A Subconference of the World Design Cities Conference

Designing AI's
Boundaries:
Respecting Human Agency

Shanghai · September 26–29, 2026

Venue
SUES & Tongji University
Format
Dialogue · No Papers
Chair
Michael Lissack

Is There Too Much AI?

Most conferences are occasions for talking: papers are delivered, panels perform, an audience listens, and everyone leaves with a tote bag and their opinions intact. This event is built on the opposite premise. There are no papers and no spectators — every person in the room (and every person joining remotely) is a working participant in a sustained, structured dialogue, supported by AI tools that handle the retrieval and synthesis so the humans can do what only humans can: question, judge, and decide. And the work aims at something usable. Over three days, each participant builds a concrete takeaway — an action plan for designing, in their own domain, the balance among three kinds of reliance: when to rely on AI, when to rely on human perception and judgment, and when to rely on the context and information particular to the situation at hand. The difference is what you leave with.

Every discipline now confronts the same recursive problem: AI systems are redesigning the people who are supposed to govern them. This conference refuses to treat that as a technical puzzle with a technical solution. Instead, we ask a design question — how do we draw boundaries that respect what only humans can do?

Hosted within the World Design Cities Conference, this three-day subconference brings together complexity scientists, philosophers of mind, AI researchers, educators, and a playwright to hold that question in dialogue — through Buberian dialogues, keynote exchanges, and AI-augmented working groups designed to produce insight that no single discipline could reach alone.

The intellectual engine is the concept of the AI Dramaturg: the recognition that every discipline has a dramaturgical function — the framing, staging, and interpretation of its own questions — that AI can partially supply but cannot replace. The irreducible remainder is situated human judgment.

Beneath that boundary lies a question about reasoning itself — about abduction, the act of deciding which explanation is worth reaching for. A model surveys an enormous field at once and weighs relevance in a single currency, learned similarity across everything in view: comprehensive, but flat. A person works from a small foreground, steered by something the model lacks — affect, bodily salience, the trained feel for what matters that can reframe a problem on a basis no text contains. A model can make the products of that judgment legible; it cannot perform the judgment. Drawing that line is the work of these three days.

DesignX

Complex, sociotechnical problems that resist traditional design methods. AI's relationship to human agency is a DesignX problem par excellence — the observer is part of the system being designed.

Second-Order Science

The presuppositions governing how a field knows what it knows. We surface them, question them, and ask what changes when AI enters the epistemic loop.

The 60/40 Split

AI can supply roughly 60% of what any discipline's dramaturgy requires. The 40% it cannot — the situated judgment, the ethical framing, the interpretive act — is the subject of this conference.

Habitus & the Tacit

A field's habitus — its durable, embodied dispositions — and its tacit know-how decide what a practitioner treats as the obvious move. A corpus-trained model can surface the sediment they leave behind, not the disposition that produced it. What it flattens is what we mean to name.

The AI Dramaturg

Every discipline has a dramaturgical function — the framing, staging, and interpretation of its own questions. AI can partially supply this function but cannot replace it. The remainder is what makes us human.

In theater, the dramaturg is the person who holds the intellectual architecture of a production — researching context, questioning choices, ensuring coherence. The dramaturg doesn't direct, doesn't act, doesn't write. The dramaturg holds the space in which meaning is made.

AI is becoming every discipline's dramaturg. It can retrieve, synthesize, pattern-match, and propose — reasoning, in effect, over a learned surrogate of human practice. But it cannot make the situated judgment that transforms information into understanding. This conference asks: where exactly is that boundary, and how do we design systems that respect it?

60%
AI Can Supply
40%
Requires Human Judgment
3
Curated Research Corpora

The 26th opens with filmed provocations, Harry Collins's keynote, and a reciprocal Buberian trialogue; the DesignX working sessions then run across the 26th and 27th, building the takeaway with GraspingAI, and on the 28th the conversation reaches the WDCC main stage on the Bund, where David Snowden presents and then joins Lorenzo Magnani in a single Buberian dialogue whose two strands — abduction beyond the machine, and sensing before solving — interweave before the full WDCC audience.

Keynote

Harry Collins

Distinguished Research Professor, Cardiff University

For half a century, Harry Collins has been the definitive voice on what knowledge resists being made explicit. His work on tacit knowledge, expertise, and the sociology of scientific knowledge is the ground on which this conference's central question stands: what can a machine be told, and what can it never be told? His keynote opens the first day.

