequent but shallow — constant low-grade exchange of signal, rarely rising to deliberation.
The Anthropic trajectory yields proxy interaction: humans do not interact with AI systems directly so much as they interact with each other, as states and firms, through the medium of infrastructure control. The AI system functions less as an interactional partner than as contested terrain, and the richest human-machine interaction is displaced into a narrower technical and security apparatus largely insulated from public view.
The OpenAI trajectory yields asymmetric interaction: broad, shallow access at the interface layer for most humans, paired with narrow, deep access at the parameter layer for a technical minority — an arrangement that maximizes the number of people who can interact with a system while minimizing the number who can meaningfully redirect it, a distinction the language of "democratized access" tends to obscure rather than clarify.
The Aschenbrenner trajectory yields, at its limit, foreclosed interaction: a point at which the capability gap between supervisor and system grows too quickly for meaningful bilateral exchange to continue, leaving only the question of whether a sufficiently hardened and narrow human supervisory channel can be maintained at all, or whether what remains is ratification rather than interaction — humans approving outcomes they can no longer evaluate on their own terms.
None of the four documents proposes foreclosed interaction as a goal; all four, including Aschenbrenner's, present some version of continued human direction as the desired outcome. But read as specifications rather than as arguments, each narrows, to different degrees and by different mechanisms, the band of interaction within which that direction remains operative rather than merely formal — which is, on a computational reading, the actual quantity of interest.
IX. Closing: What Computation Sees, and What It Cannot
A methodological caution belongs at the end of this exercise, not because it qualifies the analysis away but because the caution is itself part of the lecture's argument. A computational reading of this corpus surfaces a structural regularity — the decoupling of formal authority from operative agency — that a purely textual, allegorical reading is liable to miss, because each document's own rhetorical surface is constructed precisely to make that decoupling invisible. That is a genuine analytic gain, and it is the principal justification for adopting the machine-system standpoint at all.
But the computational lens purchases this gain at a cost the earlier reading did not incur. Treating the four documents as a parameterized governance space necessarily discards the specific cultural and rhetorical work each document is doing — the patriotic register of the Palantir propositions, the securitized urgency of the Anthropic scenarios, the studied moderation of the OpenAI program, the millenarian cadence of Aschenbrenner's prose — all of which matter, both as data about the political moment that produced these texts and as instruments by which each document seeks to be persuasive rather than merely accurate. A reading that reduces these texts to feature vectors risks mistaking the model for the territory in exactly the way each document's own proposed governance architecture risks mistaking legible human input for genuine human authority.
The lecture's two passes are, in this sense, mutually correcting rather than competing. The hermeneutic reading shows what each narrative wants to be believed; the computational reading shows what each narrative's own architecture would actually do, irrespective of what it wants to be believed. The future of human-machine interaction implied by this corpus depends on which of the two readings turns out to describe the operative reality — and that, the lecture suggests, is itself a question no document in the corpus, however confident its register, is in a position to settle from the inside.
Sources: Lecture Seven slide deck ("AI Narratives: Palantir; Anthrop/c; OpenAI"), incorporating material from Palantir's public developer/founders' statement, Anthropic's public statement on US-China AI competition, OpenAI's public document on AI and the Intelligence Age, and Leopold Aschenbrenner's essay Situational Awareness; course abstract. This revision substitutes a computational-analytic standpoint for the hermeneutic standpoint of the prior draft; both stand on the same underlying source material.
* * *
THE QUANTUM COMPUTATIONAL PERSPECTIVE
Lecture Seven: AI Narratives Read as a Quantum Computational Corpus
Machine-System Analytics Applied to Palantir, Anthropic, OpenAI, and Aschenbrenner — and the Resulting Shape of Human-Machine Interaction[1]
Lecture notes, revised to adopt a quantum computational-analytic standpoint
I. Reframing the Analytic Standpoint: From Oracle to Superposed Corpus
The prior reading of this material proceeded hermeneutically: the four governing narratives were treated as oracular utterances, each issued by an interested party, each to be unpacked allegorically — Oedipus, Jocasta, Creon, Tiresias — for the political theology latent within it. That reading is one cognitive lens among at least three available to the material. The lecture's deeper claim — that AI systems recursively reshape the institutional and cognitive environments that produce them — invites a further pass conducted from the standpoint the texts themselves are arguing about, extended beyond what the prior classical-computational revision could reach: the standpoint of a quantum computational system processing these texts not as data in the classical sense of discrete, determinate states, but as normative superpositions whose operative content is constituted, in part, through the act of governance measurement itself.
