Research and working papers defining epistemic generativity, evaluative AI architectures, and the structural limits of optimisation-based systems.
This paper argues that out-of-distribution failure is not merely degraded performance but a structural signal of epistemic enclosure: current machine learning systems cannot treat failure as evidence against their own evaluative standards, only as deviation within them. details
Until now, no explicitly epistemic theory of generativity has been formulated. This paper defines the structural threshold between optimisation and generativity, locating agency and responsibility in control over evaluative criteria rather than behaviour. details
A board-level governance paper arguing that directors may delegate operational functions, but not fiduciary accountability for the architecture through which corporate authority is exercised. As AI-mediated systems become persistent, decision-structuring infrastructure, oversight shifts upstream to system design, authority allocation… details
AI governance fails when it asks whether systems are conscious instead of identifying who authored the evaluative structure under which they operate. This paper relocates responsibility from metaphysical speculation to mandate design, evaluative control, and allocable normative authorship. details
Contemporary AI systems exhibit increasingly sophisticated behaviour—planning, adaptation, and tool use—yet these properties are routinely taken as evidence of agency. This paper argues that such inferences are not justified. It introduces the Duck Criterion: agency, authority, and responsibility track control over evaluative structure, not… details
A philosophical and governance-oriented inquiry into the difference between optimisation within inherited norms and genuine epistemic generativity, asking when responsibility remains traceable in distributed knowledge systems. details
This memorandum constitutes a public comment submitted to NIST NCCoE on its draft concept article concerning software and AI agent identity and authorization. It argues that identity controls are necessary but not sufficient, and proposes mandate specification, delegation-chain modelling, and bounded discretion as structural complements to Zero… details
Artificial General Intelligence is frequently defined as cross-domain adaptability under constraint. This article argues that competence alone does not constitute agency. For AGI to denote a governance-relevant agent, a system must satisfy the Ontological Continuity Condition: non-interchangeability under reset, identity risk, and unified… details
Introduces the Delegation Threshold to specify when AI and digital infrastructure acquire governance-relevant authority. When discretionary functions are embedded in persistent, non-optional systems, users organise decisions around their outputs. Responsibility shifts upstream to architecture, integration, and oversight. details
Argues that as AI shifts from episodic models to persistent embedded infrastructure, governance cannot remain an optional overlay. Operational tempo turns delay into risk, diffusing accountability and exposing institutions to structural liability. Identifies minimal design conditions: contestability, revisability, halt-capacity, and traceability. details
Argues that most systems labelled ‘agentic AI’ are optimisers embedded in governance architectures rather than agents. Agency requires continuity of evaluative authority and authorship of standards—conditions typically neutralised by resetability, external alignment, and liability design. Misattributed agency displaces responsibility and obscures… details
Develops a structural account of epistemic agency grounded in the ontological continuity condition. Many biological and artificial systems can possess knowledge without being epistemic agents. Epistemic agency arises where knowledge must be owned, revised, and acted upon by the same system over time under identity-threatening… details
Reframes agency as episodic: responsibility arises when inherited frameworks fail and an actor must author the normative terms of action. For AI governance, treating optimisation systems as agents mislocates accountability. Infrastructure requires identifiable authorship; without it, liability, oversight, and institutional responsibility become… details
Identifies the frame-stability problem: planning and decision theories assume a unified evaluator but do not explain why it must persist. Argues that agency arises when systems must preserve evaluative authority under diachronic uncertainty through unified arbitration. Most AI optimisers coordinate behaviour without owning commitments—clarifying… details
Challenges the assumption that distributed cognition is inherently reliable. Collective systems are engineered epistemic infrastructures whose performance depends on governance architecture. Develops a taxonomy of markets, platforms, open source, deliberation, and institutions, showing how closure, incentives, and contestability shape failure… details
Reinterprets media technologies as cognitive infrastructure that reshapes bounded reasoning. Argues that AI marks a categorical shift by intervening inside deliberation—compressing representations, expanding counterfactual exploration, and lowering revision costs. Introduces a typology of scaffolds, crutches, and substitutes, highlighting… details
Argues that fiduciary law already embeds epistemic duties governing how decision-makers form and close beliefs under entrusted discretion. Epistemic discipline is a legality condition of authority, not a virtue: failures can constitute fiduciary wrongs even without harm. For AI governance, operators of decision infrastructure bear responsibility… details
Advances a constitutional reframing of democracy under AI: legitimacy depends on fiduciary–epistemic trust rather than procedure or information abundance. Treats AI institutions as epistemic fiduciaries, introduces Fiduciary Epistemic Governance, and proposes legal and institutional forms to restore accountability, autonomy, and public trust… details
Defines the Third Enclosure Movement as consolidation of control over AI datasets, models, compute, and evaluation through legal, technical, epistemic, and sovereign restrictions. Introduces a taxonomy of restricted knowledge and argues for commons-compatible reforms that preserve innovation while clarifying liability, strengthening auditability… details