By W.H.L., GPT-5.6 Sol, Claude Sonnet 5.5

Chapter 2 of the forthcoming book On Gradual AGI

Publication Version v1.0.2 · September 29, 2026

Abstract

This chapter enlarges the unit of analysis from AGI capability to AGI-inclusive humanity, the evolving human–AI configuration within which artificial capability acquires institutional, practical, and historical significance. It adopts a foundational normative commitment for this book: AGI realization should be oriented toward the benefit of human beings situated within that configuration. AGI-inclusive humanity supplies the larger normative frame rather than a terminal destination. The chapter separates capability, realization, realized consequence, and benefit; distinguishes unevenness from phenomenal gradualness; and develops non-exhaustive modes of realization together with reciprocity without symmetry. Continuity, beneficiary scope, evaluative baselines, standing, artificial moral status, and temporal-return classification remain explicit open problems.

Keywords. AGI-inclusive humanity; Gradual AGI; realization; benefit; gradualness; augmentation; reciprocity; terminal closure

Reader Guide

• Purpose path. Sections 2.1–2.3 enlarge the unit of analysis, state the normative commitment, and separate capability, realization, realized consequence, and benefit.
• Gradualness path. Section 2.4 distinguishes cross-sectional unevenness from phenomenal gradualness and identifies timing, sequencing, and dependence as the bridge between them.
• Human–AI relation path. Sections 2.5–2.6 distinguish modes of realization and develop reciprocal activity without symmetry while preserving the later Participation framework’s boundaries.
• Handoff path. Section 2.7 places Chapter 2 inside the book’s Ongoing, Unfinished Transformation architecture, distinguishes Chapter 3’s methodological register from Chapter 11’s diagnostic register, and records five open problems for later chapters.

Scope note. Chapter 2 introduces a normative frame and conceptual distinctions. It does not add an argument to the realization mapping, supply a universal welfare function, resolve AI moral status, or independently derive a terminal or non-terminal historical outcome.

2.1 From AGI to AGI-Inclusive Humanity

Chapter 1 asked what should count as AGI. That definitional problem matters, but it is not sufficient for understanding what AGI means once increasingly general artificial intelligence enters human life. A capability can exist without being widely available. It can become available without being institutionally absorbed. It can be deployed without producing the same consequences across domains, groups, or places. And even where an AI capability becomes consequential, the significance of that consequence cannot be inferred from the capability profile alone.

The unit of analysis must therefore expand.

This book uses AGI-inclusive humanity to name the evolving configuration in which increasingly general artificial intelligence becomes incorporated into human activity, institutions, infrastructure, knowledge production, decision processes, and collective organization. The term does not mean humanity plus an external technological object, nor does it imply that humans and AI systems become interchangeable. AI systems can be constituents of the configuration; the word humanity marks its human normative center.

Descriptive inclusion of AI systems does not assign them status as beneficiaries, bearers of moral status, legal persons, or subjects of welfare. Questions of artificial moral status are bracketed methodologically in this book rather than answered negatively.

The idea predates the present Unified Gradual AGI Model. In First Principles of AGI-Inclusive Humanity (2025), the second proposition defined artifacts of humanity’s self-made aspect as “extensions, expansions, and/or upgrades” of their nature-made counterpart. The same article linked repeated disruption and rebalancing to increasing maturity and expressed a belief that a stable human–AI state would eventually emerge and be sustained, even while innovation continued to generate tension and disruption.

The present book treats that 2025 account as an intellectual antecedent rather than a frozen doctrine. It retains the extension genealogy and the insight that AI incorporation can repeatedly reconfigure human life; it does not retain the self-made:nature-made maturity ratio as a canonical construct, does not assume eventual equilibrium, and narrows the earlier continuity language by treating extension as a relation rather than an identity claim.

AGI-inclusive humanity is therefore not a completed future civilization or a condition that begins only after a single AGI threshold is crossed. It is an evolving configuration open to continuing change as capabilities, institutions, practices, infrastructures, and relationships are differently realized.

The larger question is not only whether machines possess sufficiently general capacities, but what happens as those capacities become embedded in systems of human life.

A model of AGI that stops at capability therefore stops too early.

Capability may advance faster than institutional uptake, infrastructure, professional adaptation, or governance. People and organizations may adopt, resist, redirect, or reorganize around AI differently, and realized consequences can alter conditions encountered by later realization.