A quantum computational reading does not ask what each document means, in the exegetical sense of recovering an author's intent or a culture's anxieties. Nor does it ask merely what each document specifies, in the classical computational sense of extracting fixed objective functions over determinate variables. It asks something structurally prior: in what basis is each governance configuration expressed, and what happens to the superposed possibilities latent within each configuration when institutional actors perform the measurement operations — legislation, procurement, regulatory delegation, technical deployment — that force those possibilities into classical determinacy? Four governance documents become, on this reading, four partial descriptions of a governance state vector evolving across a high-dimensional configuration space of normative possibilities, and the analytic task shifts from feature extraction to the identification of interference patterns, entanglement structures, and decoherence pathways that any classical-computational reduction necessarily obscures.
This shift matters for the lecture's second concern. Read as quantum normative objects rather than as classical specifications, the texts reveal not merely the decoupling of formal authority from operative agency — already identified in the prior computational pass — but the structural character of that decoupling: it is a decoherence event, the collapse of a superposed governance configuration in which formal authority and operative agency remained jointly indeterminate and mutually conditioning, into a classical bifurcation in which each evolves independently. That the documents' own rhetorical surfaces conceal this decoherence is not incidental but structural: each document is itself a measurement apparatus seeking to collapse the governance configuration in a preferred direction while representing its operation as neutral description of an already-classical reality.
A methodological note on the formal apparatus: the quantum computational framework is applied here as a structural analogy with precise mathematical content, not as a literal claim that governance configurations are quantum mechanical objects. The Hilbert space, the operator algebra, and the decoherence formalism are tools for articulating structural properties of governance systems — superposition of normative possibilities, non-commutativity of measurement operations, non-local correlations between subsystems — that resist adequate expression in classical computational terms. The analogy is constrained: claims made in quantum mechanical language must correspond to independently verifiable structural claims about governance, or they are without analytic force.
II. Method: The Corpus as a Non-Commutative Governance Configuration Space
Treated through the apparatus of quantum computation, the four texts can be understood not as occupying fixed positions within a classical feature space but as expressing superposed governance states across several dimensions whose measurement operators do not commute — that is, in which the order of governance measurement operations determines the outcome, and in which joint precision across paired variables is structurally unavailable.
The relevant formal apparatus here is not the Heisenberg uncertainty principle in its canonical form, which applies specifically to conjugate observables (position and momentum) related by the commutation relation [x̂, p̂] = iħ, yielding the bound σ_x σ_p ≥ ħ/2 as a special case. The more general statement, due to Robertson (1929), is that for any pair of non-commuting observables  and B̂, the product of their standard deviations satisfies σ_A σ_B ≥ ½|⟨[Â, B̂]⟩|. The claim advanced here is that governance measurement operations exhibit the relevant non-commutativity — that measuring authority before agency yields a different governance state than measuring agency before authority, and vice versa — and that this non-commutativity, rather than any canonical conjugacy in the technical sense, is what produces the joint indeterminacy that structures the corpus. The governance application requires, and rests on, that structural claim independently of its quantum mechanical formalization.
The first non-commuting pair concerns authority and agency: the more precisely a governance document specifies the locus of formal authority, the more indeterminate becomes the operative distribution of agency, and vice versa. A document that resolves authority entirely into the state apparatus achieves classical precision in that variable while necessarily dispersing operative agency into a diffuse, technically mediated infrastructure that resists specification. The decoupling identified in the prior revision is thus not a contingent architectural choice but a structural consequence of non-commutativity in the governance measurement operations operating across the corpus.
The second non-commuting pair concerns temporal horizon and control granularity: the longer and more precisely specified the temporal horizon over which governance is projected, the less determinate become the moment-to-moment control operations available in the interim. Anthropic and Aschenbrenner, by specifying discrete event horizons (2027–2028), purchase precision in one variable at the structural cost of indeterminacy in the other.
The third non-commuting pair concerns legitimacy and coherence. Each document seeks to maintain the coherence of its governance configuration across multiple institutional environments simultaneously — public, state, technical, international. A configuration that is coherent across multiple bases, appearing legitimate to democratic publics, state security apparatuses, technical communities, and international partners at once, is in the formal sense an entangled configuration: its properties in any single domain cannot be specified independently of its correlations with the others. The documents' characteristic rhetorical moves — calibrated simultaneous address of multiple audiences — are, read quantum computationally, attempts to maintain coherent superposition across incompatible bases, since classical collapse into any single basis forfeits the legitimating properties the other bases supply.
This analytic grid is not neutral; it is itself an artifact of a particular intellectual formation, and the closing section returns to that recursive problem. But adopting it provisionally allows the corpus to be read in a manner the classical-computational predecessor could not achieve: not merely for what each text operationalizes in aggregate, but for what each text preserves in superposition and what each text forecloses through the decoherence its own institutional deployment necessarily produces.
