Twenty minute talk / A speculative curriculum

Five Speculative Courses for a World with Decentralized Intelligence

Four courses come from the published talk. I am adding one more because projects now change while we are building them.

01 / 35

Opening, 35 sec. Start with motion, not explanation. The point is not that universities are obsolete. The point is that the old reason for many programs is weaker when analysis, drafting, and first-pass research become cheap.

Intelligence report / cold read

Five minutes into the future

Status board
27 May 2026

01 / Global use

Consumer AI is mass market.

Large assistants are used for search, writing, coding, translation, tutoring, planning, and image work. Weekly accounts across major tools likely exceed one billion; unique people cannot be measured cleanly.

02 / Organizations

Adoption is broad; redesign is uneven.

Surveys show most organizations now use AI in at least one function. Many teams still run pilots, copilots, and isolated workflows instead of changing reporting lines, budgets, risk review, or service design.

03 / Europe

Europe is moving, but slower.

EU enterprise AI use was 13.5% in 2024 by Eurostat's measure for firms with 10+ employees. A 2026 working estimate is 20-25%, led by large firms and Nordic/Benelux markets.

04 / Data conflicts

Training data is contested.

Publishers, platforms, artists, universities, states, and companies are fighting over scraping, private records, synthetic data, attribution, data residency, and removal rights.

05 / Regulation

Rules are becoming operational.

The EU AI Act is shifting from text to enforcement. General-purpose AI, transparency, prohibited practices, AI literacy, standards, and high-risk systems are now practical compliance problems.

06 / Pressure points

The next fight is control.

Open models versus closed services. Compute and chips. Energy and water. Export controls. Workplace monitoring. School assessment. Procurement. Liability when an AI-assisted decision fails.

Assessment: use is ahead of governance; governance is ahead of measurement; curricula are behind both. Confidence: medium / figures are mixed by source and definition.

02 / 35

Cold read, 60 sec. Read this like a briefing, not like a keynote. The point is a sober present: AI use is already widespread, institutional change is uneven, Europe is moving more cautiously, and the conflicts are mostly about data, control, responsibility, and rules becoming operational.

Source: Anthropic Economic Index / Anthropic

02 / 35

Graph, 45 sec. Let the image do the work. The useful question is what higher education trains when AI is already inside professional tasks.

The old bargain

University trained the person who could find, explain, and produce.

Find.Gather material that was hard to access.

Explain.Turn scattered material into a clear argument.

Produce.Write the paper, plan, model, or deck.

02 / 35

Context, 45 sec. Keep this simple. The university used to create value by giving people tools for scarce knowledge work. AI does not erase that, but it changes where the hard part is.

If answers are cheap, the program has to train better questions, better choices, and better responsibility.

03 / 35

Thesis, 40 sec. This replaces broad claims with a practical test. What would a course train that is still hard when students have strong AI from day one?

The artifact

Think of this as a real 30 ECTS studio semester.

Five six-credit courses. Twenty-four students. Seminars, labs, fieldwork, and a public review at the end.

Level

MA / MSc
/ MDes

For design, management, policy, HCI, and digital humanities students.

Format

Studio
+ seminar

Three contact days per week, plus field work and partner work.

Credits

30
ECTS

Five courses at 6 ECTS each, taught as one integrated block.

Assessment

Public
work

Briefs, protocols, prototypes, and working plans, not only essays.

04 / 35

Artifact frame, 60 sec. Make it feel administratively plausible. This is not a manifesto slide. It is a proposed curriculum that could sit in a course catalogue.

Four plus one

Original course 1

Systems
Governance

Who sets rules for systems no one fully sees?

Added course

Planning on
a Moving Substrate

How do projects survive when the platform changes mid-build?

Original course 2

Decision
Architecture

How do groups choose when analysis is endless?

Original course 3

Meaning
Design

How do people make shared sense when content is cheap?

Original course 4

Seeing
& Foreseeing

How do we notice what is not yet obvious?

Why this order

Rules, projects,
choices, meaning, futures.

First stabilize the ground. Then decide what to build.

05 / 35

Map, 60 sec. Name the change clearly. The published talk has four. This version adds Planning on a Moving Substrate, placed second because project work is where many people will first feel the problem.

I / Systems Governance

Course title: The Sovereignty Problem. 6 ECTS. Six weeks.

06 / 35

Transition, 20 sec. Open the first course with the practical question: who has authority when the system crosses vendors, departments, cities, and bodies?