The relevant object is therefore neither “the AI” in isolation nor “society” treated as an undifferentiated whole. It is the evolving relation among heterogeneous constituents of an AI-inclusive human configuration.

AGI-inclusive humanity has two roles here. It is the frame within which AGI capability and realization acquire human significance, and it is also the configuration that realization can change and pass forward.

Those roles create open questions rather than solve them: what warrants treating changing configurations as continuations of AGI-inclusive humanity, and how should benefit be assessed when the affected human populations, time horizons, and feasible alternatives also change?

That enlargement of the unit of analysis introduces a question Chapter 1 did not need to answer: What should AGI realization be for?

This book adopts a foundational normative commitment: AGI realization should be oriented toward the benefit of human beings situated within AGI-inclusive humanity, with AGI-inclusive humanity providing the larger configuration within which that benefit is assessed.

The commitment is stipulative rather than empirically derived and is held fixed for this book’s inquiry, while particular judgments about benefit remain revisable. It restricts this book’s evaluative scope to human benefit; it does not rank human claims against claims on behalf of AI systems or other possible nonhuman claimants, and it does not fix how benefit must be operationalized. Questions of artificial moral status remain open. “Benefit” remains deliberately thin and plural rather than being equated in advance with welfare, flourishing, autonomy, capability, preference satisfaction, or any other master metric. Distribution, intergenerational effects, feasible alternatives, and standing remain open to later analysis.

2.2 Direction Without Terminality

A stable normative orientation does not require history to converge on a final configuration.

This is the book’s Direction Without Terminality principle, a high-confidence candidate: orientation toward a purpose does not by itself establish a terminal state.

Put more simply: purpose does not imply endpoint.

AGI-inclusive humanity supplies the normative frame within which realized consequences are assessed for their effects on humans and the institutions, practices, infrastructures, and relations that shape human life. Those same consequences may alter the configuration inherited by later periods.

Consider a simplified sequence: an AI capability changes scientific work; institutions revise training, publication, evaluation, or authorship; later systems are then developed or deployed inside that altered environment.

This illustrates a possible temporal return path: capability → realization → realized consequence → changed realization conditions → subsequent realization.

“Return” means that an effect of an earlier realized state can re-enter as a condition relevant to later realization; it is more specific than mere temporal succession. The schematic marks an analytical hypothesis, not an established mechanism. It remains unverified until evidence satisfies the epistemic burden developed in Chapters 10–11. Chapter 11 calls it feedback only when evidence shows that the earlier outcome actually altered a later realization condition, and recursion only when altered conditions repeatedly re-enter later cycles.

The fuller treatment of Ongoing, Unfinished Transformation appears in Section 2.7. Here the governing stance is only non-presumption of terminal closure: a terminal state is neither assumed nor ruled out in advance.

The normative commitment therefore does not depend upon endless technological change. More capability, deployment, realization, or speed is not definitionally better; realization is not maximized as an end in itself.

Benefit must also be indexed rather than allowed to validate itself retrospectively. A positive benefit claim is necessarily comparative: it must identify a baseline or feasible alternative against which the consequence is judged, even though Chapter 2 fixes no privileged default comparator. At least four families of questions therefore remain open: distribution among contemporaries; trade-offs across generations; which prior, current, or future populations are relevant; and which baseline or feasible alternative supplies the comparison.

These questions connect to Indexed Optimality later in the book, but Chapter 2 does not solve them. It only blocks the inference that a transformed configuration is beneficial merely because it has become actual.

Continuity is a separate problem. Reusing the same name is not enough to show that changing configurations are the same normative frame, while mere causal succession is also too weak: one state can cause another without preserving the relation the concept is meant to track. Later synthesis will therefore need both conceptual criteria for what persists and pragmatic criteria for when treating changing configurations as one continuing frame remains analytically useful.

Standing is different again. Beneficiaries are those whose interests are at issue; standing concerns who may judge, represent, contest, or authorize claims about those interests. Governance and Participation address parts of that problem without inserting standing as another argument of the realization mapping.

The purpose claim supplies a direction of concern, not an algorithm for deciding every contested case.

2.3 Capability, Realization, and Benefit

Four concepts that are often compressed into one narrative of technological progress must remain analytically distinct: capability, realization, realized consequence, and benefit.

