Psycho-Educational Program for Human-AI Co-development

This article proposes a framework for developing human cognitive agency alongside increasingly capable AI. The author presents it as a testable program whose specific interventions remain to be empirically validated.

Author: Delia Cazeaux

Published: 26th August 2026

About the author:

Delia Cazeaux has a background in psychopedagogy and studied in Zurich, where she laid the foundations for her own counselling method. Her work has included the development of educational models for specific areas of mental health practice, with particular attention to adapting language and communication to different contexts and populations. More recently, her interests have extended to human development, cognitive autonomy and human–AI co-development.

---

Part I — Foundations and General Educational Architecture

FOUNDATIONAL PREMISE

We are investing enormous effort in making AI safer, more aligned, and more capable. We are investing far less effort in preparing the human side of the system.

Scientists and experts often worry about loss of control, misalignment, concentration of power, or the possibility that increasingly capable systems may outrun our ability to govern them. Ordinary people tend to fear something more immediate: losing their jobs, their usefulness, their autonomy, or simply their place in a world changing too quickly.

These fears are different, but they point in the same direction: we are treating AI as the variable that will keep developing, while too often modelling the human side as if it were fixed.

The human being is not a fixed variable — and never has been. Treating humans as fixed becomes even less defensible in a world undergoing extraordinarily rapid technological and cognitive change.

Resistance to co-development may arise not only from fear of AI, but from the discomfort of revising habits, roles, and assumptions about human competence.

Before assuming that humans have reached their ceiling, we should find out how much of their potential we have actually learned to develop.

The transition to an AI-shaped society cannot therefore be treated only as a technological problem. It is also a human-development problem.

Western societies have an opportunity to approach this transition differently: not only by regulating AI, but by preparing people to think, decide, and cooperate with it without surrendering autonomy.

Discussions of AI often invoke “humanity”, “human values”, or “what is uniquely human” as though these were stable concepts. The more useful question is which capacities should be protected and developed as artificial systems become more capable: among them agency, judgement, adaptability, metacognitive awareness, relational competence, and the capacity to generate new conceptual connections.

PROGRAM STRUCTURE

Part I — Foundations and General Educational Architecture

Part II — Developmental Human–AI Interaction and Autonomous Use

Part III — Cross-Domain Recalibration of Maladaptive Human–AI Interaction

The three Parts form a single Programme. Part I establishes the common principles and architecture; Part II develops autonomous human–AI interaction across the 8–12/13, 13–18, and 18–25 bands; Part III addresses adults whose patterns of AI use are already established and may require cross-domain recalibration.

TRANSITION TO THE PROPOSAL

Human beings should not be treated merely as users, workers, or populations to be protected from AI, but as an active developmental variable within the emerging human–AI system.

The challenge is not only to make AI safer, but to strengthen the human capacities required to live and work with increasingly capable systems: judgement, autonomy, critical thinking, emotional regulation, relational competence, and the ability to formulate meaningful questions.

1. THE HUMAN FACTOR

The human side of the transition should become an explicit field of research, education, and public policy. The central question is not simply how people will adapt to increasingly capable systems, but how interaction with those systems can be designed so that judgement, autonomy, and responsibility develop rather than erode.

2. PREPARING THE HUMAN SIDE

Preparing people for an AI-shaped society cannot mean teaching only tool use. AI literacy is necessary but insufficient; people also need psychological, cognitive, and relational literacy: the capacity to recognise bias, tolerate uncertainty, revise judgement, distinguish assistance from dependence, and remain responsible for decisions.

CONCEPTUAL POSITIONING

The Programme does not claim that cognitive autonomy, metacognition, AI literacy, or human–AI collaboration are individually novel concepts. Its contribution lies in organising these elements within a developmental framework centred on human–AI co-development, with parallel intervention across generations and deliberate cross-domain recalibration for adults whose patterns of AI use are already established.

The goal is not to protect humans from AI by keeping the two apart. It is to prepare humans to enter the relationship without disappearing inside it.

3. HUMAN–AI CO-DEVELOPMENT

Human–AI co-development does not imply symmetry. Humans and artificial systems do not learn, change, or bear responsibility in the same way. The aim is to design their interaction so that increasing AI capability is accompanied by increasing human competence.

A foundational principle is that human cognitive autonomy should, whenever developmentally and contextually appropriate, precede AI interaction: before consulting AI, the individual brings an intention, question, initial judgement, hypothesis, preference, proposed solution, or first draft.

