The learning contract of an AI school

Serious AI education is a joint production process: the school supplies structure, feedback, and standards; the student supplies preparation, effort, revision, and independent work.
Admission creates an opportunity to learn, not an entitlement to completion or a promise that the work will be easy.
A clear learning contract protects students from hidden requirements and protects the meaning of the qualification.

Why a Learning Contract Is Necessary

Students and institutions can enter the same program with incompatible expectations.

A student may believe that tuition purchases a credential, that recorded lectures guarantee flexibility without deadlines, or that prior academic success ensures technical progression. The institution may assume that every entrant understands the workload, mathematical preparation, revision requirements, and independent dissertation standard.

When those assumptions remain implicit, predictable difficulty becomes perceived breach.

A learning contract makes the educational relationship explicit. It is not primarily a legal document. It is a statement of reciprocal academic responsibility.

Education as Joint Production

Let student learning be

where:

• $Q$ is curriculum and teaching quality;
• $P_i$ is prior preparation;
• $E_i$ is sustained effort;
• $F_i$ is feedback received and used;
• $R_i$ is revision and retesting;
• $X_i$ is external time and resource capacity; and
• $\varepsilon_i$ represents unpredictable conditions.

The school controls much of $Q$ and the availability of $F_i$. The student controls much of $E_i$ and $R_i$. Both influence preparation and route placement. Neither controls $\varepsilon_i$.

The model rejects two convenient fictions: that a strong institution can educate a student without student work, and that student effort can compensate for any institutional weakness.

What the School Owes

The school should provide:

1. stated learning outcomes;
2. an intelligible curriculum and prerequisite path;
3. accurate program-level descriptions;
4. qualified instruction;
5. access to learning materials;
6. assessment aligned with declared outcomes;
7. timely and relevant feedback;
8. consistent academic standards;
9. routes for questions, review, and appeal; and
10. honest information about workload and completion.

Table 1. Reciprocal obligations in the learning contract

Institutional obligation	Observable evidence
Curriculum coherence	Dependency map and course outcomes
Qualified instruction	Subject command, teaching review, and current materials
Fair assessment	Rubrics, comparable prompts, and moderation
Useful feedback	Diagnosis linked to required revision
Consistent standards	Common thresholds and review procedures
Student support	Clear access, response, and remediation routes
Honest communication	Published requirements and limitations

The institution should not hide a weak course behind the language of student independence.

What the Student Owes

The student should:

• complete prerequisite preparation;
• allocate the declared study time;
• attempt work independently before seeking replacement;
• identify uncertainty rather than conceal it;
• use feedback to revise;
• preserve academic integrity;
• communicate constraints early;
• contribute responsibly to group work;
• meet progression standards; and
• own every claim in the final work.

Effort is necessary but not sufficient. A student can work intensely with an ineffective strategy. The contract therefore includes willingness to change how one studies.

Admission Is Not Completion

Admission is a decision that the applicant has a plausible route to the program's outcomes. It does not establish that those outcomes already exist.

Let admission evidence be $A_i$ and completion standard be $C$. A rational admissions rule may be

for some threshold $\tau<1$.

Because $\tau$ is less than one, some admitted students will not complete. The institution should reduce avoidable failure through placement and support, but it cannot turn the completion standard into an admission guarantee.

This distinction should appear before enrollment, not after the first failed assessment.

Difficulty Should Be Declared and Designed

Difficulty has several sources:

Read as a diagnostic rather than a second scorecard, the framework becomes clear: Mathematical prerequisite — Legitimate when: Required by later claims; Institutional response: Diagnostic and foundation route. Problem ambiguity — Legitimate when: Tests formulation and judgment; Institutional response: Clear evidence boundaries and rubric. Workload — Legitimate when: Supports practice and revision; Institutional response: Accurate time estimate and pacing. Technical implementation — Legitimate when: Necessary for reproducibility; Institutional response: Scaffolding and standard tools. Language and communication — Legitimate when: Required for professional argument; Institutional response: Explicit criteria and support. Obscure presentation — Legitimate when: Rarely educational; Institutional response: Redesign.

A hard program is not necessarily a good program. Legitimate difficulty must connect to an intended capability.

The school owes students an explanation of that connection.

The Time Budget

Self-paced education can create the impression that time requirements disappear. They do not.

Let weekly study time be

If available time $B_i$ satisfies $B_i<H_i$ for a sustained period, the student must change pace, reduce other commitments, or pause.

The institution should publish realistic ranges rather than minimum viewing time. Watching a lecture is not equivalent to reconstructing its argument.

Feedback Is an Offer That Must Be Used

Feedback has no educational effect if it remains unread or produces only a corrected file.

Let initial response be $\theta_t$ and feedback direction $g_t$. Learning requires a student-owned update:

followed by a new task that tests whether the correction transfers.

The school owes a useful $g_t$. The student owes the revision and retest.

Repeated resubmission without evidence of a changed model is not meaningful progression.

Academic Freedom and Program Standards

Students should be able to disagree with instructors, select alternative models, and criticize institutional assumptions. They should not expect every answer to receive equal evaluation.

A defensible alternative must satisfy:

The rubric should reward a reasoned disagreement and reject an unsupported assertion. Academic freedom concerns inquiry; standards concern evidence.

