AI Productivity Must Count the Whole Institution

Artificial intelligence is frequently justified by a simple promise: the same work will be completed faster, at lower cost and at greater scale. That promise can be real. A lecturer may prepare an initial activity more quickly, a public officer may organise records more efficiently, and a ministry team may translate routine communication for more people. Time and money matter, especially in institutions carrying serious responsibilities with limited resources.

But a faster output is not yet public value. A polished lesson can coexist with weaker teacher judgement. An efficient assessment can conceal shallow learning. A responsive chatbot can displace human presence. A low subscription price can depend upon expensive connectivity, unpaid verification, imported infrastructure and difficult exit. Productivity accounting becomes misleading when the benefits are counted at the point of generation while the costs are transferred to learners, workers, communities and the future.

Institutions therefore need a whole-value chain: legitimate purpose → people and work affected → direct benefit → learning and capability → distribution → full cost → risk and remedy → resilience → independent verification → revision or exit. The question is not whether AI can produce more. It is whether the whole institution becomes more capable, just, trustworthy and sustainable because of its use.

The newest official report projects productivity; it does not prove it

Australia's Treasury published the 2026 Intergenerational Report on 21 September. The official publication places the artificial-intelligence revolution among five major transitions expected to shape the economy to 2065–66, alongside geopolitical fragmentation, the energy transition, population ageing and industrial change. Australian Treasury

Its official fact sheet says AI is likely to support a long-term labour-productivity growth assumption of 1.2 percent a year. It also states that there is not yet evidence of significant labour-market effects, while anticipating that some tasks will be automated and labour demand will change elsewhere. Capital expenditure on Australian data centres has nearly doubled since the 2023 report. Treasury fact sheet

Evidence status: this is an authoritative government projection intended to inform forty-year policy. It is not a causal evaluation of AI adoption, an education study or proof that projected productivity gains will occur. Assumptions about innovation, investment, skills and regulation remain assumptions. The report is consequential because it places AI inside long-term fiscal and social planning; its numbers should be used as scenarios for governance, not as realised benefits.

Productivity is a ratio whose boundaries are moral choices

Productivity compares outputs with inputs. The result changes according to what an institution calls an output, which inputs it counts and how long it observes the system. If an AI service drafts one hundred feedback comments in an hour, the visible ratio looks impressive. If lecturers then spend several hours correcting false claims, learners misunderstand the feedback, an appeal must be investigated and the institution loses assessment capability, the wider account is very different.

The measurement boundary is therefore not technically neutral. Counting documents but not comprehension privileges production. Counting staff hours saved but not unpaid learner data costs privileges the institution. Counting immediate subscription price but not connectivity, integration, verification, security, portability and exit privileges the vendor. Counting aggregate benefit while ignoring who receives it can make inequality appear efficient.

A credible business case should identify the unit of value before adoption. In education, that unit may be durable and transferable competence. In ministry, it may be truthful communication joined to accountable human care. In public service, it may be lawful, accessible and correct service with effective remedy. Output volume is useful operational information, but it should not be allowed to redefine the institution's purpose.

Education must distinguish assisted performance from learning

The OECD Digital Education Outlook 2026 distinguishes improved task performance with generative AI from genuine learning. A learner can produce a stronger immediate answer without developing the knowledge, judgement or metacognition required to perform later without assistance. OECD Digital Education Outlook 2026

A preregistered experiment involving 405 secondary-school learners found stronger comprehension and three-day retention from note-taking alone and from note-taking combined with a chatbot than from chatbot use alone. Learners nevertheless tended to prefer the chatbot and perceive it as more helpful. The study does not establish that every chatbot impairs learning; it demonstrates why convenience, preference and immediate completion cannot substitute for delayed independent performance. Kreijkes and colleagues

Educational productivity should therefore count independent recall, transfer to an unfamiliar problem, source verification, revision quality and the gradual withdrawal of support. If a tool increases completed tasks while learners become less able to explain, adapt or perform without it, the system has improved production while weakening education.

Curriculum should teach whole-value reasoning

AI literacy often begins with prompting, output checking and responsible-use rules. It should progress towards institutional judgement. Learners need to compare a claimed efficiency gain with the educational purpose, affected people, data, labour, infrastructure, alternatives, distribution of benefit, error burden and exit cost.

An authentic assessment can give learners a proposal for an AI tutor, marking assistant, translation service or administrative agent. They should identify the baseline process, calculate direct savings, test output quality, measure unsupported performance, map hidden labour, examine unequal access, estimate recurring and switching costs, design an appeal route and defend adoption, restriction or rejection before a panel.

