The AI skillset

I’m presenting an update to my reflections of April 2014 that I titled Life-changing technologies #189. That entry was an update on a report we wrote in the 1990s based on interviews with architectural practitioners. Now we are a further twelve years on from that update.

Even without the benefit of a new cache of interview responses we can say with some confidence that AI is the most recent tech that has a bearing on the way architecture is practiced and taught. 

Our main thesis in the original study and subsequent reflection was that technologies have the capacity to reveal important aspects of our practices that were not obvious before. In that study we were less interested in the “impacts” of technologies than on what they “disclose.” New technologies provide new ways of framing our current practices. 

For example, in the earlier report we argued that CAD was beginning to reveal architectural firms as custodians of digital information, not merely producers of drawings. What had previously appeared as drawing, documentation and record-keeping began to look more like database management, version control, archiving and information retrieval.

BIM (building information modelling) later intensified this reframing by presenting the building not simply as a set of drawings, but as a structured information model with relevance across the building’s lifecycle. In that sense, architectural firms could begin to see themselves not only as designers of buildings, but also as archivists and custodians, i.e. managers of information about buildings.

Similarly, for those of us involved in education, such reframings influence how we approach research, teaching and learning.

As usual, I uploaded the 2014 entry to ChatGPT with my own prompts about the emerging influence of AI on practice. The AI response seems to capture the tenor of our earlier reflections, and largely accords with my own opinions — dare I say, expressed with greater clarity and felicity.

Life-changing technologies

Helpfully, ChatGPT seemed to support our contentions:

“AI fits very well into the argument of the original post because it is not merely another useful tool. Like CAD, desktop publishing, multimedia, smartphones and social media, it re-describes practice. It discloses activities that were already there but not always foregrounded: drafting, searching, comparing, classifying, summarising, prompting, editing, judging, attributing, and taking responsibility for claims.”

Hopefully the rest of the AI insights in this article are informed by my previous conversational history with ChatGPT, so I can claim some purchase in the authorship of what follows — that and my own editing of the AI responses. 

AI as reconfiguration

The AI confirmed that the most obvious recent example of such technological reconfiguration is generative AI, especially large language models (LLMs). AI systems that enable people to do things more easily than they could do before: summarise documents, translate, draft letters, generate code, classify data, produce images, simulate dialogue, and search through large bodies of text in conversational form.

But, as with earlier technologies, the most interesting point is not simply that AI enables tasks. It also reveals something about those tasks. It shows that much of what we call professional or academic work resembles in some measure what LLMs do: recognise patterns, paraphrase, classify, compare, compress, expand, predict and re-style. These activities were never absent from writing, design, research, administration or teaching. AI makes them conspicuous.

The word processor revealed writing as an activity of revision, formatting and document management. Desktop publishing revealed architectural and academic practices as forms of publication. CAD and BIM revealed design practice as the production, coordination and exchange of structured information. Social media revealed scholarship as partly a matter of networking, visibility, affiliation and performance. 

AI now reveals that many expert practices involve the manipulation of language-like materials: reports, prompts, briefs, correspondence, captions, code, summaries, metadata, classifications and proposals.

I should say that many of the points outline here I adumbrated in my book AI and Language in the Urban Context: Conversational Artificial Intelligence in Cities published last year. (See publisher’s open access version.)

This does not mean that AI has exposed professional work as mere text production. On the contrary, it makes clearer the difference between producing plausible language and exercising judgement. The fluent answer, the polished paragraph, the generated image, the code fragment or the summary is not the end of the task. It is often the point at which the real work begins: someone, or some agent, has to check, select, reject, contextualise, cite, edit, take responsibility.

Prompting and practice

The rise of AI has also revealed the importance of instruction-giving. “Prompting” sounds like a new skill, but it belongs to an even older family of practices: to brief a colleague, set an essay question, write a design brief, specify a database query, compose a search term, instruct a less-experienced assistant, or explain a task to a student. AI makes this ordinary activity more visible. According to ChatGPT,

“In that sense, AI does not simply automate work. It changes the way work is articulated. A vague prompt produces a vague result. A precise prompt requires the user to think carefully about purpose, audience, genre, evidence, tone, exclusions and constraints. The better uses of AI often depend less on technical mastery than on the capacity to say what kind of answer would count as useful.”

This has implications for education. If students use AI to write essays, teachers and potential employers will be understandably anxious about plagiarism and assessment. But there are also more interesting pedagogical questions. What does it mean to frame a question well? What does it mean to ask for a comparison, a critique, a counter-argument, a summary, an outline, a revision, or a translation? AI brings such questions to the surface.

Deskilling and re-skilling again

AI may deskill writers, designers, coders, teachers, researchers and administrators. If a person never has to compose a sentence, structure an argument, search a database, draw a diagram, write code, or summarise a text unaided, then those capacities may weaken. Technologies can make us forget how difficult certain tasks are.