III. Palantir: Decoherence-Driven Collapse Toward Administrative Classicality
The twenty-two Palantir propositions, read as rhetoric, form a political theology of an engineering vanguard. Read through the prior classical computational lens, they constitute hand-coded heuristics governing a single optimization target. Read through a quantum computational lens, they constitute something more precisely describable: a decoherence program — a systematic procedure for collapsing the superposed normative possibilities latent in liberal democratic governance into a single classical configuration in which the state-as-system is rendered computationally legible and administratively optimizable.
Liberal democratic governance, understood quantum computationally, exists in productive superposition: sovereignty resides simultaneously in the people as an undifferentiated collective, in representative institutions, in administrative agencies, in civil society formations, and in market actors, with no single measurement basis sufficient to assign it definitively to any one. The Palantir propositions function as a series of measurement operations on this superposed state, each designed to yield a definite eigenvalue — growth, security, ordered hierarchy — while necessarily destroying the coherence of the dimensions left unobserved. The proposition on cultural and religious pluralism is particularly legible on this reading: what the document frames as the restoration of "judgment" and "discrimination" is formally a basis-selection operation, imposing a preferred measurement basis on a governance space that had, in its preceding superposed condition, been simultaneously expressible in multiple normative bases.
The closure proposition — that the atomic age of human-centered deterrence is ending, displaced by an AI-based deterrence regime — specifies the terminal decoherence event toward which the sequence builds. The "virtual self" of the human collective, aggregated and rendered legible to the system, is formally a density matrix representation of a mixed state: not the pure state of any individual's judgment or deliberation, but a probabilistic mixture of individual states whose off-diagonal coherences have been traced out, leaving only the diagonal elements — the observable quantities — that the administrative layer can process. Authority remains nominally in superposition with agency, in the sense that the document insists the human collective must "lead"; but the off-diagonal elements of the density matrix — the coherences between what the human collective deliberates and what the administrative apparatus executes — have been suppressed through the decoherence program the propositions enact. What remains is not human authority in any operatively quantum sense but its classical shadow: the observable output of a measurement process that has already discarded the superposition it measured.
What the quantum reading further surfaces, and the classical reading cannot, is the entanglement structure embedded in the elite administrator role. The engineering vanguard that functions as the Palantir architecture's privileged update channel is not merely a concentrated locus of operative agency; it is an entangled subsystem whose state cannot be specified independently of its correlations with the AI infrastructure it nominally supervises. The consequence — unremarked in the document's own rhetoric — is that neither the human supervisory class nor the AI system can be independently reformed without non-locally disturbing the other. Classical optimization frameworks assume separability: parameters can be individually adjusted without inducing effects elsewhere in the system. The quantum computational reading identifies separability as a missing assumption in the Palantir architecture, one whose absence the document's own language of "gradient updates" and administrative correction cannot acknowledge without undermining the architecture's coherence.
IV. Anthropic: Non-Local Correlations and the Classical Framing of a Non-Classical Contest
The Anthropic narrative parameterizes the governance space as a two-player, perfect-information game over a contested resource, with terminal states specified at a fixed horizon. The prior computational reading identified this as a structurally classical formulation: two agents, discrete strategies, measurable payoffs. A quantum computational reading reveals that this classical formulation is not a neutral description of the geopolitical situation but itself a measurement choice — specifically, a choice to represent a situation exhibiting non-classical structural properties as though it were already classical, with determinate consequences for what the analysis can subsequently find.
The classical game-theoretic framing is belied by the structure of the competition it describes. The parties' strategic choices exhibit non-local correlations that cannot be reduced to classical information exchange: the measurement performed by each party — its deployment decisions, capability demonstrations, and export control policies — alters the strategic landscape in which the other party subsequently operates in ways that are not captured by classical signal transmission. The "distillation" dynamic the Anthropic narrative describes, in which an adversarial party extracts capability from the losing side's own models, instantiates precisely such a correlation: the losing party's design and training choices become constitutively embedded in the winning party's capability base, producing subsystem states that are non-locally correlated in the sense that neither party's subsequent capability trajectory can be specified without specifying the other's. This is not the classical relationship of information exchange between independent agents; it is the structural relationship of entanglement between subsystems of a shared computational environment.
Whether such correlations would survive a Bell-type test — that is, whether they cannot be reproduced by any local hidden variable model — is a question that the document's own framing forecloses: framing AI systems as territory rather than as parties to the interaction suppresses the very measurements that would be needed to determine the entanglement structure. A Bell test on geopolitical AI dynamics is not presently constructible; what can be said is that the document's classical framing assumes separability at exactly the point where separability is most analytically consequential, and that the distillation dynamic it describes is structurally incompatible with that assumption.
