No records match the selected filters.
Applied Research Article
Preprint
Posted 28 September 2026
·
Author Johnny Kao
Employment is commonly treated as a source of labor income, but in modern market economies a stable job often performs a much larger distributive function. It provides current earnings, produces human capital, anchors access to social insurance and retirement saving, supports credit and asset accumulation, and gives households a recurring claim on the gains generated by productive firms. This paper develops the concept of Work as Distribution Infrastructure by formalizing an argument introduced in Kao (2026): when firms can produce more value with fewer additional workers, the economic importance of employment extends beyond the number of jobs at risk. The question becomes how productivity gains continue to reach households when job growth is a weaker transmission mechanism.
The paper distinguishes three participation channels. Work connects households to growth through wages, experience, and employment-linked benefits. Ownership connects them through returns to productive assets, directly or through pensions, funds, employee financial participation, and other intermediated forms. Creation connects them through entrepreneurship and the formation of new products, firms, and positions. These channels are complements rather than substitutes. Ownership does not replace the developmental and social functions of work, and entrepreneurship cannot be a universal solution. The framework therefore treats distribution infrastructure as a portfolio of institutional connections rather than a single policy instrument.
Korea provides a useful motivating case. Its growth model historically linked industrial expansion, large-firm employment, skill formation, and household advancement, while current household balance sheets remain concentrated in real assets. At the same time, AI may raise the return to capital, alter the labor share, and lower some organizational barriers to small-team production. The contribution is not a policy prescription for universal asset ownership or startup formation. It is a diagnostic framework for asking whether AI-led productivity growth remains broadly connected to household income, capability accumulation, and claims on future economic value.
Applied Research Article
Preprint
Posted 28 September 2026
·
Author Johnny Kao
Artificial intelligence may allow firms to expand output without expanding entry-level employment at the same rate. This possibility is usually discussed as a question of job displacement, but that framing misses a second mechanism: the weakening of the institutional pathway through which inexperienced workers acquire their first real responsibilities, accumulate judgment, and become experienced workers. This paper develops the Employment Transmission Gap framework by formalizing an argument introduced in Kao (2026): national and firm-level success can continue even as the connection between growth and career entry becomes less reliable.
The framework distinguishes productive growth from career-entry capacity. An Employment Transmission Gap arises when productive capacity, revenue, or value added grows while quality-adjusted opportunities for inexperienced workers grow materially more slowly, stagnate, or contract. Four mechanisms can generate the gap: incumbent productivity absorption, seniority-biased task reallocation, experienced-hire substitution, and the erosion of low-risk learning tasks. Korea is a useful motivating case because recent Bank of Korea evidence shows that youth employment weakness has been concentrated in AI-exposed industries while experienced-hire preferences and labor-market dualism predate the current AI wave. Comparable U.S. evidence points in a similar direction, although other recent studies show that the magnitude and persistence of entry-level effects remain unsettled.
The contribution is diagnostic rather than predictive. The paper does not claim that AI necessarily reduces aggregate employment or that entry-level tasks should be preserved unchanged. Instead, it proposes that the quality and quantity of career-entry opportunities should be measured separately from output and productivity. The relevant question is not only whether AI creates or destroys jobs, but whether an economy continues to produce the experience pathways required to create its next generation of skilled workers.
Applied Research Article
Preprint
Posted 27 September 2026
·
Author Johnny Kao
Country size is commonly proxied by territory or resident population, yet these measures can misstate the economic scale available to highly integrated small states. This paper develops the concept of accessible economic scale: the labour, demand, infrastructure, and productive networks that firms and institutions can use without those resources being located inside the political territory or resident population. The framework is motivated by Liechtenstein, where the number of employed persons exceeds the resident population and 57.4 percent of workers were inward commuters at the end of 2024. Cross-border labour is especially important in industry, while firms operate inside a highly open economy connected to Swiss and European markets. The resulting structure also helps explain the unusually large gap between GDP and GNI documented by the IMF: production occurs inside the territory while substantial labour income accrues to non-residents. The paper distinguishes resident scale from accessible scale, identifies labour, market, and infrastructure access as separate channels, and proposes a qualitative Scale-Access Test based on depth, friction, continuity, and bottlenecks. The contribution is not that openness benefits small states, a well-established result. It is the narrower claim that resident population is an incomplete proxy for the productive scale a state can mobilize, and that economic viability depends partly on access to scale beyond the border.