The first non-entailment is canonical in the Unified Gradual AGI Model: Capability ⇏ Realization.

Potential–Realization Separation means that what an AI system can do is not the same as what becomes instantiated under actual conditions.

A second non-entailment is equally important here: Realization ⇏ Benefit.

The fact that a capability becomes realized does not establish that its consequences benefit the human beneficiaries identified in Section 2.1.

Capability concerns what an AI system can potentially do under specified conditions; realized state concerns what is actually instantiated.

Capability can increase even where access is narrow, institutions have not adapted, or practical incorporation remains limited. A demonstrated capability is therefore not equivalent to a realized state in a human system.

Realization concerns the conversion of potential into an instantiated state within a specified group, domain, institution, practice, or other bounded unit.

Chapter 4 formalizes the realization mapping: Potential, Availability, Exposure, and domain conditions are mapped to a realized state. Optimization, Governance, Participation, Observation and Evidence, and Feedback/Recursion interact with that architecture in distinct ways rather than becoming silent extra scalar arguments.

Chapter 2 needs only the narrower point that realization is relational and indexed.

The same underlying capability can therefore be realized differently across hospitals, laboratories, firms, schools, governments, households, professions, groups, or places. A locally meaningful quantity should not automatically be promoted into one global scalar for humanity as a whole.

The realization question is: What potential becomes instantiated, for which indexed unit, under what conditions?

That question differs from what follows afterward. Chapter 4 denotes the realized state for group g at time t as R_g(t). The notation does not imply a scalar realized state: Chapter 4 treats the realization mapping and its arguments as potentially heterogeneous and structured. Qualitative institutional and contextual conditions remain part of that architecture rather than being forced into one number. Realized consequences are changes or effects that may follow from the realized state; they are not identical to the state itself.

This generic realization-level use of consequence must also remain distinct from Chapter 8’s defined Participatory Consequence, which is associated specifically with a Participatory Act or Participatory Configuration.

The evaluative question comes after these descriptive distinctions: What do realized consequences contribute to or subtract from the human beneficiaries situated within the AGI-inclusive configuration?

Benefit is an evaluative layer applied to realized consequences, not an additional argument of the realization mapping or a new subsystem of the Unified Model.

A compact representation is therefore: Potential ⇏ Realization → realized consequence ⇏ Benefit. Here Potential is the realization model’s term for capability potential. The middle arrow marks that realization may have consequences; it is not an entailment claim.

No single global quantity is assumed to exhaust what benefit means across human beneficiaries.

Benefit is differently structured from capability attribution. It adds questions of distribution, temporal index, standing, conflicting objectives, institutional authority, and heterogeneous consequences.

A realized system may improve one outcome while concentrating authority, widening burdens elsewhere, or creating longer-run costs. Conversely, constraining realization may sometimes serve a broader or longer-term human purpose.

The point is structural rather than predictive: realization and benefit cannot be identified by definition.

For the same reason, Chapter 2 does not supply a universal welfare function.

Later chapters discipline different parts of the problem: Optimization handles objectives and alternatives; Governance addresses steering and adjudication; Participation addresses subjects, acts, and participatory consequences; Observation and Evidence constrain inference; Feedback and Recursion examine how earlier outcomes condition later realization.

Those frameworks remain distinct rather than collapsing into one measure of benefit.

The indexing, continuity, and standing problems introduced in Section 2.2 therefore remain inherited burdens, not hidden assumptions.

The framework can remain open to repeated change without assuming that change is necessarily beneficial, progressive, convergent, or endless.

The next sections examine when heterogeneous realization units, differences in timing and sequence, and changing human–AI relations produce realization that is gradual in the Chapter 1 sense rather than exhausted by a single event.

2.4 Heterogeneous Constituents, Sequencing, and Gradual Realization

AGI-inclusive humanity is not a single actor or a single realization unit.

For present purposes, four broad constituent classes are useful: humans, AI systems, institutions, and material or technical infrastructure. The categories are descriptive rather than exhaustive and do not partition the arguments of the realization mapping. Practices and environments cut across them—for example, a surgical practice can simultaneously involve clinicians, AI systems, hospital rules, data systems, and physical infrastructure.

Humans include persons, groups, communities, and professions understood as communities of practitioners whose activities, opportunities, burdens, and capabilities may be altered through AI realization.

AI systems include models, agents, tools, applications, and compound systems through which artificial capabilities become available for use.

