Position paper · May 2026

Human Literacy in an Age of Non-Human Intelligence

A provocation for colleges, awarding bodies, employers and policymakers

Stephen Kittner · 2026 · Comments and challenge welcome — respond here

Abstract

1. The Old Bargain Has Collapsed

The last two years have compressed what previously looked like a decade of technological change. Generative AI has moved from writing assistant to multimodal agent, software builder, research assistant, image generator, tutor, analyst and workflow automator. Stanford's 2026 AI Index reports US private AI investment reaching $285.9 billion in 2025 — a figure that illustrates not just commercial momentum but the scale of capital now driving capability growth.¹

Education systems, however, still largely behave as if human cognitive scarcity remains the foundation of economic value. Schools, colleges and universities rank, sort and credential learners according to their capacity to recall, reason, write and calculate under controlled conditions — precisely the domains where AI is advancing most rapidly. The World Economic Forum's Future of Jobs Report 2025, drawing on over 1,000 employers representing more than 14 million workers across 55 economies, identifies AI and technological change as the central forces reshaping work between 2025 and 2030.²

To be precise about the disruption: AI currently exceeds human performance in specific, well-defined cognitive tasks — pattern recognition, text generation at volume, data retrieval, statistical modelling, certain forms of coding and translation. It assists effectively across a much wider range. It continues to fall short in others — persistent physical manipulation, genuine moral reasoning, long-horizon planning under deep uncertainty, and the relational attunement that good teaching, care work and professional judgement require. Reliable judgement, contextual understanding, embodied skill, institutional wisdom, ethical accountability and genuine care remain scarce human capacities. We are not comprehensively past human cognitive supremacy. We are in a transition in which AI is removing the economic scarcity premium from a growing range of cognitive tasks — and that transition is accelerating.

What is education for when cognitive performance is no longer uniquely or scarcely human?

The weak answer is "critical thinking." The stronger question is: what kinds of human beings, communities and societies do we need when cognitive power is externalised at scale? That question applies across all phases of education but with particular urgency in further education, where disruption hits first, hardest and with the least institutional protection.

2. What Education Has Historically Done

Education has never had a single purpose. It has always been a bundle of functions and understanding that bundle is essential for deciding what to preserve, redesign and let go.

Culture transmission. Education inducts young people into language, stories, values, histories and shared references. It is society reproducing itself across generations.

Cognitive development. Literacy, numeracy, scientific reasoning and disciplinary knowledge expand perception. A learner who understands ratio, evidence, causality and material behaviour sees a different world from one who does not.

Sorting and signalling. Qualifications help allocate opportunity. Education is simultaneously liberation and gatekeeping — and the gatekeeping has historically reproduced advantage unless actively contested.

Socialisation. Learners encounter friction: disagreement, compromise, authority, failure, collaboration, belonging. This psychological curriculum is often more powerful than the formal one.

Identity and belonging. For many learners in FE, the qualification is not the primary point. The point is the first environment in which they have been taken seriously as a thinker, a craftsperson, a professional-in-training. The erosion of that experience through underfunding and credentialism is a form of waste the AI age makes more, not less, urgent to address.

These functions are now under simultaneous pressure. If AI tutors can deliver effective personalised instruction, the knowledge-transmission function changes. If employers can assess capability through AI-mediated work trials, some credential signalling weakens. If automation reduces demand for routine cognitive labour, education as workforce preparation becomes morally thin without a stronger account of purpose.

3. What AI Changes: Five Distinct Disruptions

Five distinct disruptions are frequently blurred together in AI-in-education debate. Separating them produces clearer thinking and more targeted institutional responses.

AI as curriculum content. Learners need to understand how AI works, where it fails, what it assumes and what it cannot do. This is necessary but insufficient as an educational response. Knowing how a tool works does not constitute a philosophy of education.

AI as delivery infrastructure. Adaptive tutoring, AI feedback tools, accessibility support and simulation can genuinely improve learning, particularly for learners previously underserved by one-size provision. The evidence base is promising but uneven; benefits are most clearly established in specific, well-defined domains, and the gap between laboratory efficacy and large-scale deployment — particularly for learners with complex needs — remains significant. Optimism is warranted; credulity is not.

AI as institutional operating system. Administrative burden — registers, progress reviews, compliance reporting, curriculum documentation — consumes enormous amounts of educator time. Automating the administrative to recover the human is the most clearly legitimate application of AI in under-resourced institutions. Any institution using AI to reduce learner-facing contact hours rather than administrative burden is moving in the wrong direction.

AI as labour-market disruptor. The WEF Future of Jobs Report 2025 identifies analytical thinking, resilience, leadership, social influence and AI and data literacy among the fastest-growing employer demands through 2030.² What employers increasingly need is not faster task execution but better judgement, communication and professional reasoning — precisely as AI handles more of the routine work beneath those capacities. The qualification offer must respond to this shift, not to a labour market that no longer exists.

AI as philosophical rupture. The deepest disruption is not technical but existential: AI forces a reckoning with what education was actually for. If it was always primarily about information transfer and task execution, then AI renders much of it obsolete. If it was always, at its best, about the formation of human beings capable of living and judging well — then AI clarifies and intensifies that purpose rather than displacing it.

