Why the question has changed
For much of modern education, producing an answer was itself a substantial intellectual task.
A learner had to remember, search, organize, synthesize and write before a finished response existed.
Generative AI compresses parts of that process. It can create an outline, explanation, summary or set of options in seconds.
That creates an attractive but weak inference: if the answer is easy to obtain, perhaps learning the knowledge behind it matters less.
The problem is that answer access and answer judgment are different capabilities.
A response can be fluent and wrong. It can be factually correct but irrelevant to the local situation. It can omit a constraint, hide an assumption or optimize the wrong objective.
A polished surface can make weak reasoning harder to notice.
The educational question therefore moves downstream.
The scarce human capability is not always producing another candidate answer. It is determining which question matters, what can be trusted, what is missing, what trade-offs are acceptable and who remains responsible when an answer becomes action.
Education cannot respond by pretending AI does not exist. Nor can it conclude that knowledge no longer needs to be learned.
| Abundant production | Human responsibility |
|---|---|
| Answers | Framing and context |
| Drafts | Verification and judgment |
| Options | Decisions, trade-offs and consequences |
Knowledge becomes a basis for judgment
A person needs something against which to evaluate an answer.
Disciplinary knowledge remains the foundation of competent supervision.
Someone who understands a domain can notice a misused concept, an implausible causal claim, an omitted variable or a conclusion that exceeds the available evidence.
Without that foundation, verification risks becoming procedural.
A learner may confirm that a source exists without understanding whether it actually supports the claim. They may ask the same AI system to critique its own answer without an independent basis for deciding whether the critique is adequate.
This does not imply memorizing everything a machine can retrieve.
It means building mental models rich enough to support judgment.
Mathematics, science, history, languages, economics, law, design and other disciplines still matter.
What changes is what education must ask knowledge to do.
Can a learner use it to detect error? Can they recognize a missing condition? Can they identify when the problem has crossed into another field? Can they know when another expert must enter the conversation?
AI makes knowledge valuable not only for producing answers, but for supervising them.
Learn to frame before asking
If a system can produce many answers quickly, defining the problem becomes more valuable.
Real problems rarely arrive with a perfectly specified prompt.
Someone has to decide what the problem is, whose problem it is, what outcome matters, which constraints are real and what evidence would count as improvement.
That work requires context.
This is why “learning to ask” cannot be reduced to prompt engineering.
A technically excellent prompt can still operationalize a poor question.
The more consequential intellectual move may happen before the AI system is opened.
Next Human Learning begins with the capacity to revisit the frame: question assumptions, unlearn a weak model and reformulate the problem when evidence shows that the original question was inadequate.
Learn to verify before believing
Abundant answers create an epistemic problem: plausibility can masquerade as certainty.
Fluent language lowers friction. That is useful, but it can also reduce the impulse to inspect what sits underneath the prose.
Human learning therefore needs the ability to pause between answer and belief.
What is supported? What is inferred? What source would be appropriate? What context is missing? How confident should we be? What happens if this assumption is wrong?
These are questions of epistemic judgment.
Real-Life Readiness describes preparedness partly as the ability to distinguish evidence, assumptions and plausible claims, and to know what requires verification before action.
In an AI-rich environment, that becomes a practical literacy for everyday decision-making.
Learn to cross boundaries without losing standards
AI systems can combine language from many domains fluently.
Real-world problems also cross domains — but they do not erase the different standards those domains use.
An energy solution can be technically sound and economically impossible. A medical intervention can be clinically effective and fail because of behavior or access. An education technology can function perfectly while creating poor pedagogical incentives.
Humans therefore need two complementary capabilities: depth and breadth.
Depth makes it possible to recognize when something is wrong inside a field.
Breadth makes it possible to recognize when the problem has moved beyond that field.
This is not a call for everyone to become an expert in everything.
It is a call for boundary awareness: knowing where competence ends, what other knowledge is required and who should enter the decision.
Collaboration then becomes a verification architecture, not merely teamwork.
- Depth
- Detects errors within a domain.
- Breadth
- Detects the boundaries of knowledge.
- Collaboration
- Adds missing expertise.
