This question is also personal. As a student in Portugal, I encountered what I would call the adversarial layer of education: tests and examinations were often designed as confrontations between students and institutions. Difficulty itself was treated as evidence of rigour.

I survived that system partly because I had access to excellent books. Many equally capable colleagues did not and faltered. What appeared to be meritocracy was also, to some extent, unequal access to the resources needed to decode what education had failed to make clear.

The problem persists today. We continue to confuse opacity with rigour and failure with demanding education. In some subjects and examinations, very high failure rates coexist with an expanding dependence on private tutoring. When success requires families to purchase a parallel educational system, the examination is no longer measuring only what the school has taught.

When I later studied in the United States, I encountered something different. The demanding dimension of education had not disappeared, but it was more transparent. Students understood what they were expected to master and how they would be assessed. There were bridges between theoretical formulations and practical applications reducing abstraction.

Examinations consequently felt easier not necessarily because the intellectual standards were lower, but because difficulty was not created artificially through ambiguity.

At more advanced levels, that demanding layer acquired an entirely different purpose. It became a bridge to new ideas. Knowledge was not the destination; it was the structure that made discovery possible.

Artificial intelligence, simulation, software, sensors, robotics, 3D printing and printed electronics now allow us to take the next step. Students can move beyond understanding existing ideas. They can generate original hypotheses, model their consequences, fabricate prototypes and test them against reality.

There is also greater awareness today of what can happen after a prototype works: a research project may become a product, a company, a public intervention or even a new industry. In this respect, much of the world has learned from MIT and Stanford.

But one essential participant has remained comparatively absent: Nature.

Nature should not merely supply examples for science classes. It can become a source of strategies, materials, constraints, intelligence and surprise. It establishes the conditions within which Humans and Machines must operate.

Nature Inspirer, a proposed system through which a student could develop a nature-inspired product for the home of 2035, is presented in a concept site.

The Human defines the problem, purpose and values.

Nature provides strategies and establishes limits.

The Machine searches knowledge, connects distant ideas, generates alternatives and simulates consequences.

The student then fabricates a prototype and allows reality to answer.

Assessment must therefore ask more than “What knowledge did you acquire?”

What did you imagine?
What did you build?
Does it work in the real world?
What resistance did you encounter?
Who benefits—and who might be harmed?
Can you defend your decisions?

The prototype is not the answer. It is a hypothesis. Reality is the examiner.

None of these principles is new to my own practice. They have informed my teaching during four decades at university and now guide our work with younger students at Seixal Criativo.

What is new is the scale of what has become possible.

We can finally build an education in which Nature sets the limits, Humans define purpose and assume responsibility, and Machines enlarge imagination and foresight.

That, I believe, is the education of the future.

The report introduces a particularly apt expression: “cognitive surrender.” Obtaining a correct answer quickly can create an illusion of learning, leading students to hand over to the machine the effort, doubt and resistance through which imagination, judgment and intellectual autonomy are developed.

However, I believe the report is still stronger at protecting education from AI than at imagining a new form of education made possible by AI.

Its dominant question appears to be: How can we preserve genuine learning when machines can already perform much of the cognitive work traditionally required of students?

A second, more ambitious question is missing: What can students now imagine, model and build that would previously have been impossible?

A university prepared for the age of AI should not limit itself to educating competent and responsible users of these tools. It should educate builders capable of combining human intelligence, AI models, agents, simulations, sensors, robots, institutions and natural systems to intervene responsibly in the world.

The fundamental unit of education may therefore gradually cease to be the “AI-aware” discipline and become the Human–Machine–Nature project.

Instead of asking only for answers that a machine can produce, we can challenge students to build living models of an estuary, a city, an organization, an ecosystem, a sports team or a community. These models should integrate data, represent hypotheses, simulate possible futures, make uncertainties explicit, confront different perspectives and, above all, be tested against reality.

In this form of education, the central question of assessment will no longer be only, “What knowledge have you acquired?” It will also ask: What did you imagine? What did you build? Does it work in the real world? What resistance did you encounter? Who benefits? Can you defend your choices and understand their consequences?

MIT correctly identifies the first great imperative: we cannot allow students to surrender their capabilities to AI.

I would add a second: we must use AI to radically amplify the human capacity to imagine, understand and build worlds that neither humans nor machines could create independently.

I am certain that MIT students will do precisely this, benefiting from the programs—particularly UROP, Sandbox and the $100K Competition—and the laboratories available to them.

Subscribe now

July 2026

Executive Summary

Every education system alive today was built for a world in which knowledge was scarce and procedure was the highest form of competence. That world is ending. A commercial AI model can already outscore most students on the tests schools use to sort them, and roughly 70% of exam success turns out to depend on recognising a small number of adversarial “trap” patterns rather than on understanding- a meritocracy of trick-spotting, not of thought.