The central equity risk runs through all five disruptions: AI does not enter education uniformly. For advantaged learners it offers enrichment — better tutoring, faster feedback, richer resources. For disadvantaged learners in under-resourced institutions, the risk is automated containment: AI-mediated delivery substituted for human contact, dressed as personalisation. That distinction should be the first question asked of any AI adoption decision in FE — and it is developed fully in Section 8.

4. The Evidence: What We Know and What It Implies

The case for education does not rest on time in provision alone. The World Bank's 2018 World Development Report made the distinction explicit: schooling is not the same as learning, and many systems deliver attendance without foundational skill development.³

Quality matters more than quantity. Hanushek and Woessmann's work establishes a strong relationship between genuinely developed cognitive skills and long-term economic outcomes — the "knowledge capital" of a nation, rather than years of schooling, drives long-run prosperity.⁴ As AI makes information delivery cheaper, institutions that cannot demonstrate what they uniquely develop in human beings are likely to find their educational value increasingly difficult to defend.

PISA 2022 recorded the largest single-period decline in OECD educational performance: mathematics falling 15 points and reading 10 points between 2018 and 2022 — equivalent to roughly three-quarters of a year's learning.⁵ The decline predates the current AI wave but indicates systems under strain before disruption accelerates.

High-performing systems — Singapore, Estonia, Japan, Korea — succeed through alignment: coherent curriculum, serious teacher development, stable qualification policy and national purpose operating together. For English FE, the lesson is that fragmentation degrades performance. English post-16 education has experienced repeated qualification reform cycles, funding formula changes and inspection revisions that make coherent institutional development extremely difficult.

The IFS has tracked the consequences of structural disinvestment: college spending per student fell by 14% between 2010–11 and 2019–20, with wider post-16 funding pressures continuing to constrain the system.⁶ That figure translates directly into a staff recruitment and retention crisis. FE college teachers already earn on average £7,000 less per year than their school counterparts⁷ — and that pay gap is widest precisely in the specialist vocational areas where industry salaries dwarf FE scales: engineering, construction, digital and advanced manufacturing. These are exactly the departments colleges most need to protect, because they house the professionals qualified to teach the capabilities that survive automation — engineers who understand design intent and safety accountability; software developers who can teach system architecture and code review; project managers who can model professional judgement under constraint. When these posts sit unfilled for months or are covered by fractional contractors who cannot carry a full pastoral relationship, the temptation to deploy an adaptive learning platform is not a failure of leadership character. It is an existential structural pressure built into the funding model — and it is targeted, whether by design or default, at the highest-value, hardest-to-replace teaching in the system.

Understanding this matters because it changes the moral framing of automated containment: the executive teams reaching for AI platforms are often not choosing convenience over principle; they are choosing institutional survival over a philosophical position the funding settlement gives them no safe room to hold. The policy implication is direct — the funding methodology must change before we can reasonably expect institutions to absorb the risks that genuine human formation requires.

5. Why "Critical Thinking" Is Not Enough

The current policy reflex — AI makes critical thinking more important, so teach more of it — is partially right and strategically insufficient.

It is partially right because learners do need to interrogate AI outputs, understand algorithmic bias and avoid automation complacency.

It is insufficient for three reasons.

First, it is poorly defined in practice. In most institutional settings "critical thinking" means "evaluate sources" or "write a balanced argument." It does not reliably mean reasoning across competing incentives, incomplete evidence, values and long-run consequences. When a curriculum document claims to develop critical thinking, it usually means something narrower than the phrase implies.

Second, it depends on domain knowledge. A learner cannot critically evaluate AI-generated engineering analysis, medical reasoning or financial advice without enough subject understanding to recognise what matters and identify what is wrong. Skills without substance collapse into vibes. The answer to AI-generated content is not scepticism in the abstract; it is deep enough internal fluency to know where to probe.

Third, AI will increasingly perform the critical operations itself. Advanced systems already compare evidence, detect contradictions, model uncertainty and generate counterarguments. The human role cannot be to serve as a slower fact-checker of a faster intelligence. That is not an educational outcome; it is a holding position.

Critical thinking asks whether something is valid. Judgement asks what matters, who is affected, what risks are acceptable, and what kind of decision should be made.

The future requires judgement. AI may assist with analysis; humans remain accountable for the consequences of acting on it.

6. The New Purpose: Human Formation

Human formation is the organising concept this paper proposes. It means the deliberate integration of knowledge, skill, identity, ethics and agency so that learners can act responsibly in real contexts.

It is not a new label for personal development, character education, employability skills, PSHE, tutorial programmes, Gatsby benchmarks or "British values" compliance — though it encompasses the best of all of them. Its distinctive feature is that it changes the organising question of education from "Can the learner complete the task?" to "Is the learner becoming capable of responsible judgement in a world where tasks can increasingly be automated?"

One tension requires explicit resolution before the five components are described. This paper argues both that deep domain knowledge matters (Section 5) and that the value of subjects shifts away from task execution towards habits of mind (Section 6.5). These are not contradictory positions, but the relationship between them must be stated precisely: internal fluency is a prerequisite for external judgement, not an alternative to it. A learner who relies entirely on AI for foundational technical calculations cannot develop the professional intuition — the gut-check that precedes and enables expert critique — that distinguishes competent practitioners from sophisticated prompt operators. The shift described in this paper is not from knowledge to process, or from content to skills. It is from knowledge-as-terminal-product to knowledge-as-foundation-for-judgement. Civilisational literacy and domain fluency are not the least important components of human formation; they are the substrate without which all other components lack purchase.