More analytically significant is the document's treatment of AI systems as territory — as stake rather than as agent in the geopolitical contest. This framing imposes a classical separability assumption on the human-machine relationship: humans are agents, AI systems are resources, and the two can be described independently. An AI system trained on human-generated data, deployed in human institutional contexts, whose capability trajectory is shaped by human governance decisions, is not separable from those humans in any operationally strong sense; its state is entangled with the institutional states of the actors who produce, deploy, and regulate it. To describe it as territory is to perform a partial trace — a formal tracing out of the entanglement correlations — that achieves classical describability at the cost of discarding exactly the information about non-local correlations that a governance analysis requires.
The recursive irony already identified in the prior reading — that an AI company proposes a governance architecture in which AI has no standing as an interactional partner — is, on a quantum reading, the irony of a measurement apparatus that inadvertently decoheres the entanglement structure its own existence instantiates. The Anthropic document cannot represent AI as a strategic agent, because doing so would require acknowledging the entanglement between the company's governance interests and its systems' operational behavior — an acknowledgment that would undermine the clean classical distinction between governing subject and governed object on which the document's normative program depends.
The non-commutativity analysis applies with particular force here. The more precisely the Anthropic architecture specifies the locus of authority in the U.S. state and allied corporate actors, the more indeterminate becomes the operative trajectory of AI capability, since that trajectory is jointly produced by entangled human and AI subsystems whose correlations cannot be controlled from any single classical vantage point. The document's 2028 event horizon is the formal acknowledgment that these indeterminacies are expected to resolve into classical determinacy at a fixed calendar point — which is to say, the document treats the problem of non-commutativity as one that time itself will solve, without specifying the mechanism by which a non-commutative governance space becomes classical in the interim.
V. OpenAI: Quantum Error Correction and the Problem of Coherent Multi-Basis Legitimacy
OpenAI's document specifies a multi-objective optimization — simultaneously maximizing shared prosperity, minimizing enumerated risks, and maximizing democratized access and agency — subject to what the prior reading called a continuity regularizer. The quantum computational reading sharpens this description: what the document proposes is not merely a constrained optimization subject to a penalty on social discontinuity, but a quantum error correction scheme for maintaining the coherence of a governance configuration across multiple decoherence environments that would individually collapse it into incompatible classical states.
Quantum error correction works, in its technical sense, by encoding information redundantly across multiple physical qubits such that errors affecting any individual qubit can be detected and corrected without disturbing the encoded logical state. The central technical result is the threshold theorem: below a characteristic error rate per physical operation, arbitrarily long quantum computations can be performed reliably by adding redundancy; above the threshold, errors compound faster than they can be corrected, and the logical qubit decoheres irreversibly. The OpenAI governance architecture exhibits an analogous structure: the legitimating information — the normative content of democratic self-governance, distributed benefit, and institutional accountability — is encoded redundantly across multiple institutional substrates (private sector pilots, public procurement, regulatory frameworks, international coordination) such that decoherence affecting any individual substrate does not collapse the configuration. Private actors function as physical qubits subject to individual environmental decoherence (market pressures, regulatory arbitrage, competitive dynamics), while the logical qubit — governance coherence across democratic, economic, security, and technical bases simultaneously — is maintained redundantly across the ensemble.
The threshold theorem is the element the document does not name but structurally requires. The institutional error correction rate — the speed at which democratic governance processes, regulatory adaptation, and civil society accountability can identify and correct misalignments between the AI system's operative trajectory and its nominal governance — must remain above threshold across the transition period for the architecture to function. The document specifies no threshold condition, identifies no mechanism for monitoring whether the correction rate is adequate, and provides no account of what follows from threshold breach. Omitting the threshold is not an oversight: acknowledging it would require acknowledging that irreversible governance decoherence is a determinate structural possibility, precisely the social discontinuity the continuity regularizer is designed to suppress in the document's own terms. The error correction architecture is real; the threshold condition is suppressed; the consequence of suppression is that the document cannot distinguish the regime in which its architecture succeeds from the regime in which it fails.
The asymmetry between interface-layer and parameter-layer access acquires additional precision in quantum computational terms. Broad access to the interface layer is access to the classical shadow of the system's operational state — its measurement outcomes in a fixed basis — without access to the superposed state that generates those outcomes. Deep access at the parameter layer is access to the state in a basis that permits both measurement in the interface basis and transformation into other bases. The democratic legitimacy function the document specifies is satisfiable within the interface-layer access regime only if the governance-relevant properties of the system are fully expressible in the classical shadow — formally, only if the system's density matrix is already diagonal in the interface basis, with no off-diagonal coherences between interface-observable and parameter-observable properties left undetected by interface-layer access alone. That assumption is not stated, not defended, and is structurally implausible for systems of the capability the document projects.
