From Black Box to Visible Chain: Teaching Architectural Foresight Through AI-Curated Scenarios.

What if a first-year architecture student could ask a machine about the future of an entire city, then trace for themselves how the machine reached its answer? That simple question condenses an entire educational project now taking shape in architecture schools worldwide: an introductory course teaching prospective thinking – what the French school calls la prospective – while a large language model plays curator, organising evidence and documenting every analytical step in a knowledge system the student can audit. This innovative approach to education puts a spotlight on understanding complex processes. The idea is close to turning architectural research from a black box into a visible chain of decisions, and the wager is a whole generation of architects who understand scenarios not as magical predictions but as methodical constructions that can be taken apart.

Why tomorrow’s architects need a methodology of the future

Architecture is by nature a futuristic profession: every project drawn today is built years later and inhabited for decades. Yet its teaching all but ignores foresight. The founding literature for this course begins with Voros’ classification of prospective methods, the field’s most important reference framework. Voros distinguishes modes of thinking – evolutionary and extrapolative versus revolutionary and discontinuous – from five levels of interpretive depth: event, trend, system, worldview, and historical context. The pedagogical value is immediate: a student who learns to tell these levels apart discovers that most architectural debates about the future are stuck at the trend level, while the real analytical power lies in descending to the worldview level where the deep assumptions that govern what we consider possible are formed. Voros himself confirms that his classification framework was designed as a teaching aid to introduce students to the range of prospective methods available – precisely the context of a 101 course.

Masini completes the frame with her three levels of futures thinking: prognosis by extrapolation, visions of desirable futures, and project building that synthesises both. This three-part model gives a first-year student a simple but powerful mental map of what prospective thinking can achieve, with Masini’s insistence on the futurist’s ethical responsibility and on listening to the seeds of change. Coyle adds a practitioner’s inventory of methods: inductive reasoning, graphical techniques, influence diagrams and numerical methods on the analytical side, then narratives, Delphi, field anomaly relaxation and multiple scenarios on the judgemental side, with worked examples that serve as templates for student exercises with machine assistance.

The French school and causal layered analysis

No course on prospective thinking can bypass Godet, founder of the French school that gave the field its name. Godet condenses his thought into seven pillars: clarifying present action in light of the future, exploring multiple uncertain futures, adopting a global systemic approach, accounting for qualitative factors and actors’ strategies, remembering that information and forecasts are never neutral, opting for plurality and complementarity of approaches, and questioning preconceived ideas. His most famous tool, structural analysis through influence and dependence matrices, offers a rigorous qualitative method that architecture students can execute with AI assistance: identifying the governing variables of an urban scene through a visible mathematical matrix rather than floating intuition.

The most teachable tool, though, is Inayatullah’s causal layered analysis, which works on four levels: the litany of daily headlines, social causes, discourse and worldview, and myth and metaphor. An architecture student who learns these layers discovers that a surface trend like urban growth is merely the expression of deeper worldviews and root metaphors governing our image of the city. The method’s requirement to move up and down the levels creates a natural workflow for an AI system that documents and displays each layer, so every analytical level can be recorded as a separate node in an auditable knowledge network. The five case studies in the original paper, from Bangkok traffic to university funding, prove that layered analysis produces richer scenarios than conventional approaches.

The bridge between foresight and urban design

The question forces itself: what does any of this have to do with architecture specifically? Childs answers in his essay on urban design foresight: the future of urban design should include more robust tools for envisaging potential futures, and sketching multiple iterations of alternatives is essentially a scenario thinking tool. That sentence condenses the disciplinary justification for the entire course: architecture already practises prospective thinking through design iteration, and the course simply makes this practice explicit and systematic. Sepe adds that teaching in the context of rapid urban change faces a real challenge – it is not easy to foretell what cities will be like even a short time from now – which demands a shared basic vocabulary and reproducible methods, exactly what a documented knowledge system provides.

Scenarios: from the boardroom to the classroom

The course inherits its scenario tools from both the academic and the corporate worlds. Ruff documents how scenarios function in major organisations as both mental models and methodology, through three documented case studies, linking foresight directly to organisational renewal. Oliver and Parrett offer a simple, disciplined teaching methodology built on the Garvin and Levesque approach: from the focal issue through PESTLE analysis, driving forces, critical uncertainties, a two-by-two matrix, scenario narratives, and strategic options. The method is structured enough for machine-assisted execution and simple enough for first-year students, with a YouTube case study walking the entire process step by step.