Five components constitute human formation in an AI age.

6.1 Agency

When AI can perform almost any defined task on request, the scarce human capacity becomes wanting wisely — forming intentions, not merely completing briefs. Traditional education too often trains compliance: wait for the task, meet the criteria, submit the work. A learner whose only capability is executing clearly specified tasks has no durable advantage over a well-prompted language model. Agency requires learners to frame problems worth solving, choose goals worth pursuing, sustain effort when the path is unclear and take responsibility for what they produce.

6.2 Ethical Judgement

AI creates abundance: infinite explanations, drafts, images and simulations. Abundance generates ethical questions that scarcity never forced: What should be made? What should be automated? What should remain human? What should not be optimised, and on whose behalf? Ethics must become a core operating system running through every curriculum area. In engineering, this means safety culture, professional liability and consequential responsibility. In business, stakeholder impact beyond shareholder return. In health and social care, understanding what cannot be delegated to a machine without moral loss.

6.3 Professional Identity

Learners need a durable sense of who they are professionally: not just what tasks they can perform, but what standards they hold, what responsibility they carry, and what it means to be a competent and trustworthy practitioner. Professional identity is what survives when the specific task portfolio changes. It requires embodied experience, real consequence and the judgement of other practitioners over time — none of which AI can replicate or shortcut.

6.4 Relational Competence

Human life remains embodied and relational regardless of what AI can do. Learners need emotional regulation, conflict repair, care, collaboration and the ability to read and respond to other people in real time. In most vocational contexts these are not soft extras — they are the core professional skill. A health and social care worker whose relational competence is poor cannot be rescued by technical knowledge. An engineer whose communication fails can cause harm regardless of technical ability.

Vocational education has understood the integration of hand, mind and professional identity better than academic education. That instinct is not a concession to learners who cannot manage academic work; it is a sophisticated educational model that the academic mainstream should be borrowing from.

6.5 Civilisational Literacy

Learners need enough science, mathematics, history, economics, ecology, politics and technology to understand the systems they inhabit — not to outperform AI in each domain, but because democratic agency requires an intelligible world. A population that cannot understand energy systems, climate science, statistical reasoning, public debt or AI itself is governable but not free.

The case for individual subjects becomes stronger, not weaker, in the AI age — but the justification changes. Mathematics matters less as computational execution and more as a discipline of structure, abstraction, proof and logical constraint. Writing matters less as text production and more as the clarification of thought, voice, intention and selfhood. When a learner struggles to articulate what they mean, they are not failing to operate a text-production tool; they are encountering the irreducible difficulty of becoming someone with something to say. Science matters less as fact retrieval and more as disciplined doubt, model-building and intellectual honesty.

6.6 Human Formation Across Vocational Pathways

The five components do not apply equally across all vocational contexts — and being honest about the differences matters for curriculum design. Physical and relational trades — construction, electrical installation, plumbing, engineering, early years practice, health and social care — operate in domains where embodied presence and physical judgement remain genuinely difficult to automate. These pathways are not permanently safe from disruption, but they currently afford time to develop the human formation curriculum before displacement arrives in full.

The situation is categorically different for office-based, digital and analytical vocational pathways. Business administration, accountancy, legal services, entry-level data processing and routine digital production face an immediate cliff edge. Table 1 outlines how traditional tasks across key vocational pathways must evolve into domains of higher-level human judgement.

For these pathways, incremental curriculum adjustment is insufficient. What is required is a reconception of what the qualification is actually developing: the professional judgement, ethical reasoning and relational intelligence that sits above the task and remains human even when the task does not.

The Architecture of Human Formation

The five components are not independent modules. They form an integrated structure in which civilisational literacy and domain fluency constitute the substrate, relational competence the connective tissue, professional identity and ethical judgement the operating principles, and agency the governing capacity — all converging in human judgement: the accountable practitioner who can act responsibly in conditions of genuine uncertainty.

Table 2 translates this architecture into inspectable FE practice — giving leaders the evidence markers that correspond to each component so that human formation can be named, tracked and protected rather than assumed.

7. Regulation, Proxies and the Limits of Current Accountability

For understandable reasons, national regulation has relied heavily on measurable proxies: retention, achievement, attendance, compliance and progression data. These indicators matter. They reveal real problems when they deteriorate and genuine strengths when they improve. But they are not the same as the deeper educational formation this paper describes, and for a generation the sector has risked conflating spreadsheets with standards.

The current policy response to AI is reinforcing this tendency rather than challenging it. Four specific tensions deserve direct naming.

The Literacy and Mathematics Problem. The DfE continues to treat functional skills, and GCSE resits as the primary arbiters of post-16 literacy and numeracy. The case for writing and mathematics is, in fact, stronger in the AI age — not weaker. But the justification must change. Mathematics matters not as computational execution but as structure, abstraction and proof. Writing matters not as text production but as clarification of thought, voice and intent. Current resit policy can inadvertently narrow mathematical and written learning towards exam survival rather than genuine reasoning — training learners to compete with large language models on the LLMs' home turf, in conditions where those learners have already demonstrated they struggle. Maintaining rigour while changing the method is not a softening of standards. It is the appropriate response to a changed environment.