Applied Research Article
Preprint
Posted 27 September 2026
·
Author Johnny Kao
Small states are commonly analyzed through a capacity problem: they face high fixed costs, thin administrative pools, and vulnerabilities that make it difficult to reproduce the institutional apparatus of larger states. Yet this framing can obscure an important distinction between possessing a capability, directly providing it, and retaining political authority over the arrangements through which it is obtained. This paper develops a provision-authority separation framework for small-state capacity. It argues that a state can access monetary, customs, administrative, security, or other capabilities through durable external arrangements without thereby transferring all political authority over the relevant policy domain. Liechtenstein provides a structured case because its statehood combines extensive external provision with continued international sovereignty and demonstrable instances of policy divergence from its principal partner, Switzerland. The paper distinguishes three analytical questions: whether a capability is reliably accessible, who supplies or administers it, and where politically meaningful authority remains. It introduces the concepts of provision-authority separation and externalized state capacity, together with a qualitative capability-architecture test. The contribution is integrative rather than a claim that dependence, delegation, or external shelter are new phenomena. The central claim is that state capacity and state self-provision are not the same thing: the institutional location of a capability does not by itself determine where political authority resides.
Published 27 September 2026
·
Author Johnny Kao
Agricultural artificial intelligence can create substantial operational value without producing an equally strong technology business. A robot may reduce a scarce task, a decision system may improve reliability, or an autonomous system may expand feasible production, yet the supplier can still fail to identify the right payer, overcome adoption friction, deploy repeatably, or retain enough of the created value to sustain the offering. This paper develops an Agricultural AI Value-Capture framework that treats commercialization as a linked sequence from technology to constraint removal, economic value, buyer identification, adoption, deployment, and value capture. The framework is positioned against recent research on digital-agriculture adoption, interoperability, platformization, servitization, market power, and business-model innovation. Existing work establishes that value creation and value capture are distinct, that profitability and technical support shape adoption, and that digital agriculture increasingly relies on service- and platform-based models. The contribution here is narrower and operational: it makes deployment an explicit strategic link between adoption and capture, defines Deployment Replicability as the ability to reproduce a stable solution with bounded site-specific adaptation, and introduces Buyer-Value Alignment and Value-Capture Leakage as diagnostic concepts for agricultural AI businesses. The framework implies that superior AI is neither necessary nor sufficient for a scalable agricultural technology business. Stronger positions arise when a technology removes a binding constraint for an identifiable payer, creates value large enough to justify operational disruption, enters heterogeneous farms through a repeatable deployment system, and retains an economically viable share of the value after service, financing, competition, and ecosystem effects. The central claim is that agricultural AI should be evaluated not only by what it can technically accomplish, but by whether the path from constraint removal to repeatable value capture remains intact.
Published 27 September 2026
·
Author Johnny Kao
Agricultural technology increasingly makes temperature, humidity, light, irrigation, carbon dioxide, airflow, and other production variables technically controllable. Yet technical controllability is not the same as economic desirability. Each additional layer of control can raise yield, quality, timing reliability, or loss avoidance while also adding capital, energy, maintenance, sensing, integration, and management costs. This paper develops the Profitable Control framework by formalizing the argument in Kao (2026a) that the relevant question is not whether a variable can be controlled, but whether it is worth paying to control. The framework treats control as an economic choice over intensity rather than a binary technological capability. The paper connects greenhouse optimal-control research, controlled-environment agriculture, vertical-farming economics, techno-economic benchmarking, and recent work on AI-enabled climate control. Existing studies already optimize yield, energy use, profit, and resource efficiency within particular systems. The contribution here is integrative rather than a claim to a new control-theoretic method: it introduces a portable vocabulary for comparing control decisions across crops, technologies, and locations. Four concepts organize the framework: Control Value, Control Burden, Profitable Control, and the Economic Control Boundary. A lightweight formulation states that tighter control is justified only when its expected incremental economic value exceeds its full incremental control burden. Recent evidence illustrates why this distinction matters. Autonomous greenhouse systems can choose among very different lighting, heating, carbon-dioxide, irrigation, and harvest strategies while being judged on profit rather than biological output alone. Model-predictive control can materially reduce energy use and improve economic performance without necessarily making an otherwise unattractive production system viable. Vertical-farm studies likewise show that optimal settings depend on electricity prices, climate, crop physiology, capital structure, and the value of the crop being sold. The central implication is that AI does not repeal the economics of physical control. Its value lies in shifting the cost-value relationship of control: using existing equipment more selectively, responding to changing prices and crop states, and concentrating precision where the crop can pay for it.