Main Stage · 28 September

David Snowden

Founder & Chief Scientific Officer, The Cynefin Company

Creator of the Cynefin framework and the founder of naturalistic sense-making, David Snowden has spent decades mapping where ordered approaches to knowledge end and complexity begins. On 28 September he presents on the WDCC main stage at the Convention Center Riverside, carrying the subconference's question — what must remain situated human judgment — to the full WDCC audience on the Bund.

Main Stage Dialogue · 28 September

Lorenzo Magnani

Professor of Philosophy of Science, University of Pavia

Lorenzo Magnani returned abduction — Peirce's logic of forming explanatory guesses — to the world in which it actually happens. His eco-cognitive model shows that abductive reasoning is not a formal calculus but an activity distributed across bodies, instruments, and environments: we think through doing, and our guesses carry the weight of lived consequence. That insight anchors the first of the working groups' six cases, and on 28 September he brings it to the WDCC main stage in a Buberian dialogue with David Snowden — two interweaving strands on whether abduction can, or should, be formalized computationally. He also serves as a learned listener in the opening day's trialogue.

Video Provocations
Markus Buehler
MIT
Sherry Turkle
MIT
Don Norman
UC San Diego · Design Lab
Cameron Tonkinwise
UTS Sydney
Ken Friedman
Tongji University · She Ji
Lucy Suchman
Lancaster University
Joanna Bryson
Hertie School, Berlin
Thomas Powers
University of Delaware
Buberian Trialogue — Protagonists
Jonathan Boymal
RMIT University
Owen Matson
Feng Tao
Nankai University
Buberian Trialogue — Learned Listeners
Ramón Alvarado
University of Oregon
Lorenzo Magnani
University of Pavia
Ricardo Baeza-Yates
Northeastern University · UPF Barcelona
Working Sessions
Hugo Letiche
ISTEC Paris
Joel Pearson
UNSW · Future Minds Lab
Fabrizio Degni
Chief of AI · AI ethics & governance
Thomas Biedermann
Senior Lecturer · games & systems
Marco Palombi
Entrepreneur · human-centric platforms
John Hawkins
Chief AI Officer · ML researcher
Aaron Kagan
GraspingAI
Dan Zhu
Design PhD · Generative-AI Artist
Muhs & Stankowski
Design science · DFKI
Scholars & Participants
Jacqueline Sullivan
Rotman Institute · Western University
Mina Ekramnia
Max Planck · Harvard Global Health
Elizabeth Ricker
Author, Smarter Tomorrow
Mythili Kolluru
College of Banking & Financial Studies, Muscat

Invited

Invitations are extended to the scholars, designers, and artists below; the roster will firm as acceptances arrive.

Keynote
Andy Clark
University of Sussex
Yann LeCun
Meta · NYU
Ethan Mollick
Wharton, U. Pennsylvania
Conor Grennan
NYU Stern
Arvind Narayanan
Princeton University
Vincent Blok
Erasmus University Rotterdam
Luciano Floridi
Yale University · Digital Ethics Center
Bernard Baars
Florida Atlantic University · Center for the Future Mind
Tacit Knowledge & the Legible
Erik Rietveld
University of Amsterdam
Rachel Ankeny
Wageningen University
Johan Heilbron
CNRS–EHESS, Paris
Joseph Rouse
Wesleyan University
Tor Nørretranders
Copenhagen
Philosophy of AI & Mind
Cameron Buckner
University of Houston
Sara Aronowitz
University of Toronto
Lisa Messeri
Yale University
Susan Schneider
Florida Atlantic University
Anil Seth
University of Sussex
Nicholas Shea
Institute of Philosophy, London
Been Kim
Google DeepMind
Roman Yampolskiy
University of Louisville
Jörg Noller
LMU Munich
Tony Seale
Knowledge Graph Guys
Design — International
Kees Dorst
TU Eindhoven · UTS Sydney
Ezio Manzini
Politecnico di Milano
Birger Sevaldson
Oslo School of Architecture & Design
Naoto Fukasawa
Tokyo
Wolfgang Jonas
Design researcher, Germany
Kun-Pyo Lee
KAIST
Additional Scholars
N. Katherine Hayles
UCLA · Duke University
Pascale Fung
HKUST
Yi Zeng
Chinese Academy of Sciences
Huang Minlie
Tsinghua University
Shen Weixing
Tsinghua University
Yu Zhang
Jilin University
Dramatic Arts
Stan Lai
Playwright & Director, Taipei