The Assessment Contradiction. At the precise moment authentic, contextualised assessment becomes most educationally necessary, regulatory concern about AI malpractice is driving a retreat towards invigilated, pen-and-paper examinations. That retreat prioritises the security of the credential at the expense of its validity. Institutions must navigate this contradiction honestly — maintaining controlled assessment where required, while building the assessment models that genuine capability now demands.

The Accountability Blind Spot. Ofsted's renewed inspection framework, which came into effect on 10 November 2025 and replaced the previous single-word grade with a multi-area report card format, contains concepts that could accommodate human formation — curriculum coherence, personal development, behaviour and attitudes, leadership and governance.⁸ The problem is not total absence. The problem is that AI-age formation is not yet named clearly enough, evidenced consistently enough, or protected sufficiently from subordination to more familiar performance indicators.

This paper argues that human formation outcomes should sit explicitly within two of the new report card judgement areas: the Quality of Education judgement, evidenced through curriculum design that deliberately sequences knowledge towards professional judgement rather than task competence; and the Personal Development judgement, evidenced through learner agency, ethical reasoning, professional identity and the capacity to use powerful tools responsibly.

The A-Level Problem. There is a deeper sorting dysfunction that FE understands better than any other part of the system — but has been reluctant to name publicly. A-levels were never primarily a learning qualification. They were a sorting mechanism: a reliable proxy for the social background and prior educational advantage that correlates with degree completion and professional entry. Universities use them not because they are the best predictor of capability, but because they are a convenient filter.

This argument requires precision. Invigilated, closed-book A-level examinations — particularly those requiring multi-step mathematical proof, unseen textual analysis under strict time constraints, or applied scientific reasoning — retain genuine integrity value against AI substitution in ways that coursework and extended essays do not. A student cannot prompt an AI in a supervised examination hall. That controlled integrity is real, and this paper does not dismiss it. The problem is not that invigilated A-level exams are worthless. The problem is that they measure a particular kind of high-pressure compliance performance — recall, mark-scheme application, examination technique — which is not the same thing as the capability they are used to signal, and which AI has now made replicable in every unsupervised context outside the examination hall. The credential's sorting function depended on its scarcity and its correlation with social advantage, not on its validity as a measure of genuine formation. That distinction matters when deciding what to preserve and what to challenge.

This matters for FE because the consequences of this exposure flow downwards. A strong vocational learner with a well-evidenced project and a compelling professional discussion may be demonstrably more ready for university-level work than an A-level student with three Bs who has never defended a position under pressure or produced anything with real-world consequences. The T Level was supposed to begin closing this gap. It has not, largely because universities did not trust it and UCAS points allocation kept it subordinate. Assessment reform cannot succeed without confronting the structural status hierarchy it is embedded within.

8. Assessment, Qualification and Equity

The Assessment Paradox

The direction of authentic assessment is clear: T Level industry placements, apprenticeship End-Point Assessments, portfolio models and project-based qualifications all point towards evaluation of real capability in context. But awarding bodies and regulators, alarmed by the undetectability of AI-assisted submission, are retreating towards closed-book, invigilated examinations as the primary guarantor of integrity. JCQ guidance (revised April 2025) requires centres to address AI use in malpractice policies and take reasonable steps to protect assessment integrity.⁹ That requirement is legitimate. The response it is producing is not. The system is not protecting standards. It is protecting a ritual.

The Grade Inflation Spiral

Before any new model can be proposed, an uncomfortable structural truth must be named. The current unit-by-unit accumulation model does not just fail to measure genuine capability — it actively produces the appearance of capability where it may not exist.

The logic is systemic and everyone inside the sector understands it. Colleges need achievement rates for Ofsted and funding. Lecturers need pass rates to justify their programmes. Learners need qualifications to progress. So, everyone has a structural incentive to move learners through the gate — and the qualification becomes a processed product rather than a genuine threshold. Nobody acts in bad faith. The incentives produce the outcome collectively without anyone choosing it individually.

The consequence is a cohort that arrives at the next level unable to do the thing the previous qualification claimed they could do. The next level then inherits the same pressure and repeats the process. AI has made this worse. A learner who could previously be moved through a written submission with modest support can now produce a fully compliant piece of work with minimal intellectual engagement. The gap between what qualifications claim and what learners can do is widening precisely as the consequences of that gap are becoming most severe.

The Threshold Model: Assessing Destination Readiness

The solution is not simply tougher marking of the same model. It is a different question.

Current qualifications answer: did the learner complete the programme?

The right question is: is the learner ready for what comes next?

This paper proposes a threshold model of qualification, drawing on and extending the apprenticeship End-Point Assessment framework. Rather than assessing against internally defined unit criteria, assessment should be organised around three legitimate destination thresholds:

• University readiness — the learner can engage with degree-level study in their domain: framing problems, sustaining independent enquiry, defending reasoning and handling intellectual uncertainty
• Employment readiness — the learner can operate as a productive, professional and accountable practitioner in their vocational area at the level the labour market requires
• Apprenticeship progression — the learner has the foundational knowledge, work-readiness and professional awareness to benefit from structured development within an employer context

At Level 2, a fourth threshold applies: Educational progression readiness — the learner has the domain fluency, study skills and professional orientation to succeed at Level 3.

These thresholds relocate the validity question. Instead of a qualification validating itself against its own internal criteria — a circular and increasingly meaningless exercise — it is validated against the real-world destination it claims to prepare learners for. That is a standard that is far harder to game and far more honest about what the qualification is actually for.