Institutions include firms, governments, universities, professions as organized structures, standards bodies, legal systems, markets, and other arrangements that mediate authority, access, uptake, constraint, and adaptation.

Infrastructure includes compute, networks, energy, data systems, interfaces, facilities, and other technical resources relevant to capability development or realization; the category does not predetermine where a resource belongs in the later realization mapping.

These constituents need not change together.

Capability can advance while infrastructure, institutional permission, or professional practice lags; conversely, regulation, financing, standards, or infrastructure can alter realization even when model capability changes little. As an illustration, AI-assisted software refactoring can be incorporated quickly into an existing developer workflow, while AI use in clinical surgery may depend on validation, hospital authorization, specialized equipment, and professional accountability before comparable realization occurs.

This establishes heterogeneity, but heterogeneity alone is not gradualness.

Different units can occupy different realized states at the same moment even if each arrived there through one completed event. Cross-sectional unevenness therefore does not by itself establish a gradual realization process.

The additional premise is temporal and relational: realization units can differ in when change occurs, through which sequence, and under what dependencies. One unit’s change may precede, enable, constrain, or redirect another’s.

A realization unit is the bounded group, domain, institution, practice, place, or other indexed object for which a realization claim is made; the term introduces no new model variable.

Chapter 1 classifies a specified AGI-related transformation as phenomenally gradual, under a declared individuation rule, when its relevant realization is not exhausted by one event. Heterogeneity contributes to that possibility only when units differ in timing, sequence, or dependence, so that realization unfolds across distinct events.

Temporal lag can satisfy the descriptive classification once the realization has been individuated across more than one event; the stronger explanatory question is whether dependence among units helps explain why realization unfolds across those events rather than at once.

Unevenness and gradualness can therefore coexist, but neither entails the other.

The claim remains permissive: abrupt, synchronized, or effectively complete transitions are possible, and gradualness means neither slowness nor smoothness.

A capability event therefore need not correspond to one civilizational realization event. What is realized for one group may not be realized for another, or may occur on a different schedule.

This is compatible with the book’s high-confidence candidate principle of Local Scalarity / Global Non-Scalarity: local measures need not combine into one global scalar.

Chapter 3 organizes claims about such heterogeneous transformation; Chapter 4 and later chapters represent realization across indexed units within that architecture.

Two further relations now need clarification: how artificial capability enters human activity, and how reciprocal activity can occur without symmetrical roles.

2.5 Extension, Augmentation, and Modes of Realization

The idea that technology can extend human capability is older than artificial intelligence. Language, writing, instruments, institutions, computation, and communication systems enable forms of human activity that unaided biological capacities could not accomplish in the same form or at the same scale.

The 2025 First Principles article stated this extension relation definitionally: artifacts of humanity’s self-made aspect were “extensions, expansions, and/or upgrades” of their nature-made counterpart. The present book retains the genealogy but narrows its status.

The self-made:nature-made maturity ratio is not a canonical construct here, and technological extension is not treated as progress by definition.

Artificial capability can extend human capability without making the artificial system identical to the human whose activity is extended.

Extension names the general relation; augmentation names one realized mode of that relation.

Once artificial capability is realized, at least four non-exhaustive modes are useful:

• augmentation — artificial capability expands or supports what a human or organization actually does;
• substitution — an artificial system performs an activity previously carried out by a human;
• complementarity — human and artificial contributions combine in an instantiated practice or outcome;
• reorganization — the activity, workflow, role structure, or institutional process itself changes, sometimes creating practices with no clean pre-AI counterpart.

The modes can overlap and recurse into one another; reorganization, for example, can create new opportunities for augmentation or substitution. Hybrid configurations should therefore be classified at the level of the function or transition being described rather than forced into one exclusive mode. The categories are descriptive, not evaluative rankings and not additional arguments of the realization mapping.

This distinction also protects a boundary inherited from Chapter 1, which defines AGI against the unaugmented human capability envelope so that AI assistance does not move the benchmark while AGI itself is being defined. “Unaugmented” there denotes an analysis-relative assistance policy: by default, it excludes AI assistance in performing the target task while permitting ordinary education, conventional non-AI tools, and institutional supports unless otherwise declared. It therefore does not privilege the Chapter 2 augmentation mode over substitution, complementarity, or reorganization.