The machine as curator: ethics of AI-assisted qualitative analysis

Here lies the course’s contested heart. Paulus and colleagues’ study of how the role of AI is constructed in qualitative analysis identifies five discursive patterns researchers trade in: portraying qualitative analysis as chronically problematic, describing it as easily automated, presenting AI as disruptive yet not replacing humans, treating hybrid human-AI methods as the field’s future, and decentering ethical concerns. The study’s sharpest warning is against the habit of describing qualitative research as time-consuming, labour-intensive and subjective, because that description primes the machine to be received as a saviour. The alternative the study proposes, and this course adopts, is to present AI as a tool for deepening interpretive rigour rather than shortcutting it, while documenting both machine and human analytical decisions together.

Smirnov’s study supplies tested practical protocols: documenting prompts and outputs, repetitive classification with averaging, simplifying outputs, separating exploratory and confirmatory phases, and reporting machine-learning metrics. Its headline finding – the language model achieved perfect accuracy in identifying meaning in narrative data and reliable causal coherence assessment – matters less to the course than the transparency protocol itself: every analytical step recorded and reviewable. The Gümüş and Oğuz platform provides the engineering reference model: a web platform combining a fine-tuned language model with structured expert input to generate scenarios through the STEEP framework and Jim Dator’s four futures archetypes – growth, collapse, discipline, transformation – with evaluation panels and expert feedback loops that make every step visible. He and colleagues’ paper on urban planning in the era of large language models adds the third dimension: conceptualisation, generation and evaluation, where models tested on professional planning exams outperform the top ten percent of human planners, and the paper describes a conversational iteration between planner and machine as a transdisciplinary consultant.

How futures are actually taught

The pedagogical layer has its own literature. Heinonen and colleagues experimented with introducing causal layered analysis as a deliberate disruption in the middle of an advanced scenario course: a reflective pause inside the linear process that prevents path-dependent lock-in and increases reflexivity, with students rating the improvement of dialogue at 4.4 out of 5. Damhof and Mazzeo delivered a complete instructional design model for building futures literacy through three learning outcomes – enhanced perception, embracing complexity, and a new sense of agency – drawing on UNESCO’s Futures Literacy Labs and the Discipline of Anticipation. That sense of agency is the outcome closest to an architecture student who will design the built environment for decades to come: the feeling that the future is something you make, not something that happens to you.

From classroom to market: corporate foresight

The course does not close its door on commercial application. Ruff’s second paper documents twenty-five years of corporate foresight practice at DaimlerChrysler’s Society and Technology Research Group, identifying five fields of practice: strategic futures research, tomorrow’s markets, prospective evaluation of innovations, business processes and organisational change, and scanning and monitoring. Saheb and colleagues contribute the processed model: from scanning to analysis to interpretation to strategy formulation to implementation, with defined external and internal capabilities and an institutional maturity model. The educational value of this literature is a realistic map of what awaits the graduate: companies genuinely seek people who can turn uncertainty into decisions, and the scenario-trained architect is better qualified for that role than most assume.

Toward a new educational contract

This course gathers three threads that were separate: rigorous foresight methodologies from Voros to Godet, transparent AI-assisted qualitative analysis, and an architectural education that recognises designing place as already designing time. The thread joining them is documentation: every analytical step, whether completed by student or machine, is recorded in an auditable knowledge network, forming an accountable chain of reasoning from data to scenario. This is the true meaning of the word curator: not the machine that answers, but the machine that shows its work. In this the course proposes a model for teaching architecture in the age of AI that neither hides the tool nor surrenders to it, but makes it an object of critique and accountability from the first day of study – a skill that will define the coming architect’s value more than any drawing tool.