Program Overview

Download the program overview (PDF) →

1
September 26 · Saturday · Tongji or SUES

Opening Dialogues & DesignX

Opening
Video Provocations
Buehler · Turkle — the machine as partner set against what erodes when it stands in; two provocations that stake the day's poles
Midday
Keynote
Harry Collins on tacit and explicit knowledge — what machines cannot be told
Afternoon
Buberian Trialogue
Boymal · Matson · Feng Tao and Alvarado · Magnani · Baeza-Yates — two trios who trade places partway through, each serving in turn as protagonists and as learned listeners who describe what they heard
Late Afternoon
DesignX: Framing the Orthogonal Space
The working sessions open — Lissack frames the problem: computational capability and situated human judgment as independent axes, not one scale, and the groups form
2
September 27 · Sunday · SUES or Tongji

DesignX Working Sessions

Opening
Video Provocations
Norman · Suchman · Bryson — designing for agency, situated action that escapes the plan, and the governance the process itself demands; provocations to open the working sessions
Morning
Working Groups I — with GraspingAI
Tentatively confirmed participants — Pearson, Degni, Biedermann, Palombi — and attendees map the space between AI and human thought, working from six shortcut cases: six practices where AI stands in for a human capacity
Afternoon
Working Groups II — the Journey Method
Building a way to plot a route through that space fitted to a given user's context
Late Afternoon
Consolidation
The groups bring their maps and journey drafts together, readying the takeaway carried into the 29th
3
September 28 · Monday · WDCC Main Stage, the Bund

On the Main Stage

Morning
WDCC Opening
The WDCC plenary addresses — not part of this subconference's program
Afternoon
Main Stage Address — David Snowden
The creator of Cynefin on sense-making at the boundary between computational reach and situated judgment — the subconference's question put to the full WDCC audience
Afternoon
Buberian Dialogue — Two Strands, One Encounter
One Buberian dialogue on the main stage, opening with video reflections drawn from the prior two days, in which two strands interweave — and the audience's role is to find the commonalities and usable takeaways as they do
“Abduction Beyond the Machine: The Eco-Cognitive Model and the Limits of Computational Inference” — Lorenzo Magnani (University of Pavia), in dialogue with David Snowden. Whether abductive reasoning — as scientific discovery, model-based cognition, and distributed sense-making — can or should be formalized computationally, and what is lost or gained in the attempt
“Sensing Before Solving: Complexity, Constraint, and the Case Against Premature Formalization” — David Snowden (Cynefin Centre), in dialogue with Lorenzo Magnani. Why complex, unordered domains resist the pre-specified models that computation and best-practice both depend on — and what this demands of human sense-making before, and instead of, formal inference
4
September 29 · Tuesday · Tongji or SUES

The Takeaway

Opening
Video Provocations
Friedman · Tonkinwise — what the work owes as knowledge set against what design should refuse; provocations before the map and the method
Morning
The Orthogonal Map
The working sessions consolidated into one shared picture: computational capability and situated human judgment drawn as independent axes rather than a single scale from worse to better — a space you can point at, where a model's reckoning is strong, where it is empty, and where the two must meet
Late Morning
The Journey Method
A repeatable way to plot a route through that space for a given context — which work to hand to the model's calculative reach, and which to hold back as embodied, situated judgment; the roughly 60/40 line drawn not in the abstract but for a real case brought into the room
Midday
Carrying It Forward
Each participant leaves having applied the map and the method once to their own domain — the takeaway is that instrument, a way of seeing and a way of deciding, not a proceedings volume

Six Shortcuts, Six Capacities at Risk

Every stage of inquiry now has its AI shortcut. Hypotheses arrive pre-formed. Meaning arrives pre-structured. Users arrive pre-simulated. Artifacts arrive pre-built. Judgments arrive pre-rendered. Decisions arrive pre-optimized. None of these shortcuts is the work of fools.

Each is the product of competent, well-intentioned practice — and each rests on presuppositions its practitioners rarely stop to examine. On the 27th, the working groups take six such practices seriously enough to question them, asking the same three questions of each: What is being uncritically presupposed? What human capacity is quietly being placed at risk? And what boundary would protect it? Together the six trace the full arc of knowledge-making — from generating a hypothesis to deciding what to do.

