DR McCullochHOMESHOPABOUTLETTERS

I make work about the systems that now sit between us and the rest of our lives, the systems that mediate love, sex, work, family, politics and sometimes our own interiority. A programmer for many years and a CTO in tech and startups, I have been in board rooms as we gamify attention and engineer more clicks. The work is my thoughts about this expressed without words.

I work in layers. I build a ground on board or canvas, or start with a line etched into tile, then scrape, wash and dissolve until forms appear. I might attack an existing painting with an oil stick, or work a small tile for a day with brush pens until a landscape resolves. The process is physical and gestural and it often feels like the material is guiding me; it is automatic. I sometimes get a sense that I’m working with the material to discover what should already be there.

I photograph the work and feed it through a generative pipeline, which misreads the marks and renders figuration that promises more on the screen but when printed, is always less. I sometimes embed this in clear gesso and respond in oil, acrylic, ink or photoshop. Sometimes I use a laser to burn away my marks with the AI-Slop. I call it algorithmic frottage: my mind rubbing against a statistical model.

My practice began as master’s research into emergent narrative at Glasgow School of Art and then prompt engineering within a fine art practice at Central Saint Martins and has come as a surprise to me. I thought I’d make digital work but instead, have become enchanted by materiality.

Luciana Parisi — for thoughts
Issa Samb — for objects and deep-time
Lisa O’Neill — for lyricism
Zanele Muholi — for fearlessness
Alfred Gell — for art anthropology

Education

MA Art and Science, Central Saint Martins, University of the Arts London, 2024.

My research was prompt engineering within a fine art practice.

MSc, Glasgow School of Art, 2019.

Emergent narrative, serious games and VR.

Write to me

About a painting, a show, a commission or anything else.

Fragments

My notes, drafts and titles, broken into single lines and read against each other by a language model.

Since October 2023 I have been sending myself WhatsApp messages about the work, and the same ideas turn up there, in my MA submissions at Central Saint Martins and in seventeen drafts of my artist statement, often in nearly the same words. I broke all of it into single lines and asked Claude Sonnet 5 to read them against each other and say which of them echo, sharpen, contradict or deflate one another.

How it was made →

Show relations the model scored at leastout of 9

Rest the pointer on a cell, or tap it, to read what runs from the row’s theme to the column’s. Or walk the lines one at a time →

Key

Each row and each column is a theme, one of 26 groups of lines with similar meanings that a small model on my laptop found and Claude Sonnet 5 then named and gathered into five bands. A cell counts the relations running from lines in its row’s theme to lines in its column’s theme, and carries a letter for every kind of relation among them, so one cell can hold several. A bigger letter means more relations of that kind.

EOne line echoes another when it says the same thing in other words.

SIt sharpens the other when it is a tighter, better version of it.

RIt rhymes with the other when they share an image or motif without saying the same thing.

COne line contradicts another when it pulls against it.

UIt undercuts the other when it deflates it, usually with a joke.

NOne line names another when it gives a name to what the other describes.

GIt grounds the other when it is the lived experience behind it.

BA note or idea becomes the piece named by the other line when it turned into that piece.

The model wrote the theme names, the band names and the one-line reasons, and I have left them as it wrote them.

How far to trust this

This is one model’s reading of my writing. When it read three themes a second time, 48% of its relations came back the same, and two other models agreed with it less often, though the rankings hold up better, with 17 of the 20 most central lines unchanged when the weakest relations are dropped. Any single relation is best read as a suggestion.

That is why the matrix opens on the strongest relations only, the ones the model scored 7 or more out of 9. The numbers above the grid bring in weaker ones, down to 4, and the letters, counts and examples change to match. Below 4 I have left everything out.

What it shows

The model’s reading of the lines is a graph: every line a point and every relation a link between two of them. These are the things I found when I measured it. The theme and band names in this section are the model’s.

How these were found →

WHAT THE WRITING KEEPS COMING BACK TOThe lines that name the rest

The model read these three lines as naming what more of the other lines describe than any others do, counting the strength of every naming relation that runs out of each.