I / Why it exists

Someone must set limits before the system causes harm.

The hard part is not writing a principle. It is deciding who can pause, audit, change, or shut down a live system.

Week 1

Make the system visible.

Map actors, vendors, users, data flows, and hidden decision points.

Week 2

Draw boundaries.

Decide what counts as inside the system and what counts as outside.

Week 3

Assign responsibility.

Build an accountability map for shared failures.

Week 4

Design consent.

Explain what people can agree to when they cannot read the model.

07 / 35

Course I detail, 90 sec. Avoid abstract governance language. Say: students learn to draw the map, name who can act, and design a process that works before an incident becomes a scandal.

I / What students make

Case

Public AI service

A city benefit tool, school tool, transit tool, or health triage tool.

Main artifact

Governance brief

Rules, pause points, review cadence, and a sunset clause.

Evidence

Risk walk-through

Show how one failure would move through the system.

Review

External reader

A policy, legal, or procurement partner responds to the brief.

08 / 35

Deliverable, 60 sec. Make the output feel real. This course ends with a document a municipality or company could actually review, not a generic ethics essay.

I / Systems Governance6 ECTS / course plan

LECTURE 01 / seeing the system Exercise: map one public AI service as actors, data, vendors, decisions, and affected people. Reading: Elinor Ostrom, rules-in-use; Langdon Winner, "Do Artifacts Have Politics?" LECTURE 02 / where authority sits Exercise: mark who can approve, pause, audit, repair, or shut down each part of the service. Reading: Julie Cohen, Between Truth and Power; Sheila Jasanoff, Technologies of Humility. LECTURE 03 / harm before scandal Exercise: write a failure story from first complaint to public consequence. Reading: Ruha Benjamin, Race After Technology; Cathy O'Neil, Weapons of Math Destruction. LECTURE 04 / consent and explanation Exercise: design a consent and appeal moment for a person who cannot inspect the model. Reading: Helen Nissenbaum, contextual integrity; GDPR guidance on automated decisions. LECTURE 05 / governance brief studio Exercise: draft rules, review cadence, pause points, and a sunset clause. Reading: Model cards and system cards from major AI labs. LECTURE 06 / external review Exercise: defend the brief to a policy, legal, procurement, or community reviewer. Final artifact: governance brief, risk walk-through, and revision log.

09 / 35

Course plan, 45 sec. This is the administrative proof: weekly lectures, exercises, and readings. It should feel like a course a student could enroll in.

Who can stop it?

09 / 35

Image beat, 20 sec. Let the question sit. It is simple, and it is the whole course.

II / Planning on a Moving Substrate

Course title: Planning on a Moving Substrate. 6 ECTS. Eight weeks.

10 / 35

Transition, 20 sec. This is the added course. Put it second because every organization building with AI now faces this: the thing they build on keeps changing.

II / Why it exists

The platform may improve faster than the project can ship.

By launch, the model may do natively what the team spent months building around.

Old planning

Requirements change.

Agile handles this reasonably well.

New planning

Capabilities change.

The foundation itself moves under the project.

Bad response

Chase everything.

The project never stabilizes.

Better response

Plan review points.

Decide when to freeze, defer, upgrade, or delete.

11 / 35

Course II setup, 90 sec. Use a concrete example: a team scopes a feature around one model limit. Two months later, a new model removes the limit. The team needs a way to respond without panic.

II / Tools students learn

Week 1

Assumption half-life

Estimate when each major technical assumption may expire.

Week 2

Capability horizon

Track what the platform may do by prototype, pilot, and launch.

Week 3

Reversibility score

Separate decisions that can change later from decisions that lock the team in.

Week 4

Obsolescence ledger

Keep a list of work that may be deleted when the platform improves.

12 / 35

Frameworks, 75 sec. These are invented but plausible course tools. The value is not prediction accuracy. The value is making platform change visible inside project management.

II / Studio work

Students redesign a project plan after the ground shifts.

They are tested with three capability jumps: cheaper inference, native multimodal output, and a better vendor.

Deliverable

Living project
plan

Milestones, review gates, and decision rules.

Extra artifact

Client
memo

Explain why scope changed without sounding like failure.

Assessment

Stress
test

Can the plan absorb change without becoming vague?

ECTS load

160
hours

Eight weeks, studio plus technology watch.

13 / 35

Studio, 75 sec. Keep this practical. The course trains product managers, designers, strategists, and technical leads to plan around moving capability, not pretend the platform is stable.