This changes the question education must answer. It is no longer what should we teach? It is: what must a human become so that civilization keeps generating novelty once machines can perform most existing cognitive tasks?

The answer proposed here rests on a single architecture, assembled from three converging efforts — a K-12 platform, a university-and-venture framework, and a global cognitive-gaming platform — and grounded in a deeper philosophical claim about why humans remain irreplaceable. It has:

• Three layers of knowledge applied to every concept, at every age, so that understanding always includes knowing where it breaks.

• A Human + Machine + Nature triad, in which nature is not a subject but a third teacher and a design laboratory.

• A zone-and-stage architecture carrying a learner from first self-discovery through school, university, and venture creation.

• A human development layer that treats agency- not intelligence -as the actual bottleneck for most students.

• An AI engine that is invisible infrastructure, not a replacement teacher.

• A final purpose that is neither knowledge nor employability, but the capacity to generate valuable surprises and, ultimately, to redesign reality responsibly.

AI predicts. Humans surprise. Nature generates. Communities activate. What follows is the case for building education around that sentence.

Part I — Why the Current System Is Wrong on Its Own Terms

Mathematical and scientific competence has long been measured as procedural fluency: the ability to execute a known method quickly and accurately. That measure now fails on two fronts at once.

First, it trains out exactly the capacities machines cannot replicate-creativity, lateral thinking, the capacity to be surprised-while rewarding the capacities machines are already better at.

Second, most high-stakes exams contain a hidden adversarial layer: questions engineered to trigger predictable errors rather than to test understanding. Roughly 70% of exam success comes down to recognizing a small number of these traps. A private-tutoring economy has grown up specifically to teach pattern recognition that has nothing to do with the underlying subject. That is a meritocracy of trap-recognition, not of knowledge.

A second, independent gap runs alongside the first, and it is arguably the more important one. For the strongest students, education is the bottleneck: give them better knowledge, harder problems, and they flourish. For a much larger group- disengaged, under-resourced, never shown that their choices shape their life- the problem is not intelligence but identity. They lack agency, not aptitude.

Any new system has to close both gaps at once: it must teach people to out-think exams and trained patterns generally, and it must give the majority of students, not only the top decile, but a credible reason also to believe their own choices matter.

Part II — Why Humans Still Matter: The Philosophical Root

Before describing an architecture, it is worth being precise about why the human role in this system is not merely sentimental. A model trained in everything humanity has written can already out-know almost any individual. What it cannot do is originate the reasons anything matters.

Several distinct arguments support this:

• Humans set genuinely new objectives. No system wakes up wanting to cure a disease, climb a mountain, or understand the universe. Every major human endeavor begins with someone deciding that something matters. Direction-setting is a human act.

• Humans live inside reality. All machine knowledge ultimately derives from human observation, experiment, and record-keeping. A person can notice something in a forest that has never been recorded and find out why. Machines cannot independently expand humanity’s empirical experience.

• Humans have skin in the game. Every human decision carry consequences for reputation, family, health, sometimes survival. That produces a form of judgment that reading alone cannot.

• Humans surprise each other. The future is not merely hard to predict computationally- it is unknown because billions of people are simultaneously inventing technologies, institutions, and ideas nobody modeled in advance. No amount of central planning ever anticipated the personal computer.

• Humans hold values rather than merely describing them. A model can discuss ethics without preferring anything. The questions “what should we build?” and “what future do we want?” remain irreducibly human questions.

The practical conclusion follows directly: education’s job is no longer primarily to transmit knowledge or even to build problem-solving skills, since both are increasingly commoditized by machines. Its ultimate purpose becomes maximizing humanity’s long-term capacity to generate surprises, new goals, new values, and new worlds while machines handle everything that has already been figured out.

Part III — The Three-Layer Knowledge Structure

Across every version of this architecture, the same underlying structure recurs, described in slightly different vocabularies: Core / Application / Adversarial, Fundamental / Operational / Frontier, and the Cheat Sheet layers of the Knowledge Infrastructure. Collapsed into one model:

Layer

What it teaches

Where it shows up

Foundation (Core / Fundamental)

The concept itself- definition, intuition, worked examples. “What is this?”

Settled, stable knowledge; changes slowly

Application (Operational)

How the concept is used to solve known problems; exam-style exercises and variations. “How is this used?”

Evolve every few years as tools and methods change

Frontier / Adversarial

Where the concept fails: hidden assumptions, edge cases, trick questions, paradoxes, engineering or market constraints. “How can this fail — and where does it break open into something new?”

The throughline of the entire system

The third layer deserves special weight because it is not merely defensive. At school it appears as tricky edge cases; at university it becomes an open research question; in a venture it becomes the constraint a real market imposes. It is the same cognitive instinct -find where the model stops working -aimed at progressively larger problems, and it is precisely the instinct that lets a student one day audit an AI system’s mistake rather than simply trust its output.

