Education should not only adapt to AI.
It should help shape the transition.
Human authorship
Keep intent, evaluation and final responsibility visible throughout AI-supported work.
Hybrid competence
Develop the capacity to frame, compare, validate, refine and responsibly commit.
Societal stewardship
Position education as an active shaper of the AI transition, not merely a recipient of it.
Stay connected
Want to know more about our efforts and collaboration opportunities?
Receive updates about Hybrid Intelligent (HI) Education research, practical resources, workshops and opportunities to collaborate.
The HI Manifesto for Education
Suggested citation: Sherson, J., Emruli, B., Hansen, R. L., Bang, J. B., Jung, J., Larsen, C. B., Gjaldbaek, M. S., Kristensen, S. B., Hansen, N. B., & Darwich, O. (2026). The HI Manifesto for Education: Institutionalizing societal stewardship in the age of AI. Working paper, Center for Hybrid Intelligence, Aarhus University. Retrieved from https://hi-infinity.ai/research/the-hi-manifesto-for-education/
Educational challenge
Better output is not the same as better learning
Generative AI can help students produce text, solve problems and receive feedback at unprecedented speed. Yet the same systems can also make them appear more capable while removing the struggle through which capability is formed.
Hybrid Intelligence in Education offers a different approach. Rather than treating AI as either a threat to learning or a shortcut to efficiency, it asks how human intelligence and artificial intelligence can be deliberately combined so that learning, judgment and human agency are strengthened.
The goal is not simply to teach students how to use AI tools. It is to build up learners so they can frame problems, explore alternatives, evaluate machine-generated outputs, make responsible decisions and redesign the systems and workflows in which AI is used.
Education is not only where society responds to technological change. It is one of the institutions through which society can actively shape it.
The educational manifesto is the capability-building pillar of the wider Hybrid Intelligence research program.
Protecting the expertise pipeline
AI can improve short-term performance while weakening the pathways through which professional expertise is developed.
When entry-level tasks are increasingly delegated to AI, learners and early-career professionals may lose opportunities to build the contextual and tacit knowledge required for later judgment. Organizations can become more productive in the present while becoming less capable of reproducing the human expertise they will need in the future.
Education therefore carries a wider responsibility than merely preparing individuals to use new tools. It must help society continuously reproduce the expertise required to evaluate, govern and improve AI-supported work.
THE CENTRAL RISK
A society may become increasingly capable of producing outputs while becoming less capable of developing the people who can understand, challenge and take responsibility for them.
From AI literacy to hybrid competence
Learning to work at the prediction-judgment frontier
AI literacy remains important: Learners must understand how AI works, critically evaluate its outputs and consider its ethical and societal implications. But literacy alone does not determine how responsibility should be distributed inside an AI-supported task.
AI is especially effective when work can be performed through abundant, explicit and formalized information. Human judgment becomes essential when a task depends on context, tacit knowledge, values, interpretation or accountable ownership.
The boundary is dynamic. As contextual knowledge becomes externalized and machine-readable, parts of yesterday’s judgment become today’s prediction. Education therefore cannot rely on a fixed catalog of uniquely human skills. It must cultivate the capacity to identify where AI prediction is useful, where human judgment is required and where responsibility must remain human.
Prediction points
Moments where AI can generate, classify, simulate, search or accelerate a clearly bounded part of the work.
Judgment points
Moments where learners must frame, interpret, validate, choose, escalate or take responsibility for consequences.
How this relates to existing Ai and education frameworks
The Five Big Ideas presented byAI4K12, the DigComp lineage and UNESCO’s competency frameworks has established that learners should understand how AI systems work, evaluate their outputs and weigh their societal implications.
The joint OECD / European Union AILit framework, goes further. Its Manage AI domain asks learners to decide where AI should automate or augment and to set checkpoints between human and AI roles; its Shape AI domain casts them as improvers of AI systems rather than consumers of them.