1 · Generating

The Hypothesis That No One Abduced

When a model proposes the hypothesis, the researcher receives something hypothesis-shaped — but has performed no abduction at all. Human abduction works from a limited base: cognitive constraint forces selection, and what we attend to carries valence earned through lived experience — the anomaly that nags, the guess that once failed. The constraint is the engine, not a defect. Ask a machine to abduct and both features vanish: nothing constrains attention, nothing is at stake, and valence is assigned by repetition across a corpus. Salience without a life behind it — relevance simulated by frequency.

Presupposed That a hypothesis is a text object rather than the residue of an inferential act; that frequency across a corpus is a workable proxy for what matters.
At risk Abductive competence — and the experience-earned sense of which anomaly is worth chasing.
The boundary question What must stay un-delegated for scientists and designers to keep learning to guess well?
2 · Representing

The Ontology That Closed the World

Knowledge graphs and enterprise ontologies promise machine-readable meaning: entities fixed, relations typed, semantics settled. The promise is real — and so are its buried commitments, including an unexamined circularity now that AI builds the ontologies that then constrain the AI.

Presupposed That meaning can be exhaustively represented; that identities remain stable across contexts; that the semantic world can be closed.
At risk The negotiability of meaning — re-categorizing, holding ambiguity open, noticing that the map's categories were choices.
The boundary question Where must formalization stop so that re-description remains possible?
3 · Encountering

The User Who Was Never There

Synthetic user research replaces interviews, observation, and ethnography with simulated personas — built from the very expectations the designer brought to them. A simulated user can never resist you, contradict you, or reconfigure your understanding of the problem.

Presupposed That empathy is information transfer; that the other is a distribution; that surprise is noise rather than the most valuable signal in design research.
At risk The capacity for encounter — being changed by an actual other — and with it the ethical core of human-centered design.
The boundary question Which moments of contact with real people must design refuse to simulate?
4 · Making

The Artifact Its Maker Never Touched

Prompt-to-product workflows deliver functioning artifacts whose makers never engaged the material. Yet craft has always fed theory — geometry itself grew from the practices of masons, carpenters, and surveyors — and friction in making, including the friction of debugging, is where tacit understanding forms. Experts who already carry such models wield generative tools powerfully; novices who begin with the tools may never build the models at all.

Presupposed That the artifact is the knowledge; that friction in making is overhead rather than the site of understanding.
At risk Knowing-through-making — the generative channel by which craft feeds theory.
The boundary question What must still be made by hand — and by whom — for making to keep teaching?
5 · Evaluating

The Judgment Nobody Exercised

Journals, funders, schools, and firms are adopting AI screening and review — treating quality judgment as the application of criteria. But connoisseurship is tacit: it lives in trained judgment that cannot be fully codified, and it is acquired only by being exercised.

Presupposed That evaluation is criteria-checking; that judgment assisted is the same as judgment developed.
At risk Tacit evaluative expertise — and the apprenticeship pipeline through which each generation of evaluators is formed.
The boundary question Which acts of evaluation must remain human practice, precisely so that human evaluators continue to exist?
6 · Deciding

The Future That Was Only a Forecast

Decision-intelligence systems treat the future as a distribution over the past and deciding as optimizing. Genuine anticipation is something else: running internal models of possible futures while remaining able to re-perceive the situation itself — to notice that the model's ontology of the problem is not the problem.

Presupposed That deciding is optimizing; that the model's framing of the situation is the situation.
At risk Anticipation and sensemaking — the capacities that operate precisely where models fail: novelty, ambiguity, and the moment the frame itself must change.
The boundary question Which decisions must remain acts of human anticipation rather than acceptances of a forecast?

We are not staging prosecutions. Each group is anchored by participants whose life's work names what is at risk, joined by practitioners of the very approaches under examination — invited not as defendants but as co-explorers. Each group produces a short statement of its presuppositions, its capacity at risk, and its proposed boundary. These statements become raw material for that evening's AI Dramaturg synthesis, for the overnight remote cohort, and for the Orthogonal Map drawn on the 29th.

Three Formats, No Panels

I–Thou

Buberian Dialogues

Three protagonists engage through a facilitator. Distinguished listeners respond. The audience makes meaning. No debate — genuine encounter. The format is designed to produce insight that neither monologue nor panel discussion can reach.

Keynote Dialogues

Two thinkers, one question, one hour. Not a lecture followed by Q&A — a structured intellectual exchange in which both participants are changed by the conversation. The audience witnesses thinking in real time.