A structured way to operationalize this layer: catalogue recurring “trap” patterns publicly (an open, community-editable register of adversarial patterns), tag each with the logic instruction that defeats it, and, critically, pair each one with a real-world illustration, ideally a documented case where failing to apply that exact logic had real consequences (a satellite miscalibration from a unit-conversion error; a bridge failure from dismissing measured data in favour of qualitative impression; a trading-system loss from applying the wrong logic to a boundary condition). Naming the traps publicly is itself a form of reform: once a trick is documented and taught, it stops being a paid secret and becomes curriculum.

Part IV — Human + Machine + Nature: Three Teachers, Not One

The defining move of this architecture is refusing to treat “AI in education” as the whole story. A third teacher is added deliberately: Nature.

Why Nature is a teacher, not a subject

The philosophical grounding is direct: nature should no longer appear merely as a subject -biology, ecology, environmental science-to be studied and tested on. It becomes the greatest teacher of design. Where the Fundamental/Operational/Frontier layers organize what is taught, Nature reorganizes where the material comes from. Students learn:

• evolution

• resilience

• adaptation

• cooperation

• emergence

• circularity

• ecosystems

• long-term thinking

Nature becomes the laboratory for future engineering — not a decorative example bolted onto a physics lesson, but the actual generative source of curriculum. Concrete translations of this idea, drawn across the school- and platform-level architectures:

• Bird flight → lift equations, energy optimization, path planning (physics and mathematics)

• Plant growth → resource allocation, branching logic, fractals (algorithms)

• Ant colonies → optimisation and swarm intelligence (computer science)

• Termite mounds → passive cooling systems (engineering and biology)

Every concept in the curriculum should be able to answer the question: where does this exist in nature? Simulation worlds built on this principle -ecosystems, swarms, climate systems, cities modeled on metabolic logic -turn abstract problem-solving into the modeling of living systems, and prototyping labs are pushed toward bio-inspired builds: robots that move like animals, energy systems modeled on forests, networks modeled on fungal mycelium.

The complementary triad

Putting the three teachers side by side clarifies what each contributes, and what none can replace:

Human dimension

Machine’s contribution

Nature’s contribution

Conscience

Optimisation

Ecological constraint and balance

Surprise

Prediction

Evolutionary innovation and emergence

Imagination

Algorithms

Beauty, complexity, systems wisdom

Dexterity

Automation

Biomechanical elegance and efficiency

Body intelligence

Sensors

Embodied adaptation and resilience

Shared perspectives

Multimodal synthesis

Interspecies and ecosystem interdependence

AI predicts. Humans surprise. Nature generates. Communities activate. A system that only builds the first two teachers into its architecture is still, in an important sense, unfinished.

Part V — A Zone-and-Stage Architecture: From First Classroom to Company

Three versions of this same architecture exist at different altitudes — six zones for school-age learning, seven stages spanning childhood to company formation, and a global gaming-platform layer sitting on top of both. Synthesized into a single developmental spine:

Zone -1 — Explorer Discovery. Before any subject is taught, students spend structured time discovering what they are good at, what excites them, and what kind of explorer they are (archetypes commonly include Builder, Scientist, Artist, Caregiver, Steward of Nature, Storyteller, and others). Every student is guaranteed one visible early success and one public presentation within the first month-small victories that create momentum, momentum that creates identity, identity that creates aspiration. This zone exists because, for the widest population of students, agency rather than ability is the actual constraint.

Zone 0 — Nature Lab. Real-world patterns become the entry point for concepts across physics, biology, algorithms, and engineering, as described in Part IV.

Zone 1 — Knowledge Grid. A navigable, zoomable map of every concept, each built on the three-layer structure, linked to its Nature analogue and to its entry in the open adversarial register.

Zone 2 — Challenge Arena. Students face adversarial problems drawn from real exams and cross-domain puzzles; an AI layer classifies errors in real time and generates personalized variants targeting individual weak spots. A game-like ranking structure (naming, badges, seasons) gives this zone a motivational engine that “beat the trap” language supplies on its own — students are not selling themselves on “better test prep,” they are joining a status game built around mastering hidden systems.

Zone 3 — Simulation Worlds. Playable causal models-cities, ecosystems, economies, health systems -where changing one variable exposes its consequences elsewhere. This is where imagination becomes operational rather than merely aspirational.

Zone 4 — Prototyping Lab. Where digital understanding meets physical reality: sensors, code, and builds inspired by nature.

The loop that ties the zones together is not linear: Discover → Observe Nature → Understand → Stress-Test → Simulate → Build → Reflect → Repeat. A student may join at whichever point fits what they need that day.

