This is my talk from May 2026. There is also Oracle Machines an upcoming exhibition on this general topic – booking now for the free opening – Oracle Machines, CAS at BCS Moorgate London – 21 October.
INTRO
Hi, this is Geoff Davis. Today’s talk is called An Unexpected Story and concerns AI text, literature, poetry and text art. We will also briefly look at AI and the publishing industry.
Over the last few years, when I’ve given talks about AI and art, the general mood was curiosity rather than hostility. But now there seems to be a growing backlash. I’ll mention some of the reasons why.
ART
We’ll look at programmed art that is sometimes labelled AI, although not in the modern commercial Large Language Model sense that currently dominates discussion. There are many kinds of AI art, and some of the work we’ll discuss uses procedural and generative systems developed long before today’s commercial platforms.
I’ll show some of my own AI-related artworks. Then Dr. David Link will discuss his famous Poetry Machine, which has been exhibited internationally and used for music and installation works across Europe and China.
As part of the Computer Conservation Society, David also recoded the very first computer text generation system, Love Letters by Christopher Strachey from 1952. This used procedural methods similar to those I later used in my own Story Generator from 1985.
AI BACKLASH
At the moment there appears to be a significant backlash against AI, particularly within creative industries. Artists, writers and musicians increasingly argue that AI systems are damaging livelihoods through copyright issues, imitation and automation.
There are also concerns about environmental costs, labour issues, concentration of power, misinformation and the general effects of algorithmic culture. Critics range from artists and publishers to AI researchers themselves, including commentators such as Gary Marcus.
Despite Large Language Models being an groundbreaking and universal technology with many applications, disappointment is a recurring theme after years of hype about superintelligence saving humanity or solving climate change. The most tangible effects so far are fake media disrupting fragile ecologies and economies, and job losses through vibe coding and AI agents. Protein folding may have been solved, but what practical benefits follow, given that medicine involves far more than protein structure alone? This is a pure science advance that does not affect people right now, as they see the internet descend under 40% AI slop posts.
Even Phreak Magazine for computer art is asking for a ‘reset’ of an ‘evolving computational climate that’s becoming more and more hostile to its users’, and want to ‘reclaim digital sovereignty’.
REACTION
The UK Society of Authors has launched a Human Authored initiative, arguing for clearer distinctions between human and AI-generated writing. There are also campaigns focused on copyright, consent and compensation for training data used by AI companies.
Part of the problem is that AI systems were trained by scraping enormous quantities of online material. While companies argue this constitutes legal training data, many creators feel their work has effectively been absorbed into commercial systems without permission or payment.
This differs slightly from earlier internet disruptions. File sharing and streaming damaged income models for music and publishing, but AI introduces a new issue: generated material can now imitate or replace aspects of the original creative work itself.
GENERATED WRITING
Generated books are not entirely new. Since the rise of ebooks and self-publishing platforms such as Kindle, there has long been a market for automatically generated or “spun” text. These systems often used primitive techniques to rapidly produce topical books designed to exploit search algorithms.
But contemporary AI systems are much more sophisticated.
Literature already contains strong genre structures — crime, romance, science fiction and fantasy all rely on recognisable tropes and patterns. AI systems can now generate stories using combinations of themes, styles and prompts almost instantly.
You could ask for a story involving Spanish galleons, domestic animals and bananas, and the system would construct something around those elements. In that sense, AI may be moving toward what could be called ambient fiction: stories generated on demand for temporary reading experiences.
You might eventually say:
“I’ve got 20 minutes — generate a 20-minute story.”
The system could potentially adapt to personal preferences, reading level, pacing and subject matter automatically. This is still somewhat futuristic, but many of the technical components already exist.
At the same time, it is arguable that strong human authors will continue to attract audiences because readers value distinctive voice, personality and lived experience.
SOCIETY OF AUTHORS
The Society of Authors provides an interesting example of how the publishing industry is responding. Their Human Authored proposal allows AI use for brainstorming, outlining and research, but not for generating the final prose itself.
This immediately raises difficult questions. Most writers already use spelling correction, grammar tools, search engines and digital editing systems. Translation software and AI-assisted audiobooks complicate the distinction even further.
I contacted the Society of Authors to ask about procedural text generation systems using entirely self-written material. For example, my Story Generator from 1985 used words and structures manually entered by me. It generated endless variations from a seed story I originally wrote myself.
Their response suggested that even defining “human authored” becomes complicated once generative systems enter the process.
But the chill is out there. The indie publisher that puts out the AI Creative Anthologies, which I’ve talked about before on these CAS talks, now is making a separate unrelated channel for AI art books, since they might get banned or penalised, affecting their normal books.
And the Pattern City novel, I’ll discuss soon, will be print published with new hand drawn illustrations, rather than AI generated images, so the AI animations will now only exist as art.
The art world has accepted AI and other digital tools going forward from Paintbox such as Quantel and Photoshop in the 1980s. In fact Sasha Stiles, a poet, and Refik Anadol are very popular and have huge immersive shows all over the world.
FINE-TUNING AI
Another contemporary method is “fine-tuning,” where artists train or adapt AI systems using their own material — texts, images or sounds.