Published 27 September 2026
·
Author Johnny Kao
Artificial intelligence in agriculture is often described as a substitute for labor or as a tool for improving prediction, precision, and control. Yet many agricultural systems are constrained by a different scarce resource: experienced judgment. A farm may have land, capital, equipment, workers, and demand, but still depend on a limited number of people who know what to notice, how to interpret changing conditions, and when to intervene. This paper develops the Scalable Agricultural Intelligence framework by formalizing the argument in Kao (2026a) that intelligence can be treated as an agricultural input alongside more visible inputs such as land, labor, water, energy, and capital. The term does not imply a standardized factor of production. It identifies the operational capacity to observe conditions, interpret their significance, and select useful actions. The paper connects agricultural extension, decision-support systems, human-centered AI, agricultural expertise, and the economics of knowledge in organizations. It introduces four definitions: Agricultural Intelligence, Judgment Bottleneck, Expertise Reach, and Scalable Agricultural Intelligence. The central mechanism is that AI can reduce the amount of routine expert judgment required per unit of production, route exceptional cases toward scarce experts, lower the knowledge required at the point of entry for less experienced operators, and partially separate expertise from physical location. A lightweight formulation expresses expertise reach as a capacity relationship rather than a calibrated production function. Evidence from digital agricultural extension, autonomous greenhouse experiments, human-centered agricultural AI, and recent workplace AI research supports components of this mechanism while also showing its limits. Digital advice can expand knowledge reach without automatically raising output; agricultural skill contains embodied and contextual elements that are not readily codified; and AI benefits may be largest where routine decisions can be supported while humans retain responsibility for unusual conditions. The paper therefore argues that AI need not replace agricultural experts to change agricultural economics. It may be sufficient for the same scarce expertise to support more production, more locations, or less experienced operators, provided that the system preserves reliable escalation to human judgment.
Applied Research Article
Preprint
Posted 27 September 2026
·
Author Johnny Kao
Artificial intelligence in agriculture is commonly evaluated through changes in yield, input use, labor cost, profitability, sustainability, or adoption. These measures are necessary, but they can miss a distinct economic effect: a technology may change whether a production configuration is economically viable at all. This paper develops the Agricultural Viability Frontier framework by formalizing the concept of an agricultural frontier introduced in Kao (2026). Building on research on digital agriculture, technology adoption, intensive and extensive supply responses, farm viability, and the economics of autonomous machinery, the paper distinguishes optimization from frontier expansion. Optimization improves a crop-location-scale-production configuration that was already viable; frontier expansion changes membership in the viable set. The framework defines a production configuration, an economically viable set conditioned on technology and the surrounding economic environment, and frontier-expanding technological change. It further proposes a Frontier Expansion Test centered on four questions: whether the activity would occur in substantially the same configuration without the technology, which constraint is binding on viability, whether the technology materially relaxes that constraint, and whether complementary conditions remain sufficient. Recent evidence on autonomous field machinery, labor scarcity, robotic harvesting, and autonomous greenhouse control illustrates why technical success or positive return on investment is not by itself evidence of frontier expansion. The contribution is integrative rather than a claim that viability boundaries or extensive-margin effects are new: the paper extends a prior geographic use of a farm viability frontier into a technology-conditioned, configuration-level framework for evaluating when AI makes agriculture better and when it makes additional agriculture economically possible.
Applied Research Article
Preprint
Posted 27 September 2026
·
Author Johnny Kao
Artificial intelligence can produce work that is functionally correct at delivery yet difficult to inspect, reconstruct, or modify later. This paper develops the concept of decision debt for AI-generated work: the future cost created when an acceptable output is retained without preserving the assumptions, evidence, dependencies, and interfaces needed to understand and safely change it. The concept extends but does not replace technical debt. Whereas technical debt traditionally concerns implementation and architectural compromises that increase future maintenance cost, decision debt focuses on lost decision structure across software, analytical work, and agentic workflows. The paper proposes four preservation properties - observability, traceability, modularity, and changeability - and a future-changeability test for evaluating whether AI-generated artifacts remain governable after delivery. It also offers a qualitative decision-debt checklist and a risk-adjusted preservation heuristic for deciding how much reconstruction evidence to retain. It also distinguishes external reconstruction evidence from private model reasoning: maintainability does not require access to an AI system's hidden chain of thought, but it does require sufficient records of sources, actions, tests, assumptions, and artifacts to reconstruct consequential decisions. The contribution is integrative rather than a claim that provenance, maintainability, or technical debt are individually new. The central claim is that correctness is a present-tense property, while changeability is a future-tense property; reliable AI-generated work requires both.