But AI also re-skills. It rewards other capacities: asking good questions, comparing outputs, detecting made-up “facts,” checking references, recognising genre, identifying bias, specifying constraints, and deciding when not to use AI assistance. 

“It may also revive older virtues. Slow reading matters more when summaries are cheap. Citation matters more when fabricated references are easy to produce. Judgement matters more when fluent prose is abundant. Authorship matters more when text can be generated without experience, commitment or accountability.”

I recall a time when poor spelling, grammar and expression automatically put a writer (e.g. a student) in a lower assessment bracket. AI assistance means that everything can be word perfect, requiring that assessors and reviewers pay more attention to what is actually being said. As suggested already, perfect syntax does not equate to perfect sense. It’s a skill to identify the difference.

My AI counsellor here even went so far as to say that AI does not simply remove skills. It redistributes them. As is the case of writing this article, it shifts attention from production to evaluation, from drafting to revising, from searching to verifying, from expression to responsibility. 

“That shift is not necessarily benign, but nor is it simply a decline.”

AI and architectural practice

In architecture and design, AI extends a line of development already evident in CAD, rendering, BIM and parametric modelling. Earlier technologies revealed architecture as drawing, assembling databases, coordinating, simulating, publishing and presenting. AI reveals it additionally as a practice in which an agent describes, classifies, produces images, generates scenarios and prompts iteratively.

A design office may use AI to generate mood boards, draft planning statements, summarise regulations, produce precedent studies, prepare public consultation material, or explore visual alternatives. These are not marginal activities. They are part of how architectural work is framed, communicated and justified. 

AI therefore reveals something about architecture that may have been under-acknowledged: that design practice is deeply rhetorical. It involves persuading clients, planners, consultants, communities and colleagues through words, images, diagrams and narratives.

But AI also disturbs the status of those representations. A rendered image once implied a certain investment of time and intention. A generated image may be produced in seconds. That does not make it worthless, but it alters the evidential and rhetorical force of visual material. The question becomes: what does this image commit us to? Is it an exploration, a promise, a deception, a provocation, or atmosphere?

Disruption without inevitability

You could say that AI is a disruptive technology. It disrupts assessment, authorship, professional services, search, publishing, image production, coding and administration.

At this point in its assistance, ChatGPT helpfully opined that “disruption is not the same as replacement.” It elaborated that earlier claims about paperless offices, virtual campuses, frictionless global communication and fully automated design were never realised in simple form. Technologies enter practice unevenly. They are resisted, absorbed, modified, domesticated and sometimes abandoned.

“The same will happen with AI. Some uses will become routine and almost invisible: spelling correction, transcription, translation, formatting, summarisation. Others will remain contested: generating academic prose, producing design imagery, marking student work, writing peer reviews, simulating human interaction, or making decisions about people. Some uses will be prohibited; others will be encouraged; many will sit in a grey zone governed by convention, disclosure and institutional policy.”

As in the 1990s, the main issue is not simply what the technology can do, but what practices it reveals and reconfigures. To repeat, AI exposes the extent to which contemporary work depends on language, classification, prediction, pattern recognition and performance. It also exposes the limits of those operations.

“A plausible answer is not necessarily a true one. A coherent paragraph is not necessarily an argument. A synthetic image is not necessarily a design. A summary is not necessarily understanding.”

What AI reveals

So what does AI reveal about us? It reveals that we are not only tool-using, mobile, image-making social entities. We are also creatures who prompt. We ask, instruct, query, revise, correct, compare and narrate. We seek assistance not merely to save labour, but to externalise thought. We use tools to show us versions of what we think, or wanted to think.

“AI also reveals the fragility of many distinctions on which institutions depend: original and derivative, author and assistant, search and synthesis, reading and summarising, drafting and editing, public and private, human and machine production. These distinctions have never been absolute, but AI makes their instability harder to ignore.”

Our earlier report on computers in architectural practice argued that information technology constituted a perturbation in the praxis of the firm. That still seems the right word. AI perturbs. It interrupts settled habits. It does not simply add efficiency to existing routines. It prompts us to ask what a routine is, what counts as expertise, and where responsibility lies.

“The task, then, is not to ask whether AI is useful. It is. Nor is it enough to ask whether it is dangerous. It can be. The more interesting question is what forms of practice AI brings into view, what it obscures, and how we might incorporate it without surrendering the capacities it appears to emulate: judgement, care, interpretation, and accountability.”

Note

In this and other newsletter entries (I’m rebranding my blog as a newsletter now!) I’ve been experimenting with a mode of writing that acknowledges the LLM I am using as an advisor or co-author — and in the text itself. Will it catch on as a writing method, or does its self referentiality read as tortured meta fiction?