Hybrid Intelligence education departs from the existing frameworks in two respects. Where AILit anchors the human contribution in a catalog of capacities: Emotion, ethical reasoning, critical thinking, originality, we anchor it in the knowledge structure of the task, which does not age each time a capability get progressively better. And where AILit prescribes iteration as a general principle, we specify a governed procedure, FERC, that makes framing, comparison and accountable commitment assessable rather than dependent on individual disposition.
Policy landscape reviewed August 2026.
The Calibration dilemma
What should education preserve, and what should it redesign?
Not every traditional learning activity should disappear because AI can perform parts of it. Learners still need independent mastery wherever professional practice requires unaided reasoning, embodied skill or real-time judgment.
At the same time, education should not prepare students only for workflows that are already being transformed. Institutions must continuously assess which capabilities should remain independently mastered and which should be developed through structured human–AI collaboration.
Preserve independent mastery
Retain deep knowledge and unaided competence where future practice still depends on direct human reasoning, physical skill or immediate judgment.
Redesign for hybrid practice
Rework learning activities where AI will be part of professional practice, so students learn to frame, validate, refine and take responsibility.
THE GRADUATE THIS PRODUCES
The faglig kaospilot
The graduate this produces is a faglig kaospilot: A domain expert trained not just to survive labor-market turbulence, but to steer it. By grounding the living-laboratory mission in real organizational experimentation, this program yields graduates who carry the capacity to transition workplaces from basic AI automation toward durable human–AI collaboration.
Instead of education trailing behind market shifts, these graduates are the direct mechanism by which institutions actively lead the transition.
Institutional transformation
The HI Educational Transformation Playbook
The playbook connects institutional purpose, shared mindsets, educational practice and formal assessment. It helps institutions move from isolated experiments towards a coherent capacity for Hybrid Intelligence.
Four connected tracks move education from AI as an add-on towards institutionalized hybrid capability.
Governed human-AI collaboration
FERC keeps the learner in the author’s seat
Many approaches to AI in education focus on writing better prompts. FERC instead structures the full interaction through four phases: Frame, Explore, Refine and Commit. The process begins with human intent, creates space for comparison and refinement, and ends with accountable human commitment.
FERC is developed as a separate HI framework with its own research, practical guidance and examples.
Frame and Commit anchor the cycle in human intent and responsibility. Explore and Refine form the iterative middle.
From designed learning to learner designing
Educational AI often arrives as finished furniture: an approved chatbot, a fixed prompt template or a predefined set of rules. This is an appropriate starting point, but it should not be the endpoint.
Learners who are expected to enter AI-rich organizations must also practice designing the conditions under which AI is used. They should be able to define task boundaries, configure instructions and criteria, identify uncertainty, create escalation points and decide where human review is structurally required.
This is the idea of seamful co-creation. Instead of hiding every limitation and handover, the learning design makes important seams visible and actionable. Friction is not added everywhere; it is placed where thinking, evaluation and responsibility need to occur.
From friction to seamful co-creation
Two approaches to structuring human–AI interaction in education
| Aspect | Friction | Seamful co-creation |
|---|---|---|
| Core idea | Introduce pauses or resistance to slow automation and promote reflection | Make AI’s limits and handover points visible to structure human–AI interaction |
| Primary focus | Regulate student behavior and prevent shortcuts | Engineer an explicit division of labor between prediction and judgment |
| Relation to technology | Technology constrained through rules and restrictions | Technology treated as malleable infrastructure to be configured and shaped |
| Role of teacher and student | Teacher moderates; student reflects and evaluates | Teacher designs workflows; student acts as co-creator and citizen developer |
| Uncertainty and trust | Surface ambiguity to encourage critical questioning | Define escalation points, responsibility boundaries and verification practices |
| Educational trajectory | Classroom-level intervention to improve learning quality | Institutional architecture that cultivates durable hybrid intelligence |
Seamful co-creation moves beyond regulating the learner towards jointly shaping the human-AI system.