The practical instrument for assessing these thresholds is a single substantial learner-derived project, culminating in a human-led professional discussion or viva. This mirrors the EPA model that apprenticeship already uses — and which nobody seriously questions as a valid measure of occupational competence. The professional discussion tests ownership, reasoning, uncertainty, ethical judgement, domain fluency and the learner's ability to respond when challenged. It cannot be faked. Either the learner can hold a professional conversation about what they have done and why, or they cannot.

One objection must be confronted directly here: if the economic and regulatory incentives remain unchanged — if colleges still live or die by achievement rates — what prevents a pressured assessor from passing a borderline viva just as they currently pass a borderline portfolio? The threshold model reduces but does not eliminate this risk. What it adds, critically, is an auditable evidence trail. The AI-supported preparation layer — evidence mapping, gap flagging, structured briefing — creates a documented record of what the assessor saw and what questions they asked. Internal quality assurance and external moderation can scrutinise that record in a way that no portfolio of unit submissions currently enables. The economic gravity does not disappear. The accountability architecture around it becomes significantly stronger.

The Role of AI in Assessment Infrastructure

The resource objection to this model is real and must be answered honestly. A twenty-minute professional discussion with four hundred Level 3 Business students represents over 130 hours of assessor time for a single assessment component. Under current funding, that is simply not viable for most colleges.

AI does not solve this by replacing human judgement. It solves it by making human judgement more efficient and better prepared. AI can ingest a learner's submitted project, map evidence against threshold criteria, identify gaps and inconsistencies, flag ownership concerns, generate targeted lines of questioning for the assessor, and summarise the learner's developmental trajectory across the programme. The assessor arrives at the viva with a detailed, structured briefing rather than having read 40 pages of portfolio from scratch. One guardrail is non-negotiable, however: the AI briefing supplements but does not replace direct assessor engagement with the learner's work. An assessor who conducts a viva having read only an AI-generated summary — without independent reading of the portfolio itself — risks presiding over a closed loop in which AI audits AI, with two human beings as front-facing proxies. The assessor must bring their own cognitive ownership of the evidence into the room. The AI prepares the conversation; the human leads it.

As AI capability develops, there is a longer-term case for multi-agent assessment councils: specialist agents analysing different dimensions of a submission — domain accuracy, reasoning quality, professional judgement indicators, evidence of genuine ownership — producing a structured, auditable verdict that a human assessor can interrogate and either confirm or override. The key feature of such a model is its transparency. Unlike a black-box score, a council produces a traceable reasoning chain. That transparency is what makes AI involvement in high-stakes assessment regulatorily defensible in principle.

This model also integrates directly with existing Internal Quality Assurance (IQA) frameworks. If AI reduces the portfolio-trawling burden for both the assessor and the IQA lead — producing structured evidence maps, flagging inconsistencies, generating standardisation prompts — the business case for shifting staff hours from marking written unit submissions to conducting targeted vivas becomes practically viable rather than aspirationally attractive. The resource argument against authentic assessment weakens significantly when AI is carrying the coverage work that currently consumes that resource.

AI should carry the burden of coverage. Humans should carry the responsibility for judgement.

The Equity Imperative: Automated Containment as a Class Divide

Any leadership strategy that uses technology to decrease a learner's time with a human educator must answer one public question: which learners will receive less human contact, and what is the educational justification?

The most insidious danger of the AI transition in English further education is not a failure of technological adoption. It is the emergence of a stark, class-based pedagogical divide that launders inequality as innovation.

Advantaged learners in well-funded environments will receive a human-rich, relationally intensive education built around the five pillars of human formation. Disadvantaged learners in under-resourced FE colleges risk being subjected to automated containment dressed as personalised learning. This is not personalisation. It is stratification, delivered at scale.

The financial temptation must be named directly. Cash-starved colleges, carrying a fifteen-year legacy of disinvestment and acute staff recruitment crises in engineering, digital and construction, face a structurally seductive calculation. Software vendors are aggressively marketing AI-driven platforms that promise to deliver curriculum content, track progress and auto-mark work. The temptation for an executive team under financial pressure is clear: cut guided learning hours, reduce the need for expensive specialist lecturers, and place learners on a digital platform for a significant portion of their timetable. The resulting dashboards and engagement metrics can be presented to governors and Ofsted as evidence of a modern, digital-first strategy.

This is a profound betrayal of the sector's purpose. For a learner with interrupted prior education, complex learning needs, a chaotic home environment or fragile self-efficacy, the relationship with a skilled human educator is often the only thing that anchors them to the possibility of a different future. Replacing that relationship with an adaptive algorithm is not personalisation. It is abandonment at scale — and it produces exactly the two-tier population this paper warns against: one cohort formed by human educators to exercise judgement; another managed by machines to execute routine tasks until the machines no longer need them.

The legitimate deployment of AI in an underfunded college is the mirror image of this: automating administrative burden — compliance loops, tracking systems, regulatory paperwork — to liberate and maximise face-to-face human contact time.

Equity in the Threshold Model: Neurodiversity and Language

The human-led professional discussion is proposed as the gold standard of authentic assessment — but the paper would be intellectually dishonest if it did not name a genuine equity risk in that proposal. A high-stakes oral assessment can profoundly disadvantage specific learner groups: autistic learners, those with severe social anxiety, speakers of English as an Additional Language, and others for whom the particular performance demands of a viva create barriers unrelated to their actual capability or domain fluency.