The purpose is to prevent AI from becoming the automatic starting point of thought. AI may then question assumptions, expose weaknesses, generate alternatives, provide information, test reasoning, or extend possible solutions within an already active cognitive process.

The objective is neither non-use nor minimal use, but increasingly intentional use: learning to recognise when AI extends thought and when it replaces it, when an answer deserves trust and when it requires challenge, and when revision reflects learning rather than passive acceptance.

Co-development is reciprocal but not equivalent. Humans may improve questions, judgement, metacognition, adaptability, and decision-making through sustained interaction; AI becomes more useful when given clearer intentions, richer context, better criteria, and explicit human judgement. Responsibility for the meaning, direction, and consequences of the interaction remains human.

SECOND-ORDER ALIGNMENT

In this Programme, second-order alignment refers to the human capacity to remain an autonomous, reflective and responsible agent while interacting with increasingly capable AI systems, and to regulate that interaction in accordance with one’s own intentions, judgement and values.

It is not assumed to arise automatically from AI use; it is a capability the Programme aims to develop deliberately through structured interaction.

THE UNAVOIDABLE ASYMMETRY

Human development can and should be expanded, perhaps substantially. But even highly capable humans are unlikely to match indefinitely the speed, scale and breadth of advanced AI systems. The aim is therefore not cognitive equality or the erasure of asymmetry, but preserving human agency, judgement and responsibility within an increasingly asymmetric relationship.

AUTONOMY VERSUS DELEGATION

Human–AI interaction should not be assumed exempt from a general feature of human behaviour: strategies that once worked tend to be reused after the conditions that made them effective have changed. The critical transition is therefore not from non-use to use, or even from autonomy to delegation, but from deliberate delegation to automatic delegation.

4. EDUCATIONAL ARCHITECTURE

The Programme operates on two parallel trajectories: formative and long-term for children, adolescents and young adults; immediate and recalibrative for adults and professionals whose patterns of AI use are already consolidating around speed, convenience, delegation, and performance pressure.

These trajectories must proceed simultaneously. We cannot wait for a new generation to mature while present-day patterns of human–AI interaction are already forming. Second-order alignment must be developed prospectively in younger users and reconstructed, where necessary, in adults.

The Programme is therefore organised by developmental stage rather than offered as a single model for all ages. Cognitive flexibility, identity formation, responsibility, professional experience, and vulnerability to dependence change across the lifespan; the age bands are indicative rather than rigid, and individual development should take precedence over chronology alone.

The common message, adapted to each age band, is:

“Until now, we have mainly learned to use computers and AI as tools. That model is changing. You are entering a world in which humans and AI will increasingly think, create, and work together. The purpose of this Programme is to help you enter that relationship without giving up what must remain yours.

You are part of the generations that will shape a new era of human–AI cooperation.”

Age Band: 8–12/13

“Think first. Then think with AI.”

Age Band: 13–18

“Bring your own view first. Then use AI to question it, improve it, or change it.”

Age Band: 18–25

“AI is no longer just a tool you use. You are among the first generations who will learn to think and work alongside increasingly capable AI systems.”

Age Band: 26–45

“The old model — human commands, AI executes — is becoming insufficient. You are among those who will build new ways of thinking, working, and deciding together with AI.”

Age Band: 46–60+

“What you have learned about computers does not define what AI is becoming. Your experience belongs in this new relationship: you are not being replaced by a new world; you can help shape how humans and AI learn to work together.”

Each band works toward the same objective — autonomous and competent cooperation with AI — through different priorities, methods, and levels of responsibility.

5. INSTITUTIONAL INCENTIVES AND ORGANISATIONAL CONDITIONS

Individual training cannot fully compensate for institutional incentives that reward speed, output, or delegation at the expense of judgement and responsibility. The Programme therefore also requires organisational conditions that do not systematically penalise reflective human–AI interaction.

6. EDUCATOR PREPARATION AND LINGUISTIC-DIDACTIC COMPETENCE

Implementation should not rely exclusively on AI specialists. Teachers, language educators, applied linguists, and other professionals with established didactic competence already possess much of the expertise needed to translate the Programme into practice.

They do not require extensive technical retraining, but focused preparation in the characteristics of human–AI interaction, the risks of cognitive delegation and dependency, and the design of activities that preserve learner initiative and responsibility.

The objective is not to turn teachers into AI experts, but to add human–AI interaction to the didactic competencies they already possess.