The assessment should include a delayed task that the learner completes independently. It should also require a revision after receiving evidence from an affected user. The desired capability is not enthusiasm or resistance. It is the ability to make a proportionate decision, disclose uncertainty and change course when the evidence changes.

E-learning can move work rather than remove it

A digital platform may reduce printing, travel and repetitive administration. It may also transfer work to lecturers who must structure content, moderators who verify automated decisions, students who purchase data, families who share devices, administrators who manage accounts and technicians who recover systems. A claimed saving should show whose time was saved and whose time increased.

Useful measures include completion and delayed learning; lecturer preparation and verification time; learner data and device cost; accessibility; response and appeal time; system downtime; privacy and security incidents; local-language quality; portability; and the number of people who can maintain the service independently. Measures should be disaggregated where lawful and useful so that an average does not conceal exclusion.

The September 2026 UNESCO ministerial statement calls for evidence of educational benefit, teacher participation, learner-data rights, inclusion, total-cost analysis, interoperability and the ability to change course. These requirements apply even when a system is donated or initially free. UNESCO

The Global South must count dependence as a cost

An institution may receive a capable system without gaining the authority or knowledge needed to govern it. Models, cloud infrastructure, payment channels, identity systems and technical support may remain elsewhere. Currency movement, changed terms, interrupted connectivity or geopolitical restriction can turn a productive workflow into a dependency the institution cannot repair or leave.

The African Union Continental AI Strategy connects responsible adoption with African agency, infrastructure, skills, culture, inclusion and the ambitions of Agenda 2063. The issue is not isolation from international technology. It is whether participation expands local capability and decision power rather than making African institutions permanent consumers of systems designed, evaluated and governed elsewhere. African Union

A whole-value account should therefore include currency exposure, connectivity, device replacement, cross-border data arrangements, local-language performance, staff capability, intellectual-property terms, portability, alternative providers and continuity when a subsidy or partnership ends. Cheap access can be costly if the institution loses the ability to define its problem or govern its own future.

Kenya's governance evidence shows both progress and unfinished work

The 2026 Global Index on Responsible AI evaluates 135 countries and jurisdictions using publicly documented policy, institutional and civil-society evidence. Its Kenya profile scores the country above the African average and identifies relative strengths in ethics and sustainability, public-sector skills and procurement. Global Index — Kenya

The same profile records no identified framework for labour protections or children's rights within its indicator method, a weak impact-assessment score and no identified requirement for public disclosure of government algorithmic systems. These are index findings based on documented evidence, not a complete judgement on every Kenyan law, institution or lived experience. The index itself warns that scores cannot capture implementation nuance or lived realities.

Globally, the index reports that only 18 percent of countries require disclosure of government AI systems, 29 percent have labour-protection frameworks and 27 percent address AI's environmental effects. It also finds that most environmental frameworks are non-binding. Global Index 2026. The signal is clear: skills and adoption may advance faster than the protections needed to distribute productivity fairly.

Labour productivity must not erase professional capability

AI can remove low-value repetition and release people for judgement, relationship and creativity. It can also deskill work when staff approve outputs they can no longer produce, explain or correct. The risk is especially serious in teaching, assessment, pastoral care and public administration, where professional responsibility cannot be reduced to accepting a generated recommendation.

Institutions should specify which capabilities must be retained internally. Teachers should remain able to design curriculum, interpret evidence and assess learning. Ministry leaders should remain able to listen, discern, refer and take pastoral responsibility. Public officers should understand the mandate, data, rules and reasons behind a consequential decision. Automation should be accompanied by periodic independent performance, rotation of responsibilities and opportunities to practise without the tool.

A useful productivity report should therefore show hours saved, roles changed, new verification work, training, error correction, worker voice, capability retained and whether staffing reductions shifted burden to the public. Efficiency achieved by making responsible people unreachable is not institutional maturity.

CBTE should measure support that leads to independent ministry competence

Competency-based theological education can use AI to help a learner rehearse, organise evidence, explore alternatives or receive formative questions. Yet competence must still be demonstrated contextually across knowing, being and doing under accountable human judgement. CBTE framework

A learner may draft a sermon faster with AI. The programme must still ask whether the learner interprets Scripture responsibly, verifies sources, communicates to the actual community, receives correction, protects vulnerable people and can respond faithfully when the prepared text no longer fits the situation. Mentor observation and community evidence remain indispensable.