✦ ArchUp Editorial Insight

The file these eighteen references assemble deserves careful reading from every architecture department thinking about modernising its curriculum. The core idea is not bringing AI into the studio – that is happening everywhere already – but inverting the relationship: instead of the machine being a production tool the student consumes, the machine becomes the object of training in methodological critique. A student who learns to document the full chain of reasoning, from Godet’s matrix to Inayatullah’s layers to Dator’s archetypes, acquires a double immunity: against the magical claims of technology, and against the learned helplessness before it.

The beautiful irony is that architecture practised foresight before the word existed: every design competition is a scenario exercise, every master plan a wager on an uncertain future. What this course adds is methodological rigour and documentary transparency, precisely what the design scene lacks amid its plague of seductive renderings. The largest wager remains pedagogical: the sense of agency – a first-year student feeling they make the future rather than receive it – is the most valuable thing an architecture school can plant in a generation that will spend its career repairing a full century of blind technological optimism. If this model succeeds, we may see within a decade a generation of architects who treat language models the way accountants treat calculators: a powerful tool requiring a conscious supervisor, not a prophet delivering revelation.

References

• Voros J. Introducing a classification framework for prospective methods. Foresight. 2006;8(2):44-59. doi:10.1108/14636680610656174
• Voros J. From forecasting and scenarios to social construction: changing methodological paradigms in futures studies. Foresight. 2002;4(3):5-15. doi:10.1108/14636680210697731
• Masini EB. Rethinking futures studies. Futures. 2006;38(10):1158-1168. doi:10.1016/j.futures.2006.02.004
• Coyle RG. The nature and value of futures studies or do futures have a future? Futures. 1997;29(1):77-93. doi:10.1016/s0016-3287(96)00067-5
• Godet M. Introduction to la prospective. Futures. 1986;18(2):134-157. doi:10.1016/0016-3287(86)90094-7
• Inayatullah S. Causal layered analysis. Futures. 1998;30(8):815-829. doi:10.1016/s0016-3287(98)00086-x
• Childs MC. Urban design foresight. Journal of Urban Design. 2019;24(6):813-815. doi:10.1080/13574809.2019.1706309
• Sepe M. Shaping the future: perspectives in research on, and the teaching of, urban design. Journal of Urban Design. 2019;24(6):810-812. doi:10.1080/13574809.2019.1706308
• Ruff F. The role of scenarios in strategic foresight. Technological Forecasting and Social Change. 2010;77(9):1513-1534. doi:10.1016/j.techfore.2010.06.010
• Oliver JJ, Parrett E. Managing future uncertainty: Reevaluating the role of scenario planning. Business Horizons. 2018;61(2):285-295. doi:10.1016/j.bushor.2017.11.013
• Paulus TM, et al. The construction of the role of AI in qualitative data analysis in the social sciences. AI & Society. 2025. doi:10.1007/s00146-025-02488-3
• Smirnov I. Enhancing qualitative research in psychology with large language models: a methodological exploration and examples of simulations. Qualitative Research in Psychology. 2024. doi:10.1080/14780887.2024.2428255
• Gümüş A, Oğuz M. Design and Implementation of an AI-Assisted Platform Leveraging Large Language Models for Foresight and Scenario Generation. ISMSIT. 2025. doi:10.1109/ismsit67332.2025.11268250
• He Z, et al. Urban planning in the era of large language models. Nature Computational Science. 2025. doi:10.1038/s43588-025-00846-1
• Heinonen S, et al. Drilling and Blasting to Learn Scenario Construction: Experimenting with Causal Layered Analysis as a Disruption of Scenario Work. World Futures Review. 2018;10(1):31-48. doi:10.1177/1946756718774940
• Damhof L, Mazzeo C. Mastering futures literacy in higher education: An evaluation of learning outcomes and instructional design of a faculty development program. Futures. 2021;132:102814. doi:10.1016/j.futures.2021.102814
• Ruff F. Corporate foresight: integrating the future business environment into innovation and strategy. International Journal of Technology Management. 2006;34(3-4):230-253. doi:10.1504/ijtm.2006.009460
• Saheb T, et al. Corporate foresight: developing a process model. European Journal of Futures Research. 2018;6:18. doi:10.1186/s40309-018-0147-7

Brought to you by the ArchUp Editorial Team

Inspiration starts here. Dive deeper into Architecture, Interior Design, Research, Cities, Design, and cutting-edge Projects on ArchUp.