“I like to call my approach algorithmic frottage.”“His method is algorithmic frottage: scraping himself against datasets, LLMs and PR-dressed-as-research, he captures the friction in oils, acrylic, ink, laser-etched tiles and abandoned Git repositories.”“Algorithmic Entanglement”The lines that join the rest together

More of the shortest paths between any two lines pass through these three than through any others, and under each are the bands its neighbours belong to most often.

“rubbing the ouroboros”AI and Bodies · Language and Systems · Confession and Trauma“But maybe that's the point ... I do a mix of all... I don't aim to have a cohesive single body of work... I embrace the chaos and allow myself to flit between different perspectives... rather than trying to capture the complexity of what I feel in a single form/medium/treatment ... I just interrogate it from every angle and that is the work.”Confession and Trauma · Studio Practice Inventory · Language and Systems“Pilearytix”AI and Bodies · Studio Practice Inventory · Language and Systems
WHAT THE WRITING KEEPS COMING BACK TOGroups that cut across the themes

A second grouping, made from the relations between lines, finds 23 clusters, and they only loosely match the themes. These three cross the most themes:

39 lines across 17 themesMostly from Scraping Self Against Machines, Recursive material self-becoming, Systems, Memory, and Permission“Median Humans: False Memory Edition”“They see memes as living structures inside our brains... Which all thoughts and memories are... Which are then transferred”46 lines across 16 themesMostly from How Do You Actually Feel, Ghosts of presence and absence, Recursive material self-becoming“Not darning. Not patching. Rather; a de/re/weave. The result is a consensus between their illusionary self, and their somatic self.”“the little inspirational things I see that seem to pop up in the things that I'm making”44 lines across 12 themesMostly from Technocapitalism as Artistic Praxis, Emergent narrative from fragmentary meaning, Systems, Memory, and Permission“I interrogate these latent fantasies as both auto-ethnography and catharsis; and through a process of algorithmic drag, emulates the processes, people and tools etc that coalesce as platforms and the feed-forward loops of algorithmically-mediated sensuality.”“New materialism: finding the sacred in the objects that we already have”
ARGUING WITH MYSELFI heckle myself

The pattern of three lines that turns up most often beyond chance is a chain like this one, in which one line deflates a second that deflates a third:

“A proper 8chan little shit.”undercuts ↓“I'm just doing that angry white bloke thing”undercuts ↓“My work is sometimes that of an angry white man raging with class consciousness, ready to be co-opted.”

There are 239 of them, and not one of the deflated lines deflates the other back. These three lines do the most deflating:

“Made a fing”“Start painting your own bloody life”“Constantly madyurbating, that constanza”
ARGUING WITH MYSELFWhere the tension sits

Of 683 groups of three lines that all relate to one another, 56 (8%) never settle, because two of the lines each agree with the third but contradict or deflate each other, or because all three do. Weighing the strength of every contradiction and deflation against the number of pairs the model read, the tension is greatest in these themes:

Systems, Memory, and Permission, within the thememost tension · 23.3 per hundred pairsGhosts of presence and absence, within the themesecond most · 16.2 per hundred pairsRecursive material self-becoming and Emergent narrative from fragmentary meaningthird most · 15 per hundred pairs
WHAT CHANGEDWhat is rising and fading

I compared how central each theme’s lines are (their share of PageRank, a measure of how much of the graph leads to them) among the lines dated 2023–24 and among those dated 2025–26, and these themes moved furthest:

Rising
LLMs, cognition, and consciousness +9.3 points
Algorithmic entanglement & frottage +4.4 points
Databases, dashes, and dysfunction +4.3 points
Fading
Prometheus, AGI, and Authorship −9.9 points
Scraping Self Against Machines −6.8 points
Recursive material self-becoming −5.1 points
WHAT CHANGEDIdeas that never became work

113 of my 174 ideas have no line linking them to a finished piece. These six have the most other lines grounding, naming, echoing or rhyming with them:

— The Death of Literacy— Studio Chapel From 500 Years In The Future— Site..— Make embeddings of images and map out the images that make up stable diffusion— List everyone I have met since my memories begin.— Directed Emergent Narrative Novel