CBTE productivity should be measured as the efficient development of trustworthy independent competence, not the rapid accumulation of portfolio artefacts. Useful evidence includes mentor time, reassessment, contextual transfer, ethical judgement, learning retention and the learner's progressively reduced dependence on technological support.

Digital ministry needs a pastoral value account

A ministry may use AI to prepare translations, organise public information, create study questions or draft routine communication. These uses can widen access and reduce avoidable administration. They become harmful when synthetic responsiveness conceals the withdrawal of accountable people or when sensitive prayer, counselling, health, safeguarding and children's information becomes an input to unapproved systems.

Ministry value should include theological faithfulness, accessibility, human presence, consent, correction, referral, confidentiality, relational trust and whether people can participate without the technology. Engagement time, messages sent and content volume are not sufficient ministry outcomes. A chatbot cannot receive revelation, authenticate conversion, establish calling or bear pastoral responsibility.

A pastoral value review should ask what human work was released, whether that time actually returned to people, what data entered the service, who verified theological meaning, how synthetic content was disclosed, which vulnerable person could reach a competent human and how the ministry would continue if the provider disappeared tomorrow.

Boards need a whole-value register, not a savings headline

Every consequential AI proposal should include a Whole-Value Register: purpose; baseline process; affected people; expected benefit; learning or service outcome; direct and hidden costs; data; labour changes; unequal effects; environmental and infrastructure demand; responsible owner; verification; incident and appeal routes; capability retained; recurring cost; portability; and exit.

The baseline matters. An institution cannot credibly claim time saved if it never measured the earlier process. It cannot claim better learning from higher submission rates or better service from more chatbot conversations. Measures should connect to purpose and include a comparison, an observation period and a named person responsible for collecting adverse as well as favourable evidence.

Boards should require a distribution statement: who gains time, income, access or capability; who carries new cost, surveillance, uncertainty or exclusion; and who can challenge the result. A positive average should not authorise scaling when one group repeatedly carries serious and preventable harm.

Research should test the route from efficiency to public value

A useful conceptual pathway is: AI capability → workflow redesign → task-level efficiency → distribution of work and benefit → retained human capability → educational, ministry or public outcomes → institutional resilience and trust.

Potential mediators include staff AI literacy, baseline measurement, curriculum design, teacher and mentor presence, data governance, procurement capability, worker and learner participation, connectivity, verification and leadership responsiveness. Potential moderators include language, disability, rurality, shared devices, institutional size, task stakes, vendor dependence and regulatory capacity.

An African comparative study could examine conventional practice, unstructured general-purpose AI use, bounded AI assistance and a whole-value model with independent evaluation. Outcomes should include time, full cost, delayed learning, unsupported performance, professional capability, access, local-language reliability, privacy incidents, energy and connectivity burden, appeal, switching capability and trust—not output volume alone.

Christian stewardship counts what efficiency leaves outside the frame

Christian stewardship welcomes gifts and asks how they serve God and neighbour. It does not glorify inefficiency, nor does it treat growth as self-authenticating. The biblical concern for truthful weights, just labour, care for the vulnerable, wise use of resources and accountable authority challenges every narrow account that records benefit while hiding transferred cost.

Pentecostal spirituality should add a particular warning. Visible multiplication, rapid expansion and fluent testimony can be mistaken for divine endorsement. Claims are tested through Scripture, community, character and fruit. An AI system has no calling, anointing or moral agency. Its productivity remains subordinate to the people who must govern its use and answer for its consequences.

A faithful institution should be able to say where AI genuinely helped, where it failed, who paid, what was learned and why a use was continued, repaired or stopped. Truthful limitation is not hostility to innovation. It is a form of stewardship that makes responsible innovation possible.

Count the whole institution before declaring success

The new intergenerational report is right to treat AI as a transition requiring long-range planning. Its productivity projection becomes useful when it opens inquiry rather than closing it. The central task is to connect national scenarios with evidence from actual workplaces, classrooms, churches and public services.

Institutions should begin with one bounded workflow and measure the baseline. They should define the outcome that matters, include affected people, test learning or service independently, count hidden labour and infrastructure, preserve human capability, disclose uncertainty and establish a clear stop rule. Scaling should follow verified public value, not excitement or a persuasive demonstration.

AI productivity must count the whole institution. When learning, labour, dignity, resilience, environment, local agency and remedy enter the account, efficiency can become genuine stewardship. When they remain outside it, a faster institution may simply become a more efficient way of transferring costs to people with less power.