From school into the university and the venture

Where the school-age architecture stops at Zone 4, the university-and-venture framing extends the same instinct through seven stages, because the same three-layer knowledge structure and the same Human-Machine-Nature triad apply just as well to a doctoral researcher as to a ten-year-old:

Stage

Focus

1. Build the Human

Curiosity, communication, mathematics, systems thinking, ethics, learning how to learn

2. Build the Explorer

Electronics, software, biology, fabrication, robotics, AI, design — always through projects

3. Build the Discoverer

Asking questions, designing experiments, reading papers, using AI to generate and reject hypotheses (university begins here)

4. Build the Inventor

Prototype, patent, protect, benchmark, iterate

5. Build the Entrepreneur

Customer discovery, market validation, supply chain, regulation, pricing, distribution

6. Build the Company

Funding, recruitment, culture, production, sales, operations

7. Build the Ecosystem

Mentors, investors, customers, governments, universities, manufacturers, media, research labs

Stage 7 deserves special attention: many promising founders and researchers stall not for lack of technology but for lack of the surrounding ecosystem -so ecosystem-building is treated as a taught, scaffolded stage rather than an accident of who a student happens to know.

The traditional university sequence (Knowledge → Exercises → Laboratory → Research → Publication → occasionally a company) and the traditional startup sequence (Problem → Prototype → Customer → Pivot → Company) are replaced by a third sequence in which science recurs rather than sitting only at the start:

Human curiosity → AI expands knowledge → Student explores possibilities → Science identifies what is possible → Prototype → Users → Science deepens → Patent / IP → Product → Company

Science appears twice on purpose: discovery does not stop once a prototype meets its first users, it deepens in direct response to what reality reveals.

The university reframed

The best institution to hold Stage 3 onward is not a content-delivery mechanism but society’s highest-tolerance environment for intelligent failure. A university has advantages no startup or corporate labs that easily replicates resident experts, shared instruments, a continuous supply of students, multidisciplinary knowledge under one roof, low cost, freedom to explore, and fewer immediate commercial pressures. Under this framing, its role shifts from transmitting knowledge to orchestrating the same three teachers at a higher level- becoming, in effect, society’s discovery engine, not merely its certification engine.

Part VI — Human Development: The Layer That Matters Most

If the knowledge and zone architecture answers what and how, this layer answers for whom, and why they’d bother. It is built on the observation that the most underserved population in any system organized around test scores is the group for whom identity, not intelligence, is the bottleneck.

Its mechanisms, consistent across every version of the architecture:

• Structured self-discovery — every student names an explorer archetype and answers, explicitly: What am I good at? What excites me? What gives me energy? What kind of explorer am I?

• A First Victory Programme — one visible success, one contribution, one public presentation, guaranteed within the first month.

• Life Quests — experiential challenges alongside academic work: teaching a younger child something, growing food for a month, interviewing a local entrepreneur, volunteering, organizing an event. The objective is agency, not knowledge.

• An Explorer Index — a profile tracking curiosity, persistence, initiative, creativity, collaboration, courage, adaptability, and contribution, alongside academic grades, because these are hypothesized to predict long-term flourishing more reliably than exam scores alone. This is an important caveat worth stating plainly: such a measure is a design hypothesis to be piloted and validated, not a settled predictive instrument — any serious version of this architecture should say so explicitly rather than assert it as proven.

• A three-level Mentor Corps — near-peers (18–25) who prove that change is possible within a single generation and are often the most influential figures for disengaged students; working professionals across many fields who demonstrate the full breadth of paths beyond entrepreneurship; and master mentors, reserved for work that has reached genuine seriousness and expected to challenge rather than merely encourage.

This layer exists because a system that only reaches the top 10–20% — the students for whom education was already the bottleneck — has recreated exactly the inequality it claims to fix.

Part VII — The AI Engine: Invisible, Indispensable

AI’s role throughout this architecture is deliberately understated: it is infrastructure, not a substitute teacher. Its functions unified across the source frameworks:

Function

Auditor

Detects hidden traps in problems; classify student and machine error patterns

Generator

Creates new challenges and scenarios personalized to individual weak points

Coach

Suggests the next best step; adapts difficulty in real time

Connector

Links concepts across subjects that would otherwise stay siloed

Nature Bridge

Surfaces the natural-world analogue for every concept

Explorer Guide

Tracks each student’s archetype and development stage; assigns quests; surfaces mentors at the right moment

Students are taught a simple discipline for approaching any problem, wherever the traps happen to be — in an exam question or in a machine’s answer: strip away the trick before touching the formula, name the trick once found, and verify any AI-generated answer by explaining it in plain steps. A student who learns to decode an exam trick today is the adult who audits an AI mistake tomorrow. AI shortens the path to interesting questions; it does not replace the judgment required to answer them.