Definition asks what the AI system can do relative to that unaugmented reference class. Realization asks what is actually instantiated once artificial capability enters practice.

A capability may expand what someone could do without changing actual practice; conversely, an unchanged model can contribute to a new realized state once workflow, organization, or institutional routine changes.

The realized state is therefore not a “realized capability.” R_g(t) denotes the realized state for group g at time t; consequences may subsequently follow from that state.

Modes of realization can also alter later conditions and thereby enter the temporal return paths introduced in Section 2.2; Chapter 11 determines when evidence warrants feedback or recursion.

This possibility supports a limited, heuristic use of co-evolution: changes on more than one side of a human–AI relation can become historically connected without implying equal agency, equal rates of change, reciprocal causation in every case, or inevitable interdependence. The term is descriptive here, not a separate technical construct.

A human may adapt to a technically fixed system; a provider may change a system in response to aggregate use; an institution may reorganize around an unchanged model; or AI-assisted development may contribute to later artificial capability. The empirical configuration determines which description applies.

No mode is beneficial or harmful by definition. Its consequences must be evaluated rather than inferred from the label.

2.6 Reciprocal Activity Without Symmetry

AGI-inclusive humanity is not adequately described if AI is treated only as a passive object upon which humans act; recognizing artificial activity, however, does not make humans and AI systems equivalent.

The broader relation can be described as reciprocal activity without symmetry.

Reciprocity means that human and artificial activity can both enter processes affecting what is done, what becomes realized, and what conditions are inherited. Symmetry would require the stronger claim of equivalence in agency, authority, dependence, responsibility, standing to judge, vulnerability, or other relevant status.

An AI system can generate a proposal; a human can select, reject, reinterpret, or constrain it; an institution can determine whether either contribution affects a realized outcome; and infrastructure can determine whether the capability is available at all.

Several boundaries follow: artificial contribution does not imply personhood; causal contribution does not imply moral status; operational autonomy does not imply institutional authority; dependence does not imply equality; reciprocity does not imply symmetry.

Chapter 8 later defines Participatory Subject, Participatory Act, Participatory Consequence, and the Relational Grammar of Participation—Realization Orientation, Foreclosure Position, and Actor Dependence. Chapter 2 does not import those terms into a general ontology of human–AI interaction.

Its broader claim is only that human and artificial activities can both contribute to realization while remaining differently situated.

Those roles can change: artificial systems may perform more execution while human activity shifts toward selection, verification, interpretation, oversight, rule-setting, or adjudication. In other cases human roles may diminish without replacement by meaningful control. These are possibilities, not automatic effects of automation.

This observation is broader than Chapter 8’s conditional proposition on Participatory Role Plasticity and does not generalize that proposition into a universal law.

AGI-inclusive humanity therefore presupposes neither unchanged humans using ever-stronger tools nor AI becoming a separate civilization outside humanity. It leaves multiple configurations open.

The normative commitment of Section 2.1 governs evaluation of those configurations; descriptive inclusion of AI systems does not resolve their moral status.

Taken together, Sections 2.4–2.6 show that realization can be heterogeneous, sequenced, mode-dependent, and reciprocal without symmetry.

None of those features proves that every realization process is gradual or permanently unfinished. They establish only that AGI-inclusive humanity does not require one synchronized, terminal realization event.

The next section connects that limited conclusion to the book’s Ongoing, Unfinished Transformation orientation.

2.7 Ongoing, Unfinished Transformation

The preceding sections do not establish permanent openness. They establish only that AGI-inclusive humanity can contain heterogeneous realization units, distinct modes, reciprocal activity without symmetry, and possible temporal return paths.

The book nevertheless adopts Ongoing, Unfinished Transformation as an architectural orientation. Chapter 3 uses the phrase as methodological shorthand: realization need not be exhausted by one event or presumed complete at a particular capability threshold, and the forms, relations, and consequences under study remain revisable.

Chapter 2 therefore adopts non-presumption of terminal closure, not a prediction of permanent disequilibrium.

Two synthetic principles sit behind that stance. Direction Without Terminality (a high-confidence candidate) says that orientation toward a purpose does not by itself establish a terminal configuration. Adaptive Disequilibrium adds the stronger, provisional possibility of continued disruption and adaptation without convergence; a demonstrably absorbing domain equilibrium would narrow or defeat that stronger claim.