What it looks like in practice
The classroom as a laboratory for Hybrid Intelligence
The playbook does not prescribe one technology. It provides activity formats in which AI expands the field of possibilities while learners exercise contextual reasoning, epistemic vigilance and accountable judgment.
Iterative bot or prompt co-design
Students test an AI tutor, debate partner, case coach or workflow simulator against explicit quality criteria. They identify weaknesses, revise its instructions and test it again. The activity makes system malleability visible.
Virtual stakeholder exploration
AI generates stakeholders, personas or scenario variations that widen the analytical field. Students interrogate assumptions, assess realism, surface trade-offs and decide which perspectives should shape the final proposal.
Flawed-bot detection and repair
An AI configuration contains a deliberate misconception, weak assumption or bias. Students diagnose the flaw, articulate the relevant domain knowledge and redesign the prompt or workflow to reduce the chance of recurrence.
Practice-tested formats move students from conversational tool use towards system shaping and accountable repair.
HI business process re-engineering
Students select a professional workflow, map the prediction-judgment boundary, generate alternative human-AI configurations and commit to a defensible redesign. The exercise demonstrates why domain expertise remains essential: without contextual judgment, AI proposals tend to remain plausible but generic.
A structured workflow exercise connects classroom learning with the redesign of real professional practice.
Objectives and assessment
Assess judgment, not only task completion
Generative AI weakens the link between a polished submitted artifact and the learning it was intended to demonstrate. A strong final output may conceal weak framing, uncritical delegation or little durable understanding.
HI learning objectives are therefore written as observable judgment behavior: validate, detect, justify, stress-test, integrate and assume accountability. Foundational knowledge is not reduced. It becomes the basis for informed evaluation of AI-supported work.
From traditional to HI learning objectives
Illustrative shifts from task execution to judgment-centred augmentation
| Domain | Traditional learning objective | HI learning objective |
|---|---|---|
| Trade and logistics | Process customer orders correctly in an ERP system and ensure accurate invoicing. | Validate and critically assess AI-generated order proposals; detect anomalies and justify override decisions in light of pricing patterns, customer history and operational risk. |
| Business and marketing | Produce a complete market analysis report. | Evaluate, refine and justify an AI-supported market analysis, identifying its assumptions, risks and strategic trade-offs. |
| Engineering | Calculate structural load requirements using standard formulas. | Assess and stress-test AI-assisted load calculations, identify boundary conditions, and justify final design decisions under uncertainty. |
| Healthcare | Develop a treatment plan based on patient symptoms and clinical guidelines. | Critically evaluate AI-generated treatment suggestions, integrate patient-specific factors, and assume accountable clinical decision-making. |
Across domains, the shift is from completing predictable tasks towards validating, contextualizing and responsibly finalizing AI-supported work.
PROCESS-BASED ASSESSMENT
Assessment can include the learner’s original frame, the alternatives explored, comparative evaluation, revisions, the documented role of AI and the rationale behind the final commitment.
Employability and formation point in the same direction
Problem framing, contextual judgment and accountable commitment are not only workplace competences. They also support autonomy, co-determination and responsible participation in society. HI Education therefore connects professional relevance with the broader formative purpose of education.
Re-professionalizing teaching
Teacher judgment becomes more important, not less
When submitted work no longer certifies the learning it was designed to evidence, output-based assessment loses its warrant. The answer is not better detection. It is to shift evaluative weight onto what can still be observed: whether a learner framed the problem before generating, demanded and compared genuine alternatives, revised their criteria, and committed with a documented rationale.
Evaluating how a learner frames a problem, compares alternatives and justifies a decision cannot be reduced to a judgment-free metric. It requires professional teacher judgment.