II / Planning on a Moving Substrate6 ECTS / course plan

LECTURE 01 / projects built on changing ground Exercise: list the technical assumptions behind an AI product and estimate their half-life. Reading: Donald Schoen, reflective practice; Rita McGrath, discovery-driven planning. LECTURE 02 / capability watch Exercise: maintain a weekly log of model, price, latency, and policy changes that affect the brief. Reading: Clayton Christensen on disruption; Amara's law as a planning caution. LECTURE 03 / freeze points Exercise: decide what must be stable by prototype, pilot, procurement, and launch. Reading: Barry Boehm, spiral model; Ward Cunningham, technical debt. LECTURE 04 / reversible and irreversible choices Exercise: score architecture, vendor, interface, and data choices by reversibility. Reading: Jeff Bezos shareholder letters on one-way and two-way doors; Nassim Taleb, optionality. LECTURE 05 / deleting work Exercise: build an obsolescence ledger: what should be removed if the platform improves. Reading: Marty Cagan, product risk; Amy Edmondson, intelligent failure. LECTURE 06 / client memo Exercise: explain a scope change as good project judgment, not panic. Final artifact: living project plan, review gates, capability horizon, and client memo.

14 / 35

Course plan, 45 sec. This course makes volatility teachable. Students learn to name what changed and decide what to freeze, upgrade, or delete.

Freeze, upgrade, or delete?

14 / 35

Image beat, 20 sec. This is the decision rhythm of the added course.

III / Decision Architecture

Course title: The Abundance Trap. 6 ECTS. Six weeks.

15 / 35

Transition, 20 sec. Now move from project planning to group choice. The problem is not lack of analysis. It is how a group chooses with too much analysis.

III / Why it exists

A room can have enough evidence and still avoid a decision.

AI makes every option easier to defend. That can make commitment harder.

Week 1

Recommendation layer

Study how options are pre-selected before people notice.

Week 2

Shared picture

Help groups agree what problem they are deciding.

Week 3

Decision meeting

Design roles, timing, objections, and closure.

Week 4

Uncertainty

Show confidence without false precision.

16 / 35

Decision detail, 90 sec. Use familiar workplace language. This course is for the meeting after everyone has a strong AI-generated memo and nobody wants to own the tradeoff.

III / What students make

Case

Broken decision

A delayed hiring process, product bet, budget cut, or policy choice.

Audit

Where it failed

Map missing facts, hidden fears, incentives, and unclear authority.

Prototype

New process

Run a better decision format with real participants.

Measure

Outcome delta

Compare time, clarity, confidence, and commitment.

17 / 35

Deliverable, 60 sec. Make the assessment concrete. The student does not just describe decision theory. They fix a decision process and test whether it works better.

III / Decision Architecture6 ECTS / course plan

LECTURE 01 / the abundance trap Exercise: compare three AI-generated recommendations and identify how each frames the decision. Reading: Herbert Simon, bounded rationality; Daniel Kahneman, Thinking, Fast and Slow. LECTURE 02 / who owns the choice Exercise: map decision rights, vetoes, advice roles, and hidden incentives in a real meeting. Reading: Richard Thaler and Cass Sunstein, Nudge; Annie Duke, Thinking in Bets. LECTURE 03 / making uncertainty usable Exercise: rewrite a confident memo as ranges, scenarios, thresholds, and open questions. Reading: Paul Slovic on risk perception; Tetlock and Gardner, Superforecasting. LECTURE 04 / meeting design Exercise: run a decision meeting with roles for advocate, skeptic, affected user, and closer. Reading: Priya Parker, The Art of Gathering; Roger Martin, integrative thinking. LECTURE 05 / measuring commitment Exercise: compare the old and new decision process by time, clarity, confidence, and follow-through. Reading: Peter Drucker on effective decisions; Gary Klein, Sources of Power. LECTURE 06 / decision audit Exercise: present the process, the tradeoff, the objection record, and the chosen next step. Final artifact: decision audit, redesigned meeting format, and outcome delta.

18 / 35

Course plan, 45 sec. This is where AI makes a familiar problem sharper: every option can be defended, so the course trains closure with responsibility.

Rooms that choose

18 / 35

Image beat, 20 sec. This course is about designing the room so it can move.

IV / Meaning Design

Course title: Narrative Infrastructure. 6 ECTS. Six weeks.

19 / 35

Transition, 20 sec. Now move to meaning. Avoid making this sound mystical. It is about how groups decide what words, signals, and rituals they trust.

