Speed Is Not an Educational Outcome

Digital transformation is often narrated as a race. The fastest institution appears innovative, the earliest adopter looks courageous and the leader who hesitates risks being described as resistant. Artificial intelligence intensifies that pressure because a service can be activated in minutes while curriculum, staff capability, evidence, law and governance move much more slowly.

Speed, however, is not an educational outcome. It does not demonstrate learning, formation, inclusion, trustworthy ministry or public value. A rapidly deployed system may improve access or staff productivity, but it may also move confidential information, obscure authorship, weaken independent competence, create new workload and become embedded before affected people can understand or challenge it.

Responsible institutions therefore need more than a choice between acceleration and prohibition. They need institutional pacing: the capacity to test, limit, pause, learn, repair, resume and exit. The proper speed is the speed at which educational purpose, human capability, rights protection, contextual evidence and correction can keep up.

The newest signal is an argument for pacing the frontier

Reuters reported at 04:00 UTC on 14 September that AI-linked shares fell in early Asian trading after leaders of prominent frontier laboratories called for slower development amid safety concerns. The report says Anthropic chief executive Dario Amodei urged a reduction in the rate of capability advancement, while the leaders of OpenAI and xAI publicly expressed agreement with parts of the proposal. Reuters, 14 September

Amodei's primary essay proposes three levels of action: embedded third-party evaluators with substantial access inside frontier companies; coordination among laboratories in democratic countries; and international coordination. He says pacing should not mean stopping technical progress, but creating time for safety work and requiring outside confirmation. Anthropic commits in the essay to invite an embedded external review team. Dario Amodei

These are consequential public commitments and proposals, not verified outcomes. The essay is written by the chief executive of a company with commercial, regulatory and geopolitical interests. Agreement expressed by competitors is not an enforceable industry standard, and a fall in share prices is not evidence that AI became safer. The promised reviewer access, independence, reporting freedom and practical effects will need evidence over time.

A Global South reading must ask who sets the pace

The call for pacing deserves serious attention without allowing a small group of powerful companies to define the world's risk agenda. The primary essay frames part of the problem through competition between the United States and China. African institutions should neither dismiss the identified dangers nor accept that geopolitical advantage is the natural organising principle of global AI governance.

Communities in the Global South already experience forms of technological risk that rarely dominate frontier debates: extraction of data without reciprocal value, weak local-language performance, unaffordable connectivity, imported curriculum assumptions, invisible content labour, inaccessible systems, externally controlled infrastructure and dependence on services whose terms can change without local consent.

The African Union Continental AI Strategy places African agency, inclusive development, rights, infrastructure, skills and governance within one continental framework. African Union Its significance for pacing is clear: the question is not only how quickly models become more capable, but whether African people and institutions can understand, shape, evaluate, govern and, when necessary, refuse their use.

Institutional pacing is not delay without purpose

A pause can be responsible or evasive. Responsible pacing identifies a defined risk, protects affected people, creates a bounded period for investigation, assigns authority, gathers evidence and ends with a defensible decision. Evasive delay postpones responsibility, leaves informal use untouched and produces no new capability.

Nor should every use move at the same speed. Drafting a fictional practice question presents different stakes from deciding admission, grading a final assessment, analysing a safeguarding disclosure or advising a person in crisis. Proportionality requires faster experimentation in low-stakes, reversible settings and much stronger evidence before high-stakes, intimate or difficult-to-reverse deployment.

A practical pacing chain is: purpose, affected people, risk level, minimum evidence, bounded pilot, independent review, human capability, decision, monitoring, correction, and exit. Each stage should have an accountable owner and a threshold that must be met before moving forward.

Education already has reasons to resist adoption speed

UNESCO's ministerial statement adopted during Digital Learning Week calls for evidence of educational benefit, public oversight even when AI is free or donated, teacher participation, age-appropriate safeguards, data rights, inclusion, total-cost analysis, interoperability and the ability to change course. UNESCO ministerial statement

The statement is non-binding and does not prove implementation. It nevertheless supplies an important governance principle: ease of acquisition is not permission to skip educational or public-interest scrutiny. A free account can cross the classroom faster than an institution can decide what learning it serves, what data it uses or which teacher has authority to correct it.

UNESCO's practical guide for AI in technical and vocational education recommends human oversight, institutional readiness, staff capability, data governance, evaluation and review against legal and ethical standards. UNESCO-UNEVOC These are not obstacles placed after innovation; they are the conditions that allow innovation to become responsible institutional practice.

Educational benefit must be demonstrated after the novelty fades

The OECD Digital Education Outlook 2026 concludes that generative AI can support learning when guided by clear pedagogical principles, but can improve task performance without creating corresponding learning. OECD An institution therefore cannot treat speed, usage, polished output, satisfaction or time saved as sufficient evidence of educational benefit.

