img(height="1" width="1" style="display:none" src="https://www.facebook.com/tr?id=2939831959404383&ev=PageView&noscript=1")

Why AI can only take you so far on the journey to art

Words:
Will Wiles

Human creativity can’t be imitated by machine learning, says Will Wiles, but artificial intelligence does illuminate the linguistic roots of design

Does Eero Saarinen’s work, seen here at the TWA terminal, New York,  show that talking can  be a form of authorship  in architecture?
Does Eero Saarinen’s work, seen here at the TWA terminal, New York, show that talking can be a form of authorship in architecture? Credit: Cameron Blaylock / RIBA Collections

The plot of William Gibson’s cyberpunk novel Count Zero (1986) hinges on a question of authorship. Who is the anonymous artist of a remarkable series of found-object collages? The answer is no one – no artist, anyway. They are made by a neglected artificial intelligence hiding in an orbital server farm, and their melancholy meaning must remain somewhat inscrutable. 

When I first read Count Zero 30 years ago, the thought of art made by artificial intelligence felt brilliantly exciting and subversive. Today, I fear we may already be tired of it. AI art may emit the odd twinkle of novelty or strangeness, but that doesn’t alter its fundamentally dreary, imitative nature, and its plagiaristic reconstituting of the work of real human artists with the creative insight of an auto-completed text message. It is guessing what the talented might have made of the instructions of the talentless. 

Gibson’s Boxmaker beguiled with its glimpse of the inner landscape of an alien mind. Today’s AI art just offers a bleak view of the mundane contents of human minds in the tech sector, for whom human art is unnecessary expense and human artists an obstruction. A few months ago, feeling more open-minded and curious about the technology, I argued that architecture was probably one of the better placed arts to weather this tempest of dross, given its reliance on a wider suite of practical and management skills. But with the AI industry’s apparent eagerness to sink its talons into anything faintly enriching to the human spirit, that might change. Hopefully professional bodies and the public at large will get their act together in defence of human creators. 

Yet the subject still has its intriguing aspects. Its defenders claim that creative effort is still required to get the best results, in the writing and refining of the text prompt that the machine works on. It’s a nugatory skill, far more reliant on knowing the right keywords than any descriptive flair, and hardly equals the work of an artist or architect. A client can’t claim credit for an architect’s work because they wrote the brief. A good brief surely goes a long way, but not all the way. Nevertheless, the machine’s reliance on prose expresses something about the way words underwrite other arts. 

Alberti regarded architecture as a sort of rhetoric – a way for a society to express its beliefs and virtues. It is part of a conversation, and conversation can be part of art. Eva Hagberg’s recent When Eero Met His Match examines Aline Saarinen’s contribution to her husband Eero’s career. Hagberg makes a case for the importance of talking as part of the making of architecture. Aline worked as Eero’s publicist: she discussed his projects with him, and then talked about them with the press. In Hagberg’s telling, this helped clarify the ideas and images the great Finnish-American architect employed on his later projects. 

It’s a curious feature of artistic creation – be it book, painting or building – that a fully formed idea is never fully formed. It is not present in the head in every detail, waiting to be mechanically reproduced in the real world. It’s only when you start working on it that you discover what it really is. If I experience a creative problem, and I write to someone about it, I often find that summarising the problem on the page provides the solution: the question answers itself. Expressing something in the form of a text prompt, an instruction or an ‘elevator pitch’ isn’t bad practice. But it’s where the work begins, not where it ends.


Bringing the inside out

As I write, AI enthusiasts are touting the technology’s expansion of the frame of classical paintings, showing what the painters ‘left out’. An asinine exercise with predictably dire results. But I do wonder if – just for fun, you understand – an AI application could be made to join in an architectural guessing game: figuring out the internal layout of a building by looking at it from the outside.