This is not an argument against the threshold model. It is an argument for building proportionate, principled reasonable adjustments into it from the outset — not as afterthoughts bolted on under SEND compliance pressure, but as designed features of a model that takes equity seriously. The professional discussion can take adapted forms: structured written defence, supported panel formats, extended preparation time, or assessor-led conversation replacing formal viva protocol where the learner's access needs require it. What must not be adapted is the standard being assessed — destination readiness, ownership of reasoning, and professional judgement. The instrument flexes; the threshold holds. A threshold model that advantages confident oral performers over quieter, deeper thinkers has simply reproduced a different version of the sorting dysfunction it was designed to replace.

9. The AGI Horizon: Does the Argument Still Stand?

This paper has been written in a period of rapid AI transition. It would be intellectually dishonest not to ask whether its central argument survives a more radical disruption than the one it describes.

The trajectory of AI development in 2026 makes the emergence of artificial general intelligence — systems capable of matching or exceeding human performance across all cognitive domains — a live possibility within the planning horizon of current education policy. If that transition arrives, the implicit assumptions of this paper need testing.

What the AGI horizon changes. If AGI can reason, judge, plan and operate ethically at or beyond human level across all domains, then the case for education as preparation for cognitive work collapses entirely. The professional identity, domain fluency and qualification threshold arguments in this paper are all built on the assumption that human capability retains some vocational or economic value. That assumption may not hold indefinitely.

What the AGI horizon does not change. The human formation argument survives AGI almost entirely — and in some respects is strengthened by it.

Agency — in a world where AGI can perform any defined task, the scarcest human capacity is not cognitive competence. It is intentionality: knowing what you want, what matters to you, what kind of life is worth living. You cannot outsource wanting.

Relational competence — even if AGI reasons better than any human, humans still need to live with each other. Families, communities, democratic institutions, care relationships — these are not primarily cognitive problems. They require presence, accountability, love, conflict, repair and meaning.

Ethical accountability — AGI makes decisions. Humans remain the moral agents who deploy it, constrain it, and are accountable for what it does in the world. You cannot outsource moral responsibility to a system regardless of its capability level.

Civilisational literacy — a population that cannot understand the systems they inhabit is vulnerable in any era. In an AGI era, that vulnerability is existential. Democratic agency requires an intelligible world.

The purpose that survives. Strip away every instrumental justification for education — workforce preparation, qualification currency, cognitive competitive advantage — and what remains is the oldest justification, the one that was always true and that industrial-era schooling buried under two centuries of economic utility.

The purpose of education is to form humans who can live with meaning, agency and ethical accountability in a world they did not design and cannot fully control.

AGI does not destroy that purpose. It strips away everything else and leaves it standing alone. If the most optimistic AI trajectory arrives — and it may — the human formation argument does not become obsolete. It becomes the only argument that was never contingent on human cognitive supremacy in the first place.

10. Risks and Counterarguments

Objection 1: This is too philosophical for a sector managing severe underfunding. Underfunding is precisely why clarity of purpose matters most. An underfunded system without a clear account of what it is for will use AI for cost-cutting, because that is the path of least resistance. Philosophy is not the opposite of pragmatism. It is the condition for intelligent pragmatism.

Objection 2: Employers still need specific skills and qualifications, not abstract formation. They do — and AI is rapidly changing which skills carry durable value. Employers who currently need bookkeepers will soon need people who can interpret financial risk and advise clients while AI handles the routine reconciliation. The specific task portfolio is becoming less stable; the judgement, reasoning and professional identity above the task is becoming more valuable and harder to find. The formation argument is the employer argument, properly understood.

Objection 3: Assessment must remain controlled and standardised to protect qualification integrity. Controlled assessment for regulated qualifications is necessary and will remain so. The argument is not to abandon rigour but to change its location. A threshold assessment combining a learner-derived project with a human-led professional discussion tests a wider and more relevant range of capability and is significantly harder to game.

Objection 4: Human formation is unmeasurable and therefore unaccountable. Safeguarding culture, curriculum quality, professional identity and the conditions for learning are difficult to quantify — yet inspection already assesses them through qualitative judgement, learner voice, observation and self-assessment. Human formation is not more intangible than "quality of education."

Objection 5: The threshold model would disadvantage learners who struggle with high-stakes assessment. The current model disadvantages them more profoundly — by processing them through qualifications they have not genuinely earned, producing the appearance of capability and then exposing them to failure in the environments that follow. A threshold model with proportionate formative scaffolding, structured peer defence as rehearsal, and AI-supported preparation is more likely to identify and address genuine gaps before the high-stakes moment — not less.

11. Five Policy Commitments for an AI-Age FE System

The argument of this paper has direct implications for national policy. Individual institutional ambition is insufficient if the funding, qualification, inspection and accountability frameworks reward different behaviours.

For the DfE and HM Treasury: this is not a proposal for expensive soft-skills provision. It is a national productivity insurance policy. The state currently risks funding young people to acquire task competences that will be fully automated before their student loans are written off. For the AoC and college leaders: this paper is also a sector protection instrument. Adopted collectively, the human formation framework gives the sector a unified vocabulary to resist vendor-driven dashboard education and a defensible position when inspection frameworks are contested.