A preregistered randomised study involving 405 secondary-school learners found stronger comprehension and three-day retention for note-taking alone and note-taking combined with an AI chatbot than for chatbot use alone. Learners nevertheless tended to prefer and perceive the chatbot as more helpful. Kreijkes and colleagues

One experiment cannot establish the effect of every chatbot, discipline or learner population. It does demonstrate why implementation should be paced by independent performance and delayed evidence rather than preference alone. A tool may feel useful while doing part of the intellectual work the learner needed to undertake.

Curriculum should teach responsible acceleration and responsible refusal

AI literacy is incomplete when learners can prompt, generate and automate but cannot determine whether a system should be used. Curriculum should develop a progression from operating AI to questioning it, contextualising it, governing it, creating responsibly and demonstrating competence independently.

Learners should practise identifying purpose, verifying sources, protecting data, detecting contextual bias, disclosing assistance, defining human decision rights and choosing among adoption, restriction, pause and refusal. The capacity to say 'not yet' or 'not for this task' is a positive professional competence when supported by evidence and an alternative pathway.

A systematic review of 28 empirical studies found that AI most commonly acted as assistant or content generator in teacher-AI learning-task design, with less evidence of genuinely dialogic and context-sensitive collaboration. Wang and Liu The finding supports more deliberate design, not a universal slowdown: pace should create room for teachers and learners to shape the relationship rather than inherit it from a default interface.

Assessment needs checkpoints that AI cannot silently pass

Assessment should reveal what the learner knew before assistance, how evidence was gathered, what AI contributed, how errors were identified, why revisions were made and what the learner can perform without the system. An end product alone becomes increasingly weak evidence of authorship, reasoning and capability.

Useful checkpoints include an initial independent attempt, a source record, disclosed prompts or material assistance, teacher or peer feedback, an oral defence, an unfamiliar task and later demonstration in context. The checkpoints should be proportionate; universal surveillance or unnecessary biometric verification would create a different form of harm.

Pacing also applies to assessment redesign. Institutions should not replace every essay with oral examination overnight or assume that detection software can resolve authorship. Pilot new methods, examine workload and accessibility, record false concerns, moderate judgements and provide appeal before high-stakes use.

Competency-based theological education can pace support towards independence

Competency-based theological education offers a relational way to govern AI because evidence is distributed across learning, practice, mentoring, reflection and contextual performance. The published CBTE model centres collaborative participation, contextualised outcomes, integrated formation and demonstrated competence. CBTE

AI may support rehearsal, feedback, translation, planning or access. It should progressively withdraw where the learner must interpret Scripture, exercise pastoral judgement, safeguard a vulnerable person, preach, teach, reconcile conflict or lead a ministry process. A platform can coordinate evidence; it cannot witness character or contextual transfer by itself.

Pacing within CBTE means defining the competence, permitting bounded assistance, observing supported practice, reducing the scaffold and requiring authentic performance. Faculty and mentors retain responsibility for standards, feedback, moderation and reassessment. The goal is not AI-free ministry but a minister who remains capable and accountable when the system is absent or wrong.

Theological education must distinguish prudence from fear

Christian prudence is neither technological enthusiasm nor reflexive rejection. It asks what kind of good is sought, which neighbour may be affected, what power is being transferred, what evidence is credible and who remains responsible. It can move quickly to relieve a real burden and slowly where haste threatens truth, dignity or relationship.

Pentecostal discernment adds an important account of time and authority. The Spirit can interrupt established plans, distribute gifts beyond official expertise and lead communities into new mission. Yet claims are tested through Scripture, communal judgement, character and fruit. Immediacy is not infallibility, and technological fluency does not become spiritual authority.

An AI system can generate a sermon outline, prayer or theological explanation at great speed. It cannot receive revelation, authenticate conversion, recognise calling or assume pastoral responsibility. The faster religious language becomes available, the more deliberately ministers must preserve source truthfulness, human presence and accountable discernment.

Digital ministry needs red lines before the system is switched on

Low-risk uses such as formatting a public announcement are not equivalent to processing prayer requests, counselling records, safeguarding reports, children's information or personal testimony. Digital ministry should define prohibited data and decisions before staff or volunteers begin experimenting.

Synthetic images, voices, testimonies and teaching should be disclosed when their generated nature is material. Crisis, abuse, health, legal and financial matters require referral to competent people. Automated prophecy, spiritual diagnosis and companion-style pastoral substitution should not be normalised merely because a system can sustain convincing conversation.

A ministry pause protocol should identify who can suspend a feature, preserve relevant records lawfully, notify affected people, correct public misinformation, provide human care and decide whether use may resume. A platform incident becomes a pastoral matter when a person entrusted the ministry with vulnerability.