In this approach, the base AI model already exists, but additional personal data is added so the outputs reflect a particular style or voice. Systems can now imitate style from surprisingly small samples.
I experimented with this years ago and found the results surprisingly useful creatively. The AI could produce unexpected combinations and ideas within the framework of my own themes and writing patterns.
Technically, however, this still exists in a grey area because the underlying base model may have been trained on scraped internet data.
AI, ART AND ATTENTION
I began research into AI and text at UAL in 2019, before the current explosion of public AI systems. At the time, most attention focused on image generation rather than writing.
Today, text, image, video and music generation are all developing rapidly.
One interesting shift is that online art increasingly functions within an attention economy. Viral discussion, controversy and circulation can become more important than the individual artwork itself. In some cases the conversation surrounding the work becomes the artwork.
Generated fiction may eventually function similarly — less like permanent literature and more like temporary, personalised media generated for specific moods or moments.
Whether this becomes culturally valuable, disposable entertainment, or something in between remains an open question.
MY ART
STORY GENERATOR 1985
This pioneering text generation software creates unsettling variations on an original base story by Geoff Davis. The topic is animal insurrection triggered by a human-caused disease.
It was released on Micro Arts as MA4 Story Generator, and distributed on Prestel teletext along with a lot more of my art software. It was exhibited at the time at the London Film Makers Coop, and has been shown a in UK and Europe recently, the latest was at the Essen AI Biennale November 2024. Story Generator was reviewed at the time but only in print magazines like Computer News, Sinclair News and Blitz magazine. As Micro Arts had no gallery connection, all this ground-breaking digital art was lost to art history until rediscovered by Sean Clark of CAS in 2020.
The generator methodically replaces words from my original story “Cow Boils Head” with themed synonyms, generating new narratives at a hypnotic pace. The generator was written in Sinclair BASIC for the ZX Spectrum microcomputer.
The work emerged during Britain’s BSE crisis, when the practice of feeding cattle with processed animal remains led to widespread “mad cow disease.” I transformed this animal pandemic, into a series of dark procedural fictions where different species – wolves, vultures, lizards – rise up against humanity.
Unlike modern AI’s statistical approach to text generation, Story Generator uses a carefully structured “expert system” with predefined word banks. This early form of Procedural Story Generation (PSG) contrasts with today’s Machine Learning methods, which train on vast datasets to predict likely word sequences. While contemporary curator Georg Bak describes it as “a precursor of ChatGPT,” the work’s methodical deconstruction and rebuilding of narrative creates striking juxtapositions that reveal new meanings while maintaining thematic coherence.
PATTERN CITY
The recent work: Pattern City animation system. I wrote a novel, which I finished around 2016, and then didn’t really know what to do with it — it was kind of finished but not quite.
Later, I started making AI images, experimenting with AI imagery. This was quite amazing, to see what the images would be like. I soon had far too many. So I thought I’d make some illustrations for the novel. These were then used by a image processing program which places them in a grid, and applies colour overlays, in real time. The grids can be a single cell, or many, and all can be programmed, so I had an animation system that can run continuously, using any images, in this case, the illustrations from the book.
Given the hostility to AI in the publishing industry, I might hand draw images for the print book, and leave the animation system as a standalone art piece.
The first ones were shown as still images in the 2023 CAS summer exhibition in London. They were quite dystopian looking and had overlaid text. There was an idea to make them printable, with cut out areas for folding into shapes for model building.
Stage two was much more straightforward and a bit less dystopian looking — quite cheerful looking graphics, apart from the occasional disaster scene. I decided to layer colour on to make it look more attractive. The second set was shown as a running animation in 2025 at the Drivers exhibition at CAS.
You could go through the entire novel scene by scene from start to finish and make an incredibly long animation, and then narrate the story. So this is all ongoing work at the moment.
The earlier ones weren’t really animated — they were just images that flashed up — but by using multiple images you can layer them and get this flickering effect. All of that was quite interesting.
And of course this is putting illustrators out of work, because I’m not a professional illustrator. In theory if I’d paid to do this I would have had to work with an illustrator, draw loads of images, edit them. Maybe it wouldn’t even have existed if I hadn’t had a generator..
STUDIO SQUARES
The Squares program is the most recent. You start to program, the letters and words appear, you get the graphical effect. This graphical effect is based on a 1984 artwork Studio Walking, as small feet walk around the screen, trapped forever inside the display – like now with scrolling on the phone.
The text appears letter by letter, and if a long story will appear over subsequent pages. Colours and grid can be controlled. The story can be randomly generated or a set story. This is still a work in progress.
Just a quick mention of a computer game written to go with a novel from 1994 – Nnn Goes Mobile. This was a cyberpunk novel on an indie press. The game was on the Arts Council’s Hub CDROM. This seemed modern compared to data cassettes from the 1980s Micro Arts days.
And an artist did some 3D modelling to make graphics for another novel, which was serialised in an art book called Em in the late 1990s. Here’s some images from that. All credits are in the references.
— —
So now I’ll hand over to Dr David Link, who will talk about his Poetry Machine, recoding Love Letters and Pandaemonium.