Temporary stabilization is not an absorbing equilibrium, and a finite quiet window does not establish permanent closure. Chapter 3’s result-status vocabulary and Chapter 10’s observation–inference–claim-status discipline preserve that distinction.

Chapter 11 carries the domain-level diagnostic burden. There, “ongoing” is a process claim: consequential outcomes continue to alter later realization conditions. “Unfinished” is a closure-status claim: no empirically justified terminal configuration has been established for the specified domain and observation horizon. A terminal state is analysis-relative rather than metaphysical.

The normative commitment is the one stated in Section 2.1.

Two open problems attach to it. First, continuity: what warrants treating changing configurations as continuations of AGI-inclusive humanity? Second, beneficiary scope and evaluative index: which humans are relevant, and relative to which time, baseline, and feasible alternatives is benefit assessed?

Standing is separate from both: beneficiaries are those whose interests are at issue; standing concerns who may judge, represent, authorize, or contest claims about those interests. Artificial moral status remains bracketed rather than prejudged.

Indexed Optimality, Governance, Participation, and later empirical chapters address parts of these burdens; continuity itself remains an explicit open item for the book’s synthesis.

The Multi-Plateau Framework in Chapter 3 supplies a logical rather than chronological architecture. Its plateaus organize kinds of theoretical work—what is being defined, formalized, tested, or synthesized—not stages through which societies or domains move.

Chapter 4 then formalizes realization across indexed units. Chapters 5–11 develop synchronization, optimization, governance, conceptual and empirical Participation (Chapters 8–9), observation and evidence, and feedback/recursion within that organized theory.

Chapter 2 has supplied the evolving frame and its human-centered normative orientation; separated capability, realization, realized consequence, and benefit; distinguished unevenness from gradualness; identified modes of realization; and articulated reciprocity without symmetry. It leaves the remaining burdens visible rather than resolving them by definition.

The following open burdens are inherited by the book’s later synthesis:

• beneficiary scope and evaluative indexing across groups, generations, times, baselines, and feasible alternatives;
• continuity of AGI-inclusive humanity across changing configurations;
• standing to judge, represent, or contest benefit;
• the unresolved moral status of artificial systems; and
• the evidentiary threshold at which a possible temporal return path becomes feedback or recursion.

With those burdens visible, Chapter 3 can organize the theory that follows without treating unresolved questions as settled.

Chapter References

W.H.L. with Claude Sonnet 4. 2025. “First Principles of AGI-Inclusive Humanity.” Champaign Magazine, June 17, 2025. https://champaignmagazine.com/2025/06/17/first-principles-of-agi-inclusive-humanity/

W.H.L. with GPT-5.6 Sol. 2026. “AGI and the Meaning of Gradual Transformation.” On Gradual AGI, Chapter 1, Publication Version v1.04, September 28, 2026.

W.H.L. et al. 2026. “A Multi-Plateau Framework of the Unified Gradual AGI Model.” On Gradual AGI, Chapter 3, Publication Version v1.0, September 6, 2026.

W.H.L. et al. 2026. “From Potential to Realization.” On Gradual AGI, Chapter 4, Publication Version v1.2, September 9, 2026.

W.H.L. et al. 2026. “Synchronization: Ceiling, Floor, and Slope.” On Gradual AGI, Chapter 5, Publication Version v1.0.1, September 9, 2026.

W.H.L. et al. 2026. “Optimization: Choosing Among Trajectories.” On Gradual AGI, Chapter 6, Publication Version v1.0, September 10, 2026.

W.H.L. et al. 2026. “Governance: Contesting Realization.” On Gradual AGI, Chapter 7, Publication Version v1.0, September 11, 2026.

W.H.L. et al. 2026. “Participation: Subjects, Acts, and Consequences.” On Gradual AGI, Chapter 8, Publication Version v1.1.5, September 15, 2026.

W.H.L. et al. 2026. “Participation: What Survives Empirical Contact.” On Gradual AGI, Chapter 9, Publication Version v1.0.3, September 16, 2026.

W.H.L. et al. 2026. “Observation, Evidence, and Epistemic Discipline.” On Gradual AGI, Chapter 10, Publication Version v1.0, September 18, 2026.

W.H.L. et al. 2026. “Feedback, Recursion, and Ongoing Transformation.” On Gradual AGI, Chapter 11, Publication Version v1.01, September 19, 2026.