Applied Research Article
Preprint
Posted 27 September 2026
·
Author Johnny Kao
As agentic artificial intelligence systems increasingly produce, evaluate, and act on their own outputs, review architectures face a structural problem: adding a second judgment does not necessarily create an independent basis for judgment. This paper proposes verification independence as a task-level design property for agentic AI. Verification independence asks whether the review process is sufficiently separated from the production process to surface errors that the producer is systematically unlikely to detect. The framework distinguishes four dimensions of independence: framing, evidence, mechanism, and authority. It also formalizes a four-function review architecture consisting of a producer, challenger, evidence checker, and acceptance owner. The framework is positioned relative to research on LLM self-correction, chain-of-verification, multi-agent debate, LLM-as-a-judge, correlated model errors, and emerging agentic verification systems. Existing work shows both the value and limits of additional model-based review: structured verification can improve outputs, while multi-agent agreement, self-critique, and LLM judging remain vulnerable to shared assumptions, correlated errors, evaluator bias, and noisy rubric verification. The contribution here is integrative rather than a claim that any individual verification technique is new. It treats independence as a property of the relationship between production and review, not as a synonym for reviewer count or model diversity. The central claim is that a reviewer can provide another judgment without providing another independent basis for judgment. For consequential agentic workflows, reliable verification therefore requires deliberate separation of review roles and, where feasible, external evidence, deterministic tests, independent tools, or distinct authority boundaries. The paper further introduces a qualitative independence checklist, risk-adjusted verification depth, and safeguards against erosion of independence in long-horizon workflows.
Applied Research Article
Preprint
Posted 27 September 2026
·
Author Johnny Kao
As artificial intelligence systems move from recommending actions to executing them, human oversight can no longer be treated as a binary choice between autonomy and a human-in-the-loop. This paper proposes an attention-constrained framework for designing agent autonomy around four variables: risk, reversibility, authority, and human attention. Risk, reversibility, and authority describe the consequence profile of an action; human attention describes the scarce oversight resource available to review it. Drawing on research on levels of automation, appropriate reliance, automation bias, meaningful human oversight, and recent empirical studies of software agents, the paper argues that frequent approval requests can reduce rather than increase meaningful control when they consume attention faster than humans can exercise judgment. The framework introduces an action-level oversight demand, an explicit human-attention budget, a lightweight extension for cumulative exposure across action sequences, and two design principles: exception-based oversight and approval quality over approval count. It further distinguishes four complementary modes of oversight - boundary setting, monitoring, escalation, and audit - and shows how they can be combined so that routine, bounded actions proceed without synchronous review while consequential or difficult-to-reverse actions receive concentrated human attention. The paper also links these modes to task-level delegation contracts and proposes adaptive safeguards when review conditions appear to degrade. The contribution is integrative: reliable human oversight is framed not as maximizing human presence, but as allocating human judgment where its marginal value is highest.
Applied Research Article
Preprint
Posted 27 September 2026
·
Author Johnny Kao
As artificial intelligence systems move from generating outputs to executing multi-step workflows, the control problem shifts from response quality to delegated action. This paper proposes the delegation contract, a six-field task-level framework for agentic AI that integrates semantic task specification with operational authority and escalation. A delegation contract specifies objective, context, constraints, acceptance criteria, authority, and escalation. The first four fields define semantic delegation: what the task means and how acceptable completion will be recognized. The final two define operational delegation: what the agent may do and when decision rights must return to another actor. The framework is positioned relative to principal-agent theory, meaningful human control, authenticated delegation, agent infrastructure, authorization architectures, and contemporary agent governance. Its contribution is integrative rather than a claim that the individual elements are new: it treats purpose, completion evidence, delegated action rights, and return-of-control conditions as one coherent task-level control object. The paper also outlines characteristic failure modes, a lightweight machine-readable implementation pattern, risk-adjusted specification depth, and authority attenuation in multi-agent delegation. The central claim is that reliable agentic autonomy requires more than a clear prompt and more than scoped permissions; it requires an explicit specification of what counts as success, what the agent is authorized to do, and when that authority expires.
Applied Research Article
Preprint
Posted 27 September 2026
·
Author Johnny Kao
Small states routinely rely on larger states and international institutions for markets, monetary stability, administrative capacity, security, and regulatory access. Dependence is therefore often treated as a constraint on sovereignty. Yet formal independence and external dependence can coexist for long periods, while a legal right to withdraw from an arrangement may provide little practical freedom when alternatives are prohibitively costly.
This paper develops an adjustability-centered framework for analyzing sovereignty under dependence. It distinguishes formal authority from practical sovereignty and argues that the relevant question is not whether a state is dependent, but whether consequential external relationships remain politically adjustable through negotiation, redesign, substitution, or, at the limit, exit.
Liechtenstein provides a structured case. Its postwar and European integration history shows repeated dependence on larger systems alongside episodes of reorientation, treaty modification, differentiated integration, and externally induced policy adaptation. The paper introduces the concepts of relationship adjustability and practical choice, and proposes a qualitative Adjustability Test focused on authorization, voice, redesign, substitution, and transition costs.
The contribution is diagnostic rather than classificatory: dependence becomes sovereignty-thinning when formal choice persists but feasible alternatives and credible adjustment mechanisms collapse.