IV / Why it exists

People still need a shared world to act together.

When every statement can be generated, trust has to be designed through practice, proof, and repetition.

Week 1

Meaning systems

How communities decide what counts as true enough.

Week 2

Ritual and rhythm

Meetings, releases, ceremonies, and repeated signals.

Week 3

Synthetic media

Design for audiences who doubt provenance.

Week 4

Repair work

What restores trust after an institution fails.

20 / 35

Meaning detail, 90 sec. Keep it grounded in organizations and communities. Meaning design is not making content. It is making the conditions in which people can interpret and trust together.

IV / Dialogue after AI

The machine should not end the conversation. It should improve it.

A Socratic process uses AI for questions, counterexamples, reframing, and pressure testing.

Question

What do we mean?

Clarify terms before optimizing outputs.

Probe

Why this?

Expose assumptions, values, and omissions.

Test

What resists?

Use the machine to generate pressure, not closure.

Name

What holds?

Turn the result into language people can share.

21 / 35

Socratic process, 90 sec. This keeps the user's requested idea. After AI, knowledge work becomes dialogical because the first answer is easy. The harder skill is staying with the question long enough to make shared meaning.

IV / What students make

Case

Community under strain

A school, brand, neighborhood, newsroom, or team with low trust.

Artifact

Meaning system

A ritual, editorial protocol, interface, or public practice.

Test

Four weeks live

Run it with a real group, not only in a classroom.

Assessment

Trust signals

Look for participation, clarity, reuse, and repair.

22 / 35

Deliverable, 60 sec. Again, make it real. The final work is not a poster about trust. It is a small system that tries to make trust easier to practice.

IV / Meaning Design6 ECTS / course plan

LECTURE 01 / when content is cheap Exercise: collect five messages from one institution and ask what makes each believable or empty. Reading: Clifford Geertz, thick description; Charles Taylor, social imaginaries. LECTURE 02 / dialogue as method Exercise: run a Socratic dialogue with AI where the student must record claims, doubts, and revisions. Reading: Plato, Meno excerpts; Paulo Freire, Pedagogy of the Oppressed. LECTURE 03 / rituals and repeated signals Exercise: design one repeated practice that helps a group know what matters. Reading: Erving Goffman, interaction ritual; Victor Turner, ritual process. LECTURE 04 / provenance and trust Exercise: redesign a publication or internal memo for audiences who assume text may be synthetic. Reading: danah boyd on context collapse; Kate Crawford, Atlas of AI. LECTURE 05 / repair after failure Exercise: write a public repair protocol for a mistake, rumor, or broken promise. Reading: Annette Baier on trust; Onora O'Neill, trustworthiness. LECTURE 06 / live meaning system Exercise: run a ritual, editorial protocol, interface, or public practice with a real group. Final artifact: meaning system, dialogue record, live test, and trust signal report.

23 / 35

Course plan, 45 sec. Keep the Socratic point clear: the teacher can evaluate the visible path of meaning-making, not just the final text.

Meaning is made between us

23 / 35

Image beat, 20 sec. Let this sentence carry the course.

V / Seeing & Foreseeing

Course title: Speculative Perception. 6 ECTS. Six weeks.

24 / 35

Transition, 20 sec. The last course is about imagination, but keep the wording simple: it trains people to notice early signals and make futures discussable.

V / Why it exists

The model learns from what culture has already recorded.

People still meet the raw material first: streets, habits, awkward changes, new gestures, and weak signals.

Week 1

Weak signals

Read the marginal, strange, local, and not-yet-named.

Week 2

Trained eye

Use photography, field notes, and visual analysis as method.

Week 3

Machine futures

Study what generative tools miss, repeat, and reveal.

Week 4

Make an argument

Use an object or scene to make a future discussable.

25 / 35

Seeing detail, 90 sec. This draws from the imagination talk. AI can multiply images, but the human brings raw contact with the world before the pattern is obvious.

V / Machine-human rhythm

Notice. Ask. Make. Judge. Return to the world.

The rhythm is dialogical. The machine expands the field; the person decides what is alive.

1 / Notice

Bring material.

A real signal from fieldwork, not only a dataset.

2 / Ask

Multiply views.

Use AI to create variants, tensions, and counterexamples.

3 / Judge

Choose the live one.

Use taste, ethics, context, and consequence.

4 / Return

Test in the world.

See whether anyone recognizes the possibility.

26 / 35

Rhythm, 75 sec. Keep this as the course's practical method. The machine is neither replacement nor oracle. It is part of a loop.