ψ = M

AI-Augmented Working Groups

Participants receive access to three curated research corpora and a custom interface combining Claude and NotebookLM. Groups use AI as dramaturg — retrieval, synthesis, pattern-matching — while humans supply the judgment that turns information into insight.

The Overnight Working Group

Remote participation here is not a livestream. This is a dialogue conference, and a camera pointed at a conversation twelve time zones away produces spectators, not participants. So the remote program is designed as a working role — one that enacts the conference's own thesis. Each evening, an AI dramaturg synthesizes the day's dialogues: the retrieval, the pattern-matching, the 60%. The remote cohort supplies what it cannot — the questioning, the reframing, the situated judgment. You are not watching the conference. You are stress-testing its central claim.

Three Evenings · 26–28 Sept

The Nightly Dialogue

Ninety minutes, live on Zoom, at 22:00 Shanghai time on the 26th, 27th, and 28th — hosted by Ken Friedman from Sweden and led by Hugo Letiche and Michael Lissack. The AI dramaturg presents its synthesis of the day. Two or three remote protagonists — roles rotating across the three nights — respond, question, and reframe. Open exchange follows. Capped at twenty-five participants so it remains an encounter, not an audience.

While Shanghai Sleeps

The Overnight Loop

Because your working day begins as Shanghai's ends, the remote cohort works the night shift. You receive the same three curated corpora and the same Claude + NotebookLM interface as the in-room groups, and you apply the Orthogonal Map and the Journey Method to your own domain. A distillation of the overnight work is carried into the next day's sessions in Shanghai. The conference thinks around the clock.

Afterward

The She Ji Path

Remote participants who complete the overnight loop are eligible, on the same footing as in-room participants, to develop contributions for the companion She Ji theory issue, “The AI Dramaturg,” supported by the same research infrastructure.

22:00 Shanghai
16:00 Paris · Berlin
15:00 London
10:00 New York
07:00 San Francisco

Sessions are held on Zoom; joining details follow acceptance. There is no fee for remote participation. To apply for a remote place, write to michael.lissack@gmail.com with a few lines on your domain and the question you would bring. Remote places are reviewed on the same rolling basis as in-room applications.

Shanghai, China

The conference is co-hosted by Shanghai University of Engineering Science (SUES) and Tongji University, situated in one of the world's great design cities. Shanghai's position at the intersection of Eastern and Western intellectual traditions makes it the natural home for a conversation about designing AI's boundaries.

The opening two days take place on campus — one at Tongji University's College of Design and Innovation, one at Shanghai University of Engineering Science. On 28 September the subconference moves to the WDCC main stage at the Convention Center Riverside, on the Bund, for the afternoon, and the closing day returns to campus. The assignment of specific days to Tongji and SUES is being finalized.

Institutional Hosts
SUES & Tongji University
With the support of President Lou Yongqi
Main Stage
Convention Center Riverside
On the Bund — afternoon of 28 September, as part of WDCC
Media
Major International Coverage
Press and media partnerships confirmed
Publication
She Ji Theory Issue
Companion issue: "The AI Dramaturg" — planned for publication following the conference

See the Conversation

Invitation Film

A two-minute invitation to the conversation

Watch on YouTube →

Intellectual Foundations

The ideas beneath the conference — tacit knowledge, abduction, trust, and the boundaries of the machine

Watch on YouTube →

The Academic Case

The conference's intellectual framework, in slides

View the slides →

Program Overview

The four days at a glance — a printable PDF of the full program

Download the PDF →

Join the Conversation

This conference is by invitation and application. We are seeking participants from design, complexity science, philosophy of mind, AI research, education, and the arts who want to spend three days thinking carefully about the boundaries between human and machine intelligence. No paper submissions required — bring your questions. And there are no fees: participation, whether in Shanghai or remote, is free of charge.

Apply to Participate

Applications reviewed on a rolling basis. There is no registration fee.
For inquiries, contact Michael Lissack at michael.lissack@gmail.com

She Ji: The AI Dramaturg

A companion theory issue of She Ji: The Journal of Design, Economics, and Innovation is in development, titled "The AI Dramaturg." The issue will explore the conference's central proposition: that AI's role across disciplines is best understood through the dramaturgical metaphor — and that the irreducible human contribution lies in the situated judgment that no model can replicate.

Participants in the conference's AI-augmented working groups will have the opportunity to develop contributions for this issue, supported by the same research infrastructure available during the conference itself.

設計
She Ji · Design