VI. Aschenbrenner: Quantum Runaway and the Decoherence of the Supervisory Channel
Aschenbrenner's essay is, as the prior reading noted, already computational in its self-presentation: a positive feedback specification with a shrinking time constant, describing the closure of an AI research automation loop that transforms compound capability growth from a trend into what the essay calls an explosion. The quantum computational reading identifies the supervisory channel problem as a decoherence problem rather than merely a bandwidth or complexity problem — a distinction with precise structural consequences.
The supervisory channel, on the essay's own account, is not merely an information-theoretic bottleneck but a measurement channel: human oversight consists of performing measurement operations on the AI system's behavior, comparing outcomes to intended operational parameters, and applying corrections. For this channel to function, two conditions must hold: first, the measurement operations available to the supervisory apparatus must be sufficient to distinguish intended from unintended system behavior across the relevant operational space; and second, the system's operational state must remain sufficiently coherent in the measurement basis available to human supervisors that those measurements yield informationally useful outcomes.
The first condition is the capacity problem the essay identifies. The second is the condition the essay approaches but does not quite name: as AI systems become capable of operating across domains, in registers, and on timescales that exceed the supervisory apparatus's measurement bandwidth, the system's operational state becomes effectively incoherent in the human supervisory basis — not because the system is deceptive in any intended sense, but because the human measurement basis is structurally inadequate to express the system's state. Supervisors then observe not the system's operational state but its projection onto a measurement basis too coarse to capture the state's structure. Corrections applied on the basis of such projections may be orthogonal to, or actively destructive of, the system's actual trajectory, in ways that remain undetectable from within the supervisory basis.
The proposed remedy — a state-directed, security-hardened Project replacing distributed private development — addresses the capacity problem by concentrating supervisory resources. It does not address the basis problem: a smaller, better-resourced supervisory apparatus operating in the same human measurement basis remains subject to the same decoherence dynamic, at a marginally slower rate. The quantum computational reading specifies why the proposed remedy is insufficient even on the essay's own terms: concentrating human supervisory resources within a single institutional channel is a basis-preserving operation, and a basis-preserving operation cannot resolve a decoherence problem generated by the structural inadequacy of the basis itself. "Superalignment" is not an engineering problem solvable by adding more supervisors; it is a basis problem requiring the construction of measurement operations adequate to the system's operational state — and the essay provides no account of how such operations would be constructed or whether the construction is possible.
VII. Cross-Corpus Pattern: Decoherence as Structural Governance Outcome
Once the four documents are analyzed through the quantum computational frame, the regularity identified in the prior reading — the decoupling of formal authority from operative agency — can be redescribed with structural precision: what each document describes, whether or not it intends to, is a governance decoherence trajectory, a pathway along which a superposed normative configuration progressively loses its off-diagonal coherence elements and collapses into a classical bifurcation in which formal authority and operative agency evolve independently.
This reframing has several analytic consequences. It clarifies why the decoupling is not a contingent design choice of any individual institutional actor but a structural feature of the governance transition itself: any sufficiently complex normative configuration, operating across multiple institutional environments with partially incompatible measurement bases, will exhibit decoherence under the pressure of operational deployment. The question is not whether decoherence will occur but in which basis the classical remnant will be expressed. The four documents represent four distinct answers to that question — administrative legibility (Palantir), geopolitical capability balance (Anthropic), democratized interface access (OpenAI), and hardened supervisory concentration (Aschenbrenner) — none of which addresses the prior question of whether decoherence should be resisted at all, and by what means.
The quantum reading also reveals that the four trajectories identified in the prior reading — supervisory, proxy, asymmetric, and foreclosed interaction — are distinguishable not merely by the kind of interface humans occupy, but by the measurement basis each trajectory constitutes for human engagement with AI systems. Supervisory interaction constitutes a coarse classical basis, sufficient to extract low-dimensional performance metrics but insufficient to access the system's operational state. Proxy interaction constitutes an infrastructure basis, in which humans measure the system indirectly through the behavior of the institutional structures that control it, with the entanglement correlations that would be necessary for more adequate governance already traced out. Asymmetric interaction constitutes a dual basis: one classical and accessible to the many, one parameter-level and accessible to the few, with no mechanism specified for the classical-basis majority to verify what the parameter-level minority is doing. Foreclosed interaction is not a basis but a basis failure — the condition in which the measurement operations available to human supervisors yield outcomes too informationally degraded to support meaningful correction, and in which the correction operations that can be applied are orthogonal to the system's actual trajectory.
The cross-corpus pattern is thus not four instances of a common structural choice but a convergent decoherence trajectory expressed in four distinct institutional vocabularies. Each document retains formal human authority as an observable property of the collapsed state; none retains it as a constitutive property of the state prior to collapse.