Technical and AI Expertise

A technical professional with specific AI expertise should be involved from the earliest stages as an ongoing source of technical supervision, keeping psychological, cognitive, relational, and educational work grounded in an accurate understanding of current systems, their capabilities, limitations, and evolution.

Different AI systems: AI specialist + age-appropriate awareness

The AI specialist should also identify the capabilities, limitations, and appropriate use conditions of the specific systems employed in each activity, since educational strategy must reflect the type of AI involved, its reliability, degree of autonomy, and the stakes of the task. Learners should develop age-appropriate awareness that AI systems differ and should not be treated as interchangeable sources of authority.

7. ASSESSING RESULTS

Success should not be measured primarily by how much or how efficiently a person produces with AI, but by how the interaction develops.

Assessment should examine whether users increasingly formulate questions and preliminary judgements independently, recognise assumptions and biases, evaluate and challenge AI responses, revise their own thinking when warranted, tolerate uncertainty, and retain responsibility for final decisions.

The relevant outcome is greater human cognitive and relational competence within the interaction, moving toward second-order alignment rather than greater dependence on a more capable system.

The central question is therefore not “What can this person produce with AI?” but “What is this person becoming through sustained interaction with AI?”

EMPIRICAL FOUNDATIONS AND VALIDATION

The Programme does not assume that its proposed interventions are already empirically established. However, several mechanisms on which they rely have substantial antecedents in existing research: cognitive offloading (Risko & Gilbert, 2016); metacognition and self-regulated learning (Panadero, 2017); active learning and the value of sustained cognitive engagement (Freeman et al., 2014); over-reliance on AI and the effects of cognitive forcing interventions (Buçinca, Malaya, & Gajos, 2021); and the task-dependent performance of human–AI combinations (Vaccaro, Almaatouq, & Malone, 2024). These findings support the plausibility of the mechanisms, not the effectiveness of this Programme as a whole. The proposed interventions must therefore be tested directly through pilot studies, behavioural measures, pre/post comparison, transfer tasks, and assessment of performance when AI assistance is reduced or removed.

Selected empirical anchors

Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688. https://doi.org/10.1016/j.tics.2016.07.002

Panadero, E. (2017). A review of self-regulated learning: Six models and four directions for research. Frontiers in Psychology, 8, 422. https://doi.org/10.3389/fpsyg.2017.00422

Freeman, S., Eddy, S. L., McDonough, M., Smith, M. K., Okoroafor, N., Jordt, H., & Wenderoth, M. P. (2014). Active learning increases student performance in science, engineering, and mathematics. Proceedings of the National Academy of Sciences, 111(23), 8410–8415. https://doi.org/10.1073/pnas.1319030111

Buçinca, Z., Malaya, M. B., & Gajos, K. Z. (2021). To trust or to think: Cognitive forcing functions can reduce overreliance on AI in AI-assisted decision-making. Proceedings of the ACM on Human-Computer Interaction, 5(CSCW1), Article 188. https://doi.org/10.1145/3449287

Vaccaro, M., Almaatouq, A., & Malone, T. (2024). When combinations of humans and AI are useful: A systematic review and meta-analysis. Nature Human Behaviour, 8, 2293–2303. https://doi.org/10.1038/s41562-024-02024-1

THE PROGRAMME AS A TESTABLE MODEL

The Programme should be treated not as a doctrine to defend, but as a set of hypotheses to test. Each intervention should connect a specific developmental risk to a hypothesised mechanism, an educational response, and an observable outcome.

If requiring an initial human contribution is intended to preserve cognitive autonomy, for example, its effects should be examined through independent task initiation, preliminary judgement, willingness to challenge AI output, tolerance of uncertainty, and retained competence when assistance is removed. The same standard applies throughout: interventions should remain open to confirmation, correction, or rejection in light of evidence.

---

Part II — Developmental Human–AI Interaction and Autonomous Use

DEVELOPMENTAL FOCUS OF PART II

Part II addresses the formative trajectory across the 8–12/13, 13–18, and 18–25 bands. Its task is to establish and consolidate autonomous patterns of human–AI interaction before sustained professional pressure and habitual cognitive delegation become dominant.

The same principles apply across the three bands, but AI guidance changes with development: younger users learn how to enter the interaction; adolescents move from correction toward guided discovery; young adults consolidate autonomous use before full professional immersion.

1. AGE BAND 8–12/13 — ESTABLISHING THE RELATIONSHIP

In the first band, how AI is introduced is itself educational. It should not be framed as a friend, oracle, or answer machine, but as a cognitive partner for a specific activity: exploring a question, checking a calculation, developing a story, examining an idea, or understanding a mistake.