Commitment 1: Protect human contact hours in disadvantaged provision — and change the funding methodology that makes this impossible. The funding model itself must change. A national AI deployment standard for FE should distinguish between AI uses that recover human contact time (legitimate) and AI uses that replace it (unacceptable). Ofsted should be required to evaluate AI deployment against this standard as part of Quality of Education judgements.

Commitment 2: Fund staff AI capability and educator development nationally. The priority for AI investment in FE is not platform procurement. It is educator development — building the professional capacity of teachers and lecturers to use AI critically, deploy it appropriately, resist its misuse and develop learners' own judgement about it.

Commitment 3: Require awarding bodies to design threshold-based qualification models with authenticated summative assessment. Assessment should be organised around destination readiness — university, employment, apprenticeship — not internal unit completion. Final achievement should centre on a learner-derived project authenticated through human-led professional discussion or viva.

Commitment 4: Reframe English and mathematics resit policy around reasoning, communication and agency. Maintain rigour. Change the method. Develop assessment routes that reward genuine mathematical reasoning, written communication and agency — rather than exam technique optimised for a paper that large language models can now outperform.

Commitment 5: Update inspection and accountability frameworks to recognise and value formation outcomes. Human formation outcomes should sit explicitly within the Quality of Education and Personal Development judgement areas of the renewed Ofsted inspection framework.

12. What This Means in Practice: A Framework for Institutions

What a Level 2 or Level 3 Learner Does Differently

The test of any educational argument is not what the institution believes or the curriculum intends. It is what happens in the learner's actual experience. In a programme shaped by the human formation argument, a Level 2 or Level 3 learner:

• Is regularly asked to explain their reasoning, not just demonstrate their output
• Completes some tasks without AI access, to build internal knowledge and domain fluency that belongs to them
• Uses AI for other tasks but is assessed on the quality of their direction, evaluation and judgement — not merely the output produced
• Is asked to identify where AI assistance is unhelpful, wrong or inadequate within their vocational domain
• Encounters problem-framed or employer-set tasks where defining the question is part of the challenge
• Receives feedback on professional behaviour, relational skills and ethical reasoning — not only technical accuracy
• Reflects formally on uncertainty, confidence and the limits of their own knowledge
• Practises professional discussion: defending a decision, explaining a judgement, acknowledging what they do not yet know
• Knows what their qualification threshold means — and can articulate whether they are ready to cross it

None of these require additional resource in isolation. What the human formation argument requires is that they are designed deliberately, sequenced progressively and assessed honestly.

Three Horizons for Institutional Change

Horizon 1: This Term. Rename what you are already doing — most good FE already develops human formation; it simply does not call it that and therefore cannot protect it when pressure mounts. Audit one programme for agency: take one scheme of work and ask honestly how much of it requires learners to choose, initiate, justify or frame a problem. Replace AI prohibition with AI policy — blanket bans are pedagogically dishonest and practically unenforceable; replace with a tiered policy distinguishing tasks done without AI, tasks done with AI assessed on direction and judgement, and tasks requiring learners to identify where AI fails. Make your reasoning visible: stop presenting conclusions; narrate the thinking. Learners who regularly observe expert judgement develop judgement. Learners who observe only expert conclusions develop compliance.

Horizon 2: This Academic Year. Map the human formation strand — the pastoral curriculum, enrichment programme, tutorials and employer engagement are usually doing more of this work than anyone has formally acknowledged. Build a staff AI capability framework — with practitioners, not for them. Redesign one assessment per curriculum area for judgement, starting in the formative space, where awarding body regulations do not constrain lecturer creativity. Engage employers around the new purposes — employers in every sector are confronting the same question: what do workers need when AI handles the routine?

Horizon 3: Three to Five Years. Reframe self-assessment around formation outcomes: are learners developing agency, building professional identity, and are the most disadvantaged developing these capacities at a rate that closes rather than widens the gap? Protect human contact as a strategic asset — AI deployed where it genuinely improves practice; human contact preserved where it cannot be replicated. Invest in educator development as the primary capital expenditure: the professional intelligence of teachers and lecturers is the most valuable and least replaceable institutional asset.

13. Conclusion: Human Literacy in an Age of Non-Human Intelligence

The genuine purpose of education in a world where cognitive power is abundant is not to preserve the illusion that humans will always be the most analytically capable entities in the room. Nor is it merely custodial, credentialling or workforce preparation — though it contains all three and must do them better.

The deeper purpose of education is to form humans who can live wisely with power.

AI gives humanity cognitive power at a scale never previously available. That makes education more important, not less. But it changes what education is fundamentally for. The outcome of the educated person is not the learner who can recall what AI knows, or who can occasionally catch AI in error. It is the person who can ask meaningful questions, form worthy purposes, understand enough of the world to act responsibly, relate well to other human beings, use intelligent tools without becoming dependent on them, and make judgements that are technically informed, ethically grounded and psychologically mature.

The future of education is not AI literacy. It is human literacy in an age of non-human intelligence.

And if the AGI horizon arrives sooner than current policy imagines — if the transition goes further and faster than the cautious centre of educational debate is willing to contemplate — then the human formation argument does not weaken. It becomes the only argument that was never contingent on human cognitive advantage in the first place. Every other justification for education is conditional on a world where human cognitive performance retains economic value. The human formation argument is not. It rests on the permanent fact that humans must live together, must make choices, must be accountable for what they do with power, and must find ways of belonging to something worth belonging to.