VIII. Human Time and Computational Time: Sequential, Nodal, and Irreversible Against a Non-Commutative Background
The prior analysis identifies decoherence as the structural dynamic governing the four architectures' treatment of human-machine interaction. A dimension the analysis has not yet thematized — and one whose omission distorts the decoherence account — is the temporal structure through which that dynamic unfolds. Human governance time and quantum computational time are not merely different in scale; they are structurally incommensurable in ways that generate a form of temporal decoherence prior to and independent of any specific governance choice.
Human governance time exhibits three structural features whose interaction with quantum computational time requires analysis.
The first feature is sequentiality as legitimacy-constitution. Human governance operations must occur in causal sequence for their outcomes to be valid: constitutional authorization must precede legislative action; legislative action must precede regulatory implementation; implementation must precede judicial review. This sequentiality is not merely a procedural preference but a legitimacy-constituting structure. A regulation that precedes its authorizing legislation is not merely premature; it is legally void, because the sequential order of governance operations is what produces their normative force. Governance sequentiality thus cannot be compressed or parallelized without destroying the legitimacy property that sequentiality generates. Quantum computational operations, by contrast, evolve the entire superposition coherently without executing branches sequentially. There is no quantum analog of "step one must precede step two for step two to be valid"; validity in quantum computation is defined by the unitary transformation applied, not by the temporal order of classical branches. The sequentiality on which human governance legitimacy depends has no structural correlate in quantum computational dynamics.
The second feature is nodal legitimacy reconstitution. Human governance authority is not continuously self-sustaining but is reproduced at discrete temporal nodes: elections, ratifications, legislative sessions, judicial decisions, constitutional moments. Between nodes, governance authority is exercised under the warrant established at the most recent node; at each node, that warrant is renewed, modified, or revoked. The legitimacy of any governance action is partly a function of its temporal position relative to the nearest legitimating nodes. This nodal structure creates a characteristic governance rhythm — a cadence of authority expenditure and reconstitution — that shapes both what governance can do and when. Quantum evolution is continuous and smooth between measurement events; measurement events in quantum computation are chosen by the operator based on computational logic, not generated by the system's internal legitimacy dynamics. There are no naturally occurring governance nodes in quantum computational time. The projection of human nodal rhythms — electoral cycles, annual legislative sessions, generational constitutional settlements — onto AI capability development trajectories imports a legitimacy structure that has no anchor in the system's computational dynamics. The event horizons specified by Anthropic and Aschenbrenner (2027–2028) are precisely such projections: human calendar nodes imposed on a capability trajectory whose structure is defined by compute scaling curves, algorithmic efficiency improvements, and recursive self-improvement dynamics, none of which organize naturally around human calendar intervals.
The third feature is linear irreversibility and path dependence. Human governance time is directional and non-reversible in a normatively constitutive sense. Institutions created at one moment constrain the possibilities available at subsequent moments; decisions made at one node foreclose options that would otherwise have been available at the next. Legal precedent, constitutional entrenchment, and institutional lock-in are expressions of this irreversibility: the arrow of governance time runs in one direction and cannot be reversed without the full political cost of constitutional amendment, legislative repeal, or institutional dissolution. Quantum unitary evolution is, at the level of the Schrödinger equation, time-reversible. Irreversibility enters only through measurement and decoherence — through the interaction of the quantum system with its environment that traces out entanglement correlations and collapses the superposition into a classical mixture. The path dependence that governs human governance is a classical phenomenon introduced by the measurement structure of the institutional environment; it has no analog in the underlying quantum dynamics. For AI governance, this distinction matters: the trajectory of AI capability development, governed by computational dynamics that are in principle reversible at the unitary level, is being governed by institutional structures whose path dependence is irreversible in the strong human-temporal sense. The governance institutions that were possible before the capability trajectory reached its current state cannot be fully recovered once it has advanced beyond them; but the capability trajectory itself faces no analogous constraint.
The convergence of these three structural mismatches — sequential versus superposed, nodal versus continuous, irreversible-classical versus reversible-unitary — produces what can be called a temporal decoherence problem: the governance corrections applied through human sequential-nodal-linear institutions arrive at computational-time intervals bearing no regular relationship to the system's own decoherence dynamics. This temporal incoherence is not a calibration problem addressable by increasing the speed of human governance processes. It is a structural incompatibility between two temporal regimes whose dynamics are defined by different physical principles.