Kenyan law already requires slower thinking around high-risk data

Kenya's Office of the Data Protection Commissioner applies data-protection duties to educational institutions and remote-learning services. Its education-sector guidance addresses lawful and transparent processing, purpose limitation, data minimisation, privacy by design, due diligence, automated decision-making and high-risk data-protection impact assessment. Kenya ODPC

The guidance is not an AI-specific approval list, and legal compliance does not prove educational benefit. It does mean that an institution cannot defend careless deployment by saying the technology moved faster than its policy. Where processing is likely to create high risk to rights and freedoms, assessment and safeguards belong before the decision, not after harm.

Educational and ministry leaders should therefore map the controller and processor, purpose, data, transfers, retention, security, human review, correction and complaint route. High-risk processing should not proceed merely because a lecturer, volunteer, donor or vendor has already opened an account.

Independent evaluation must be independent enough to matter

The newest frontier proposal makes independent evaluators central. Education should receive the principle while asking harder institutional questions: who selects and pays the evaluator, what access is granted, which evidence may be published, which conflicts exist and whether affected learners or communities can contribute to the evaluation.

The NIST AI Risk Management Framework organises work around governing, mapping, measuring and managing risk, while its resource centre supports testing, evaluation, verification and validation. NIST AI Resource Center The framework is voluntary and general; local leaders must translate it into educational, pastoral and legal responsibilities.

A vendor should not be the sole trainer, evaluator and interpreter of its own pilot. Institutions can use internal review for low-stakes experiments, but consequential systems need credible separation between commercial promotion, technical assessment, educational evaluation and the final authorised decision.

Governing bodies need explicit pause and restart authority

Many AI policies describe acceptable use but do not say who can stop a system. A responsible governance instrument should identify the operational owner, professional authority, data-protection role, approving body, incident lead and person authorised to suspend processing immediately when necessary.

Pause triggers may include prohibited-data exposure, unexplained adverse decisions, material hallucination in a high-stakes task, security failure, inaccessible design, discriminatory effects, loss of human review, vendor terms changing, inability to export records, staff inability to verify outputs or evidence that learning is being displaced.

Restart should require more than the passage of time. Leaders should document the cause, affected people, containment, remedy, revised configuration, independent check, staff preparation, communication, monitoring period and residual risk. Some systems should be repaired and resumed; others should remain restricted or be exited.

A seven-gate protocol can pace institutional adoption

Gate one is purpose: name the educational, ministry or public problem and consider a non-AI alternative. Gate two is proportionality: classify the stakes, reversibility, affected people and data. Gate three is readiness: establish staff capability, infrastructure, language, accessibility, cost and lawful authority.

Gate four is evidence: define the learning, service, equity and harm measures before starting. Gate five is a bounded pilot: limit users, duration, features and data while preserving an accountable human pathway. Gate six is independent review: examine results, incidents, unequal effects, workload and affected-person experience without allowing adoption pressure to predetermine the conclusion.

Gate seven is an authorised decision: adopt, restrict, pause, redesign or exit. The record should state conditions, responsible officers, review date, appeal route, continuity arrangements and the evidence that would trigger another decision. Pacing becomes governance when each gate can genuinely stop progression.

Research should compare deployment speeds, not assume one ideal pace

Research should compare immediate unrestricted adoption, fixed-period pilots, risk-tiered pacing and complete pause-and-resume governance. The central question is not whether slower is always better, but which sequence produces durable capability, protects rights and allows institutions to correct error before dependence becomes difficult to reverse.

Outcomes should include delayed independent learning, contextual transfer, source-verification accuracy, staff workload, learner access, privacy incidents, unequal effects, correction time, appeal outcomes, recurring cost, retained institutional capability and trust. Adoption rates and time saved are relevant but insufficient.

African comparative work should examine language, rurality, shared-device use, disability, donor dependence, institutional size, technical capability and cross-border data. It should include people who declined or could not access the system, not only enthusiastic users. Participatory evidence is essential because the people carrying implementation cost may see risks that dashboards conceal.

The right speed is accountable speed

The newest industry debate confirms that capability can advance faster than safeguards, evaluation and public deliberation. It does not follow that frontier-company leaders should decide the pace for everyone or that every educational use should stop. It does follow that institutions acting as passive consumers are poorly prepared for technologies whose capabilities and terms change rapidly.

Educational and ministry organisations need the authority and competence to choose their own pace within law and public responsibility. Low-stakes experimentation can remain open and creative. High-stakes uses should wait for evidence, capable people, data protection, human review and remedy. A pause should produce learning; a restart should be earned.

Speed is not an educational outcome. Responsible innovation leaves learners more capable, teachers more authoritative in curriculum, communities more able to question power and institutions more prepared to correct harm. The strongest organisation is not the one that adopts first, but the one that can explain why it moved, what it learned and when it will stop.