Learn to decide when no option is complete
AI can propose alternatives. It cannot remove trade-offs.
Many real decisions cannot maximize cost, safety, speed, sustainability and inclusion simultaneously.
Values conflict. Resources are finite. Some choices are reversible; others are not.
At that point, more options do not eliminate judgment.
This is where Agency matters.
Within the Mars Challenge framing, Agency is authorship under consequences: forming judgment, choosing, acting and accepting responsibility.
If a learner can generate ten recommendations with AI but cannot explain why one deserves to move forward, they have demonstrated production capacity rather than preparedness.
More generated options increase the need for competent authorship.
- AI output
- Verify
- Contextualize
- Integrate
- Judge
- Decide
- Act
- Reflect
Learn by making answers meet reality
An answer remains informational until it encounters the world.
Once an idea is prototyped, tested or implemented, new information appears: resistance, cost, unintended behavior, failure, adoption problems or consequences.
That feedback is part of learning.
A prepared person does not stop when the document is finished.
They can test, observe, revise and explain what changed.
They can ask who benefited, who carried the cost, what assumption failed and what should happen next.
This distinction becomes particularly important when AI makes first drafts cheap.
If the first version is nearly effortless, educational value increasingly lies in what a learner does after the first version: what they validate, reject, improve and take responsibility for.
Reality becomes part of the curriculum.
So what should humans learn?
The supplied evidence does not justify a universal, permanent curriculum for the AI era.
It does support a structural answer.
Humans need disciplinary knowledge, because judgment without a knowledge base becomes dependency.
They need to learn, unlearn and relearn, because tools, evidence and contexts change.
They need to frame problems and ask better questions, because better output cannot rescue a badly defined problem.
They need to verify and calibrate confidence, distinguishing evidence from inference and fluency from reliability.
They need to connect disciplines and people, recognizing when their own competence is insufficient.
They need to create and experiment, because the future cannot be built solely by selecting among existing alternatives.
They need to decide under uncertainty, because complete information rarely arrives before action.
They need to learn from feedback, because reality reveals what analysis misses.
And they need to remain responsible for consequences, because generating an answer is not the same as having authority to decide what should become real.
This is the direction Mars Challenge calls Next Human Learning: not less human learning, but learning redesigned for an environment in which answers are abundant and judgment remains consequential.
What the evidence says
Real-Life Readiness: A Foundational Framework (2026) provides the strongest supplied evidence base for this article.
It draws on National Academies work on transfer, where deeper learning includes the ability to use knowledge and skills in new situations.
The OECD Learning Compass emphasizes agency, values and responsibility in navigating unfamiliar contexts.
The Future of Jobs Report 2025, as referenced in the framework, describes substantial change in skill demand while continuing to foreground analytical thinking, creativity, resilience, flexibility, leadership and systems thinking.
The framework also cites 2025 research on generative AI and critical thinking examining self-reported cognitive effort, confidence and interaction with AI tools.
That research should not be converted into the claim that AI “makes people less intelligent”.
Its relevance is narrower: it raises empirical questions about cognitive offloading, trust and the role of prior expertise in critical engagement with AI outputs.
The responsible conclusion is that AI changes the conditions under which cognition and learning occur.
The supplied evidence does not support abandoning disciplinary knowledge. The framework explicitly argues that sufficient depth remains necessary to detect when a fluent answer is incomplete or wrong.
What we observe in Mars Challenge's design
First-party Mars Challenge documentation supports a claim about learning design, not causal impact.
The participant process asks teams to identify strengths, assign roles, generate and select ideas, investigate, define problems, prototype, receive feedback, revise and explain the Earth return of their solution.
The guide explicitly frames the process as building with depth rather than simply moving quickly, and links superficial work with weaker decisions and solutions.
Mars Challenge's 2026 framing of Next Human Learning adds learning, unlearning and relearning, radical creativity, ethical awareness, collaboration, resilience and planetary imagination.
This creates an environment in which an initial answer is not enough.
Participants have to make choices, expose reasoning, confront constraints and revise what they produce.
That architecture is consistent with the thesis of this article.
It does not prove that every participant develops these capacities or that participation causes durable change.