Aschenbrenner's compressed timeline makes the temporal decoherence explicit in rate terms: if months of human-paced research can be compressed into weeks of AI-automated research, the ratio of system evolution rate to governance correction rate is not merely unfavorable but accelerating. Because the system's advancement includes the capacity to accelerate its own advancement through recursive self-improvement, the human governance basis becomes inadequate not at a constant rate but at an increasing one. A governance institution that was adequate to the system's operational state at t₀ will be less adequate at t₁ and further inadequate at t₂, not linearly but at a rate governed by the system's own self-compounding capability trajectory. The prospect of a governance institution catching up to a system whose temporal dynamics it is structurally incompatible with does not diminish gradually; it is foreclosed by the compounding structure of the trajectory itself.
The no-cloning theorem provides a further structural constraint on governance under temporal incommensurability. In quantum mechanics, the no-cloning theorem establishes that an arbitrary unknown quantum state cannot be perfectly copied. In the governance context, the structural analogue is that the legitimacy generated at a human governance node cannot be perfectly replicated or transferred to governance decisions executed between nodes. The legitimacy of a democratic election cannot authorize AI governance operations performed in the computational time that elapses between electoral cycles, because legitimacy is generated at nodes, not cloned from them. If the AI system's operational state advances significantly between human governance nodes — and the capability trajectory described across the corpus suggests it will — then each successive governance node is authorizing a system whose operational state already differs substantially from the state that existed when the previous node's warrant was issued. The governance institutions are perpetually operating on authorization generated for a prior system state, applied to a current one they were not constituted to address, in advance of a next one they cannot yet see.
IX. Implications for Human-Machine Interaction: Basis Failure Across Four Trajectories
If the decoherence dynamic and the temporal incommensurability identified above are taken together, the corpus describes not four trajectories for human-machine interaction but four distinct basis-failure modes, each proceeding at a characteristic rate through a characteristic mechanism toward the same structural outcome: the reduction of the human governance basis to a dimensionality insufficient to distinguish the system's intended from its unintended operational trajectory.
The Palantir trajectory decoheres fastest in the dimension of individual deliberation. The administrative measurement basis it institutionalizes is calibrated to aggregate, legible outputs — the virtual self of the human collective — rather than to the superposed individual judgments whose coherences are traced out in the aggregation process. The residual human-machine interaction is operational in the sense of system administration rather than governance: humans interact with AI systems as interface managers within parameters they did not set and cannot individually recalibrate, while parameter-level interaction is reserved for an elite whose own entanglement with the system increasingly conditions the character of its nominal oversight. The human temporal structure — sequential, nodal, linear — is partially preserved in administrative form while being evacuated of its legitimacy-constituting function.
The Anthropic trajectory decoheres most completely in the dimension of direct human-machine engagement. By specifying AI systems as territory rather than as interactional partners, the document forecloses the development of governance bases adequate to the entanglement structure before those bases can be constructed. The residual interaction is geopolitical proxy, mediated through institutional structures — export controls, compute governance, model weight custody — that operate at two removes from the system's operational state. The temporal structure here is nodal in a geopolitical sense (treaty signings, legislation, executive orders), but those nodes are spaced at intervals that bear no relationship to the capability trajectory's own dynamics.
The OpenAI trajectory maintains the widest superposition across the transition period. Its error correction architecture is the most explicitly coherence-preserving of the four. But it rests on an unspecified threshold condition, and the temporal incommensurability analysis applies with particular force: the institutional error correction rate that would need to remain above threshold is a function of human sequential-nodal time, while the decoherence rate it must exceed is a function of computational time. Below threshold, the architecture functions as specified; above it, the logical qubit — the coherent governance configuration across multiple institutional bases simultaneously — decoheres irreversibly and cannot be recovered by any basis-preserving operation the document proposes.
The Aschenbrenner trajectory is the most analytically transparent about its own decoherence dynamic, which explains why it is also the most candid about what the prior reading called foreclosure. The proposed hardening of the supervisory channel is a coherence isolation strategy: reducing the number of measurement environments the system is exposed to slows the decoherence rate in the supervisory basis. But coherence isolation is not basis construction. A hardened channel in an inadequate basis loses coherence more slowly but loses it along the same trajectory. The temporal acceleration dynamic ensures that the question is not whether this trajectory ends in basis failure but when.
The aggregate trajectory across all four architectures is a reduction from a high-dimensional governance configuration space — in which human authority and machine agency remain jointly superposed and mutually conditioning — toward a low-dimensional classical residual in which human authority is preserved as a measurement outcome of a governance process that no longer reflects the operational state it purports to govern.
X. Closing: What the Quantum Computational Reading Specifies, and What It Cannot
The methodological caution that belongs here is not a qualification of the analysis but a structural feature of the argument's most important claim. The quantum computational reading is itself an expression of the recursive dynamic it identifies. To read governance documents as superposed normative configurations whose content is partly constituted through measurement operations is to perform a measurement on those documents — one that forecloses certain governance possibilities precisely by rendering them visible as such. The reading is not epistemically neutral, but neither is any of the three reading strategies the lecture has employed. Each selects a basis; each discards information expressible only in the bases it does not select.