V / What students make

Fieldwork

Signal
journal

Forty observed signals with photos, notes, and source context.

Studio

Future
artifact

A designed object, service scene, media piece, or space.

Review

Argument
test

Does the artifact make people argue about a real possibility?

Assessment

Not
polish

Graded on the quality of the future it makes visible.

27 / 35

Deliverable, 60 sec. The final artifact should not be a slick concept deck. It should change what an audience thinks is possible, even a little.

V / Seeing and Foreseeing6 ECTS / course plan

LECTURE 01 / weak signals are not trends yet Exercise: collect ten local signals with photos, notes, dates, and why they feel early. Reading: Ansoff on weak signals; Jan Chipchase, field research practice. LECTURE 02 / the trained eye Exercise: compare what a human notices in the street with what a model names in the image. Reading: John Berger, Ways of Seeing; James Gibson, ecological perception. LECTURE 03 / machine imagination Exercise: use generative tools to multiply interpretations, then mark what feels repeated or dead. Reading: Italo Calvino, Six Memos; Brian Eno on oblique strategies. LECTURE 04 / making futures discussable Exercise: turn one signal into an object, scene, service moment, or small media artifact. Reading: Dunne and Raby, Speculative Everything; Stuart Candy, experiential futures. LECTURE 05 / taste, judgment, and return Exercise: test the artifact with people who do not share the student's assumptions. Reading: Pierre Bourdieu, Distinction excerpts; Lucy Suchman, situated action. LECTURE 06 / argument test Exercise: present the artifact and record what future it makes people argue about. Final artifact: signal journal, future artifact, machine-human process log, and argument test.

28 / 35

Course plan, 45 sec. This course makes imagination operational: the student brings material from the world, uses AI as a partner, and returns to people for judgment.

Wild data

28 / 35

Image beat, 20 sec. This connects back to the imagination talk: the material starts outside the model.

The five courses together

This is a semester about acting well when intelligence is everywhere.

GovernanceWho may decide?

PlanningWhat may change?

DecisionHow do we choose?

MeaningWhat can we trust?

SeeingWhat is emerging?

29 / 35

Synthesis, 60 sec. Keep this very plain. These are five practical questions. They are more useful than five grand nouns.

Program calendar

Weeks 1-2

Map
the system.

Students pick partner cases and build shared maps.

Weeks 3-4

Build
tools.

Each course introduces one working framework.

Weeks 5-6

Test
outside.

Students run pilots, reviews, or field tests.

Week 7-8

Public
review.

Planning projects get extra stress tests; all work is shown.

30 / 35

Realism, 45 sec. This makes the curriculum feel operational. It could run as an intensive studio semester with shared cases and separate course lenses.

Assessment

Students pass by making useful things under real constraints.

Not every course needs an exam. The work is judged by whether it can survive contact with a partner, a group, or a public.

30%

Field evidence

Interviews, observations, logs, and system maps.

30%

Working artifact

Protocol, plan, process, meaning system, or future artifact.

20%

Public review

Response from people outside the classroom.

20%

Reflection

What changed, what failed, and what should be revised.

31 / 35

Assessment, 45 sec. This keeps the program from becoming speculative theater. Students must produce things that can be criticized by real stakeholders.

The point is not to teach people to compete with AI at producing answers.

32 / 35

Close thesis, 30 sec. Let the sentence breathe. This is the negative claim.

The point is to teach people to set limits, plan under change, choose, make meaning, and see what is coming.

33 / 35

Close thesis, 45 sec. This is the positive claim. It restates the whole program in simple verbs.

One open question

And how does academic work look like?

We do not know yet. But a Socratic dialogue seems to be one way.

Process matters.The teacher can evaluate the transparent path to a conclusion, not only the final assignment.

The teacher joins.They can enter the dialogue and ask the student to reflect on their own creation.

The record stays visible.The work includes prompts, revisions, choices, doubts, and reasons.

A typed Socratic dialogue with AI on paper

Academic work after AIDialogue as process

34 / 35

Academic work, 75 sec. This slide opens the question rather than settling it. The likely shift is that teachers assess the reasoning trail: what the student asked, how they tested answers, what they rejected, and how they responded to dialogue. The final assignment still matters, but it is no longer the only evidence of learning.

Close

Name

Sérgio Tavares

Role

PhD | Strategy and design

Web

sergiotavares.com

LinkedIn (QR)

35 / 35

Close, 30 sec. Leave this slide up for questions.