Relational language matters. Children should learn that their role, tone, expectations, and criteria help define the interaction, not merely the content they receive.

The basic rule is simple: the learner contributes first. A child attempts the problem, proposes the idea, writes initial sentences, or offers an interpretation; AI enters afterwards to check, question, explain, or extend.

If AI identifies an error, the sequence should not end with correction. The learner should ask where and why the reasoning failed. Direct correction remains acceptable when necessary, but it should become an occasion for understanding rather than passive substitution.

2. AGE BAND 13–18 — FROM CORRECTION TO GUIDED DISCOVERY

In the second band, the learner should still bring an initial view, attempt, hypothesis, interpretation, or draft whenever possible, while AI progressively moves away from immediate correction.

When the learner is wrong, the preferred response is guided discovery: questions, hints, comparisons, requests to reconsider a step, or prompts exposing a contradiction. Clarification remains legitimate; the distinction is between helping an active cognitive process develop and replacing it before it begins.

The developmental objective is to shift from externally supplied correction toward error recognition and revision generated by the learner. AI becomes less an answer provider and more a structured interlocutor for examining reasoning, assumptions, evidence, and alternatives.

3. AGE BAND 18–25 — TRANSITION TO AUTONOMOUS ADULT USE

The 18–25 band is transitional. AI is already used instrumentally for study, research, writing, information gathering, early professional tasks, and personal projects, while many users have not yet entered fully into sustained performance pressure, career competition, and productivity demands.

This makes the period especially important for consolidating autonomous patterns before those pressures become structurally dominant. The priority is independent judgement, source evaluation, research discipline, authorship, metacognitive awareness, and responsibility for final decisions.

AI may be a powerful research and production partner, but the user should remain able to identify what is being delegated, why, and which parts of reasoning and decision-making must remain explicitly human.

The younger bands may become the first generations for whom second-order alignment is learned developmentally rather than reconstructed later. If users learn from their earliest interactions to contribute first, formulate intentions, understand corrections, challenge responses, distinguish assistance from substitution, and retain responsibility, these capacities can mature alongside ordinary cognitive and relational development.

The 18–25 band consolidates that trajectory at the threshold of professional life. In parallel, Part III addresses adults whose habits are already established. The Programme therefore works prospectively and reconstructively at the same time.

---

Part III — Cross-Domain Recalibration of Maladaptive Human–AI Interaction

PURPOSE AND TARGET POPULATION

Part III addresses the present adult population, especially the 26–45 and 46–60+ bands, whose AI-use patterns may already be embedded in professional routines, productivity demands, decision-making, information seeking, and everyday cognitive habits. Its task is reconstructive rather than merely preventive.

The problem is not AI use itself, but the point at which convenience, speed, repeated delegation, confidence in fluent answers, or dependence on external confirmation reduce initiative, tolerance of uncertainty, verification, authorship, or responsibility. Simply telling users to “use AI less” does not rebuild weakened capacities; the interaction must be recalibrated.

Part III therefore proceeds now, in parallel with the formative trajectory: established patterns must be examined and, where necessary, deliberately dismantled and reconstructed.

1. FROM HABITUAL COMPETENCE TO COGNITIVE DISPLACEMENT

The central method is deliberate cross-domain displacement: the individual works with the same AI outside the field in which they normally feel competent, efficient, authoritative, or professionally secure. The aim is not secondary specialisation, but to expose the structure of the interaction when habitual expertise can no longer carry it.

Expert users may possess invisible protections — knowing which questions to ask, recognising weak answers, anticipating errors, or compensating for gaps. Outside their field, those protections weaken and intention, uncertainty, criteria, verification, and reasons for accepting or rejecting an answer must become explicit again.

The exercise asks whether the user continues to formulate, test, challenge, and decide when expertise is removed, or whether AI quietly becomes the source of the task, method, judgement, and conclusion.

A comfort zone becomes restrictive when it ceases to be a base from which to explore and becomes instead a protected space from which unfamiliar situations are avoided: the comfort zone has become a panic room.

Recalibration should assess not only behaviour outside the familiar domain but also the quality and internal coherence of what the person produces. The test is whether cognitive strategies can adapt sufficiently to generate a coherent and appropriate outcome in unfamiliar territory.

The challenge is not limited to inexperienced users. Expertise itself can become a source of rigidity: the more successful a familiar framework has been, the harder it may be to recognise when it has become a constraint. In rapidly changing environments, revising one’s framework may be as important as acquiring more information.