What this means for schools and universities
AI forces institutions to distinguish output from learning.
A polished essay can no longer be treated automatically as evidence that a learner understands the topic, can identify an error or can defend the choices inside the text.
Assessment can capture more of the process: sources used, assumptions identified, AI use where relevant, verification steps, rejected alternatives, feedback, revisions and the reasoning behind a final decision.
There may also be legitimate moments when AI assistance is restricted because the objective is to build internal knowledge or practise a cognitive process that later needs to supervise external tools.
Other tasks may deliberately permit AI and assess the quality of that supervision.
These positions are not contradictory.
Education now has two simultaneous responsibilities:
build internal knowledge and teach people to govern external intelligence.
That is a higher bar than simply teaching students to produce answers.
Related concepts
Evidence in practice
- Mars Challenge 2026 — Participant Guide
- Mars Challenge 2026 — Program Framework / Earth
No individual project is used as proof of causal effect.
Frequently asked questions
If AI can retrieve information, why memorize anything?
Because judgment requires enough internal knowledge to recognize relationships, anomalies and missing context. That does not mean memorizing every retrievable fact. It means building a conceptual base from which better questions and verification become possible.
Can critical thinking replace knowledge?
No. Critical thinking needs something to think critically about and standards by which to assess it. Without domain knowledge, a person may notice superficial inconsistencies while missing technical or conceptual errors.
Should prompt engineering become a core subject?
Knowing how to interact with AI is useful, but prompting is too narrow to organize education around. A strong prompt cannot compensate for a weak question or insufficient knowledge. The durable capability is governing the intellectual process in which AI participates.
Should AI be banned in assessment?
The supplied sources do not justify a universal rule. Some tasks may restrict assistance to build foundational knowledge; others may deliberately permit AI and assess verification, reasoning and judgment.
How does this relate to Next Human Learning?
This article develops the problem that Next Human Learning addresses. The concept extends the answer toward learning, unlearning, relearning, creativity, ethical awareness, collaboration, resilience and purposeful human action.
References
Real-Life Readiness: A Foundational Framework Adelino Sousa / Mars Challenge / Virtual Educa 2026
Education for Life and Work: Developing Transferable Knowledge and Skills in the 21st Century National Research Council / National Academies Press 2012 https://doi.org/10.17226/13398
OECD Learning Compass 2030 / Future of Education and Skills OECD
The Future of Jobs Report 2025 World Economic Forum 2025
The Impact of Generative AI on Critical Thinking Lee, H.-P. H. et al. / Microsoft Research & Carnegie Mellon University 2025
Mars Challenge 2026 — Program Framework / Earth Mars Challenge 2026
Guia Participante Mars Challenge 2026 FINAL Mars Challenge 2026
What task still requires human skills when AI supplies the answer?
Human Skills in the Age of AI become visible when a task requires verification, context and ownership of a choice. Rather than asking only for a text about sustainability, ask for a recommendation for a specific situation, supported by traceable sources, constraints and reasons for rejecting alternatives.
As a proposed editorial practice, a team may use AI to suggest options and then investigate what those options omit. Learners should distinguish available data, assumptions and missing information. Edition and institutional rules prevail; this activity does not automatically authorise a tool.
UNESCO provides an AI competency framework for students. The following activity is a Mars Challenge proposal, not an official UNESCO activity or evidence of effectiveness. Examine documented reasoning, participation and revision; fluent generated text does not demonstrate human mastery.
| Stage | Learner task | Evidence to retain |
|---|---|---|
| Verify | Check two consequential claims against original sources. | Source, date and discrepancy found. |
| Contextualise | Identify a constraint that changes the recommendation. | Local condition and revised alternative. |
| Take responsibility | Explain the choice and when it needs revision. | Rationale, limitation and review condition. |
Hypothetical example: AI recommends a water-saving solution that requires power unavailable locally. The learning to discuss is verification and the changed decision, not the quality of the first draft.
UNESCO · AI competency framework for students (2024; atualizado em 2026)
Explore next
Next step
Deepen: Next Human Learning
See evidence: Projects and Evidence
Participate: How to take part