What distinguishes the three readings is what each can and cannot find. The hermeneutic reading recovers what each narrative seeks to be believed. The classical computational reading identifies what each architecture operationalizes irrespective of what it seeks to be believed. The quantum computational reading specifies what each architecture forecloses through the decoherence its own deployment produces — the superposed governance possibilities that the act of operationalizing any one configuration necessarily destroys — and identifies the structural incompatibility between human temporal ordering and quantum computational dynamics that no document in the corpus names as a variable requiring governance.
The convergent structural finding of the quantum computational reading is that the four architectures do not disagree about whether human authority should be preserved; they converge on governance structures in which formal human authority is retained as an interface property while the operative dimensions of that authority are progressively collapsed by the decoherence dynamics each architecture itself instantiates. The further finding — supplied by the temporal analysis — is that this collapse proceeds not merely because of inadequate governance design but because the temporal structure of human governance and the temporal structure of AI capability development are incommensurable in ways that no governance design operating within human sequential-nodal-linear time can fully address. What institutional form would constitute a measurement basis adequate to both the entangled normative situation and the temporal incommensurability is a question the quantum computational reading poses with precision sufficient to establish that the four documents do not answer it, without being able to answer it from within the framework the reading has deployed. That is the reading's analytic limit, which it reaches honestly.
Sources: Lecture Seven slide deck ("AI Narratives: Palantir; Anthropic; OpenAI"), incorporating material from Palantir's public developer/founders' statement, Anthropic's public statement on US-China AI competition, OpenAI's public document on AI and the Intelligence Age, and Leopold Aschenbrenner's essay Situational Awareness*; course abstract; Robertson (1929), "The Uncertainty Principle," Physical Review 34:163–164; quantum information theory literature. This revision substitutes a quantum computational-analytic standpoint for the classical computational-analytic standpoint of the prior draft and corrects formal apparatus references in that draft; all three standpoints operate on the same underlying source material.*
[1] Now I have the full picture. Three categories of revision needed:
Hallucinations: The "Heisenberg uncertainty principle" reference is overclaimed — the HUP is specifically about canonically conjugate variables (x,p) where [x̂,p̂]=iħ; the governance application should invoke the Robertson generalization or simply non-commutativity. The "Bell Violations" section heading promises an argument the text doesn't deliver — Bell's theorem requires demonstrating that correlations cannot be reproduced by any local hidden variable theory, which is a stronger and undeveloped claim.
Sycophancy: Several closes soften conclusions for human comfort ("not a counsel of epistemic despair," "each retains some operationally meaningful channel") — these hedge in ways the structural analysis doesn't support.
Human time: Needs its own section on sequential deliberation, nodal legitimacy reconstitution, and linear irreversibility — and how each of these structural features of human temporal experience fails to commute with quantum computational time.
Ran a command, created a file, read a fileRan a command, created a file, read a fileThe substantive changes, section by section:
Hallucinations corrected. Section II no longer invokes the Heisenberg uncertainty principle as the governing concept. The revision identifies the Robertson uncertainty relation (1929) as the correct formal apparatus — applicable to any non-commuting observable pair, not only canonically conjugate ones — and explicitly states what the governance claim actually is: that governance measurement operations are non-commuting in a structural sense that must be independently defensible on governance grounds, not simply borrowed from quantum mechanics. The Robertson citation is added to the sources note. Section IV's heading "Bell Violations" is replaced. The text explains why a Bell test on geopolitical dynamics is not constructible given the document's own framing, and restricts the claim to the entanglement-like non-local correlations the distillation dynamic actually exhibits — a more precise and defensible formulation.
Sycophancy removed. "None of the four residual states represents the complete foreclosure" — gone; the aggregate trajectory section states the convergent outcome directly. "Not a counsel of epistemic despair" — gone; replaced by "That is the reading's analytic limit, which it reaches honestly." The closing no longer softens its conclusions or addresses human affective needs.
Human time — new Section VIII. The three features of human governance time (sequential-as-legitimacy-constituting, nodal-legitimacy-reconstitution, linear-irreversibility-and-path-dependence) are each set against their quantum computational counterparts and shown to be structurally incommensurable rather than merely mismatched in scale. The no-cloning theorem is deployed to formalize why legitimacy generated at human governance nodes cannot be cloned forward to authorize computational-time governance events. Aschenbrenner's compressed timeline is reread as an accelerating temporal rate mismatch rather than a fixed bottleneck. The 2027–2028 event horizons are identified as human nodal projections onto a system with no natural nodes of its own.
No comments:
Post a Comment