That cuts against three decades in which educational governance has drifted from professional responsibility towards measurable output. Hybrid education asks institutions to reverse part of that drift: to reinvest in, and re-trust, the judgment of educators. The cost is real: professional development, redesigned assessment, and protected time.
Institutions must therefore invest not only in new assessment formats, but also in teacher development and renewed trust in educators’ professional expertise. This is part of the transformation itself: Students learn what their institutions model.
Institutional maturity
From isolated use to institutional identity
Institutions begin from different positions. The maturity model provides a shared developmental horizon: from AI being peripheral, through local experimentation and operationalized practices, to Hybrid Intelligence becoming part of institutional strategy and identity.
Institutional HI maturity levels
A progression from isolated AI use to institution-wide Hybrid Intelligence transformation
Level 4: Transformative
Strategic DNA
Human–AI collaboration is embedded in the institution’s identity. Teaching is proactively redesigned to strengthen graduates’ future readiness and, more importantly, to help shape a more human-centred society. HI is not an add-on; it is strategic orientation.
How the institution talks about AI: “Human–AI partnership is our identity.”
Level 3: Operationalized
Stable hybrid culture
AI use is guided by shared institutional norms and detailed principles that ensure quality and responsibility. Prediction and judgment are systematically embedded in activities. Hybrid practice is collectively anchored rather than individually improvised.
How the institution talks about AI: “AI collaboration is professional excellence.”
Level 2: Emerging
Individual experimentation
AI use becomes iterative. Individuals refine prompts and experiment continuously, but practice remains personal rather than institutionally structured.
How the institution talks about AI: “AI is legitimate, but it depends on skill.”
Level 1: Initiating
One-shot
AI is used sporadically in isolated interactions. Engagement is shallow: a single prompt is tried, and the response is either accepted directly or abandoned.
How the institution talks about AI: “Is this cheating?”
Level 0: Inactive
AI is not used in teaching or preparation and is rarely discussed with students or colleagues. It is seen as irrelevant or peripheral.
How the institution talks about AI: “AI is irrelevant, or a threat.”
The maturity journey asks whether hybrid competence is individually improvised or collectively cultivated and governed.
Equity and access
Hybrid competence must not become a new privilege
Judgment-centred education makes a demand that AI-literacy modules do not. It presupposes learners who can reach configurable AI systems, teachers with the time and training to design judgment points into their own subjects, and institutions able to govern how those systems are used. Where those conditions are missing, the framework does not simply fail to arrive. It can widen the gap it was meant to close.
Access to configurable systems
Prompt-level access to a consumer chatbot is not the same as the ability to configure, constrain and inspect a system. Seamful work (building a flawed bot, locking an assistant to a syllabus, marking handover points) depends on tooling that is currently distributed by budget rather than by educational need.
Teacher time and capacity
Process-based assessment costs professional time: designing judgment points, reading how a student framed a problem, defending an evaluation without a machine-readable metric. Institutions that adopt HI objectives without funding that time will produce the paperwork of hybrid education and none of the learning.
The risk is a two-tier settlement: well-resourced institutions teaching students to design and govern AI systems, while others teach students to consume their outputs, a division of hybrid labor established at school and carried into working life. We therefore treat access as part of the framework rather than as an implementation detail, and we would rather see this playbook adopted partially and fairly than fully and narrowly.
Use cases
-
Ribe Gymnasium
Educational Use Case
Ribe Gymnasium
Redesigning Teaching Around Human Judgment
-
International Business College
Educational Use Case
International Business College
HI Education Transformation Playbook at International Business College
-
TietgenSkolen hhx
Educational Use Case
TietgenSkolen hhx
Teaching Students to Use AI as a Partner, Not an Oracle
-
ETH Zürich
Educational Use Case
ETH Zürich
HI Education at ETH Zürich Student Project House
Collaborators
Educational collaborators
These affiliations reflect institutions represented by the authors and early cases. They do not in themselves indicate formal institutional endorsement of the manifesto.