That was always the deepest purpose of education. Two centuries of industrial workforce preparation buried it. AI has uncovered it.

Further education is where that recovery must begin — because FE is where the disruption is already arriving, and because the communities FE serves have the most to lose if we get it wrong, and the most to gain if we get it right.

Some of this work can begin on Monday, in any institution. But doing this seriously, equitably and at scale will require a funding and regulatory settlement that recognises human contact, educator development and curriculum design as strategic national infrastructure — not optional extras to be renegotiated downwards when the budget tightens.

That is the argument. The paper invites challenge, refinement and action.

There is, however, a harder truth that purposeful action cannot entirely soften. The reform mechanisms available to England's post-16 system — qualification redesign, funding methodology review, inspection framework revision, awarding body consultation — operate on timescales measured in years. Parliamentary cycles, regulatory consultation periods, sector-wide pilots and phased implementation: these are the instruments of change, and they move at the pace institutions and governments find manageable. AI does not move at that pace. The capability gap between what the system assumes and what AI can now do is not waiting for the next qualification reform cycle. It is widening in every classroom, on every assessment submission, in every college that has not yet decided what it stands for.

The cohort bearing the cost of that gap is not a future cohort. It is the learners sitting in English FE colleges right now — assessed against frameworks designed for a world that no longer exists, in institutions whose funding settlement makes the right response unaffordable, inside a regulatory architecture that has not yet named what is happening clearly enough to stop it. A generation does not get a second chance at its formation years. The window in which this paper's argument matters most is not the window of policy debate. It is the window of now. What opens if we waste it is not a temporary setback. It is a gap in human capability and human dignity that no subsequent reform will fully close.

The authors welcome responses from college leaders, curriculum directors, awarding bodies and policymakers. The arguments in this paper are intended as a provocation for sector-wide debate, not a final position.

Stephen Kittner, Heart of Worcestershire College, 2026.

This paper was developed through iterative critical dialogue with AI tools. The thesis, arguments, structural decisions, critiques and all editorial judgements are the author's own — the product of repeated cycles of challenge, rejection and refinement. The process is itself an illustration of the paper's central argument: that the scarce human capacity in an age of abundant AI is not the ability to generate output, but the judgement to know when it is right.

Notes and References

1. Maslej, N., Toups, L., Perrault, R., Shoham, Y., Clark, P., Bordelon, M., Bengio, Y., Whittlestone, J., Manyika, J., Etchemendy, J. and Leskovec, J. (2026) The AI Index 2026 Annual Report. Stanford, CA: Stanford Institute for Human-Centered Artificial Intelligence.
2. World Economic Forum (2025) The Future of Jobs Report 2025. Geneva: World Economic Forum. ISBN: 978-2-940631-90-2.
3. World Bank (2018) World Development Report 2018: Learning to Realize Education's Promise. Washington, DC: World Bank. ISBN: 978-1-4648-1096-1.
4. Hanushek, E.A. and Woessmann, L. (2015) The Knowledge Capital of Nations: Education and the Economics of Growth. Cambridge, MA: MIT Press. ISBN: 978-0-262-02917-9.
5. OECD (2023) PISA 2022 Results (Volume I): The State of Learning and Equity in Education. Paris: OECD Publishing. ISBN: 978-92-64-97742-9.
6. Drayton, E., Farquharson, C., Johnes, R., Kettlewell, K., Sartorious, F. and Sibieta, L. (2020) Annual Report on Education Spending in England: 2020. London: Institute for Fiscal Studies.
7. Institute for Fiscal Studies (2025) Cutting Overall Education Spending Would Be Difficult Without Cutting School and College Budgets. London: IFS. The briefing notes that college teachers earn on average £7,000 (approximately 18%) less per year than their school counterparts.
8. Ofsted (2025) Education Inspection Framework. Effective 10 November 2025. Manchester: Ofsted.
9. Joint Council for Qualifications (2025) AI Use in Assessments: Your Role in Protecting the Integrity of Qualifications. Revision 2, 30 April 2025. London: JCQ.

Key Terms

Human formation

The deliberate integration of knowledge, skill, identity, ethics and agency so that learners can act responsibly in real contexts. Its distinctive feature is not that it adds more content to the curriculum, but that it changes the organising question from "Can the learner complete the task?" to "Is the learner becoming capable of responsible judgement in a world where tasks can increasingly be automated?"

Internal fluency

The domain-specific knowledge and procedural competence that must be developed within the learner, not outsourced to AI. The prerequisite for external judgement: a learner who cannot perform a foundational calculation cannot develop the professional intuition to know when the AI's answer is wrong.

Judgement

The capacity to reason across competing values, incomplete evidence, uncertainty and consequences — and to take accountability for the decisions that result.

Automated containment

The use of AI-mediated platforms to substitute for human educator contact in ways that reduce rather than enhance the conditions for human formation, particularly for disadvantaged learners.

Threshold model

An assessment framework that evaluates learners against their destination readiness — university entry, employment, apprenticeship progression — rather than against internal unit completion criteria.

Destination readiness

The demonstrated capability to cross a specific post-qualification threshold: university-level study, professional employment, or structured apprenticeship development. The organising standard of the threshold assessment model.