The Role of AI Tools

AI tools can assist with explanation, code, editing, and search. They do not transfer responsibility.

The learning contract should require:

• disclosure where required;
• verification of citations and derivations;
• preservation of intermediate reasoning;
• independent explanation;
• adaptation to unseen conditions; and
• responsibility for errors.

If the student cannot reconstruct the submitted work, the artifact may be complete while the educational outcome is absent.

Progression, Pause, and Exit

A serious learning contract needs more than pass and fail.

Possible states include:

A student may need to pause for external reasons, return to foundations, or move to a route that better matches professional goals. These options should be academically defined rather than negotiated only after crisis.

An exit can be responsible when the program is no longer the right investment. It should not be disguised as successful completion.

The Dissertation Contract

The dissertation changes the relationship.

The school provides supervision, milestones, review, and examination. The student owns:

• the question;
• the evidence;
• the method;
• the response to failure;
• the written argument; and
• the final defense.

Supervisor contribution $S$ should support, not replace, student contribution $I$:

If supervision raises document quality by taking over the student's judgment, the educational standard has not been met.

Group Work and Individual Ownership

AI projects are collaborative, but qualifications are awarded to individuals.

A group component can assess coordination and professional practice. Individual components should verify:

• understanding of the complete model;
• ownership of assigned work;
• ability to criticize team choices;
• response to an individual perturbation; and
• communication without team assistance.

The contract should state which evidence is collective and which is individual.

Informed Educational Consent

Before enrollment, students should receive enough information to make a rational commitment.

The decision can be represented as

The institution cannot calculate the expression for the student, but it should disclose the inputs it knows:

• expected workload and pacing;
• mathematical prerequisites;
• assessment forms;
• progression and completion rules;
• available support;
• dissertation obligations;
• tuition and material additional costs;
• differences among program tracks; and
• the absence of guaranteed employment.

Informed consent does not require making the program sound unattractive. It requires removing surprises that are material to the decision.

The same obligation continues when requirements change. Current students need transition rules and a reasonable opportunity to complete under a known standard.

Resolving Disagreement

Transparent procedures protect both parties.

When a student disputes an evaluation, review should ask:

1. Was the learning outcome stated?
2. Did the assessment measure it?
3. Was the rubric applied consistently?
4. Is there evidence of factual or procedural error?
5. Does the student's alternative meet the same standard?

Review is not an automatic grade negotiation. It is a quality-control mechanism.

The SIAI GSB Learning Contract

At SIAI GSB, flexible online access should coexist with demanding evidence of learning.

The school provides the curriculum, recorded instruction, cases, feedback, assessment, and dissertation structure. Students choose when and where much of the study occurs, but they remain responsible for preparation, effort, correction, and independent performance.

Different program tracks can carry different mathematical obligations. The contract should make those differences visible before enrollment.

The contract should be operational rather than ceremonial. Workload estimates, prerequisite expectations, permitted uses of AI, response times for feedback, reassessment rules, and the conditions for pausing or exiting should be stated before conflict arises. The student can then judge whether the commitment is feasible, and the school can distinguish a genuine failure of support from a refusal to engage with the work. Explicit rules do not remove faculty discretion; they give discretion a defensible boundary. In a demanding program, clarity is part of academic care because it allows difficulty to arise from the subject rather than from avoidable uncertainty about the institution. Periodic review should compare the written contract with actual student experience. Repeated deviations are evidence of a system problem, not isolated communication failures. The contract is fulfilled through repeated conduct, not merely through language published in a handbook. Consistency makes reciprocal expectations credible.

AI use makes reciprocal obligations more explicit. SIAI’s research on cognitive outsourcing argues that students must retain responsibility for verification. Why AI Detection Is Failing Higher Education shows that institutions also owe students fair process rather than unreliable accusations. The Economy’s work on teacher AI literacy adds a corresponding faculty duty: instructors must be capable of setting and explaining the boundary of acceptable use. A learning contract is credible only when responsibility runs in both directions.

Conclusion

The learning contract of an AI school is a reciprocal commitment to education rather than credential exchange.

The institution owes coherent design, qualified instruction, fair assessment, feedback, and honest communication. The student owes preparation, sustained work, revision, integrity, and independent ownership.

Clear responsibilities do not make the education less humane. They allow difficulty, support, failure, and correction to be interpreted accurately.

Admission opens a path. The learning contract defines how the path is traveled and what must be demonstrated at its end.

References

John Biggs, “Enhancing Teaching through Constructive Alignment”, Higher Education 32 (1996): 347-364.
Vincent Tinto, Leaving College, 2nd ed., University of Chicago Press, 1993.
Standards and Guidelines for Quality Assurance in the European Higher Education Area, ESG 2015, 2015.
Swiss Institute of Artificial Intelligence (2026) ‘Cognitive Outsourcing in Education: Why AI’s Real Classroom Crisis Is Verification, Not Cheating’, SIAI Working Papers, 24 July.
Swiss Institute of Artificial Intelligence (2026) ‘Why AI Detection Is Failing Higher Education’, SIAI AI Memo, 30 August.
The Economy Editorial Board (2026) ‘Teacher AI Literacy Is the Real Test of AI in Education’, The Economy Review, 22 June.