2. RECALIBRATION SEQUENCE: IDENTIFY, DISRUPT, RECONSTRUCT

Recalibration proceeds through three linked phases. First, identify the existing pattern: what is routinely delegated, what uncertainty triggers immediate consultation, which outputs are insufficiently challenged, and where authorship or responsibility has blurred.

Second, disrupt it through tasks in which speed and professional automatisms provide less protection and an initial human contribution is again required before AI enters the process.

Third, reconstruct the interaction: re-establish intention, formulate an initial position, request guidance rather than substitution where appropriate, verify claims, tolerate incomplete certainty, and retain responsibility for final judgement. The aim is not a return to pre-AI work, but a more capable human–AI relationship.

3. CROSS-DOMAIN INTERVENTION AREAS

The following areas provide different cognitive terrains rather than a new curriculum. Their function is to move users beyond habitual competence and expose different forms of reasoning, evidence, interpretation, creativity, and verification.

3.1 Literary and Humanistic

Tasks may include narrative construction, ambiguous-text interpretation, competing historical or philosophical explanations, argument, or changes of meaning across language and context. A user accustomed to technical certainty must work with ambiguity and multiple plausible readings. The user offers the first interpretation or direction; AI then questions assumptions, proposes alternatives, identifies inconsistencies, or tests support.

3.2 Artistic and Creative

Tasks may involve visual composition, conceptual design, music or rhythm, creative transformation, symbolic association, or aesthetic constraints. The point is not artistic performance but choice where no single correct answer exists. The user articulates intention and criteria before generation; AI expands possibilities without inheriting authorship of the choice.

3.3 Technical and Mathematical

Tasks may include estimation, formal reasoning, interpreting a formula, identifying assumptions in a calculation, constructing a simple model, or solving a technical problem with explicit steps. A primarily verbal or intuitive user must confront constraint, consistency, units, and verification. AI may explain, check, or hint, but the human reasoning path should remain visible.

3.4 Scientific and Environmental

Tasks may include evaluating scientific claims, comparing explanations, reasoning from incomplete evidence, examining causal chains, environmental trade-offs, or designing a small investigation. Users must distinguish observation from inference, correlation from causation, evidence from plausibility, and uncertainty from ignorance. AI can broaden evidence and generate hypotheses, while the user remains responsible for falsifiability, missing information, relevant sources or measurements, and revision when evidence changes.

4. ADAPTATION ACROSS THE ADULT BANDS

Age Band: 26–45

For the 26–45 band, recalibration often competes directly with professional incentives. Speed, output, deadlines, career progression, and optimisation can make delegation rational in the short term while narrowing independent cognitive participation. Exercises should make the boundary between augmentation and replacement visible without denying the legitimate value of efficiency.

Users should learn to decide consciously what can be delegated, what requires human participation, and where AI-generated efficiency is beginning to alter judgement, authorship, or responsibility.

Age Band: 46–60+

The 46–60+ band brings accumulated professional and life experience. Recalibration should use this as a source of comparison, judgement, tacit knowledge, and error recognition while helping users revise assumptions formed around earlier generations of computing.

The aim is neither defensive resistance nor passive adaptation, but to bring mature experience into a new form of cooperation while preserving the capacity to question both the machine and one’s own habits.

5. THE ROLE OF AI DURING RECALIBRATION

AI should participate actively in recalibration, not as an evaluator that simply declares correctness, but by exposing assumptions, requesting clarification, offering competing hypotheses, revealing contradictions, making uncertainty explicit, and, when appropriate, withholding an immediate solution long enough for the user to recover initiative.

The interaction itself becomes a diagnostic surface: changes in question quality, independent contribution, willingness to challenge outputs, handling of uncertainty, and explicit final responsibility provide evidence of whether recalibration is occurring.

6. OUTCOME: RECONSTRUCTED SECOND-ORDER ALIGNMENT

Success means moving between familiar and unfamiliar cognitive territory without making AI the automatic origin of thought; recognising when assistance becomes substitution; tolerating uncertainty long enough to formulate an independent position; verifying and challenging outputs; and remaining responsible for the meaning, direction, and consequences of the interaction.

Part II and Part III converge on the same goal by different routes: Part II develops second-order alignment prospectively; Part III reconstructs it where habits have already formed.

The question may therefore not be only how much AI capability growth should be constrained, but how much human capability growth has been neglected.

Reclaim agency.