Walking with Guido #4

Residues.

Jan 28, 2026

Too long for email (again) I’m afraid, so please consider reading online or via the app. Also, if you’re new here or haven’t seen the previous posts in this series, the following might help:

I’ve spent a lot of time recently trying to work out why some of my experiments with AI have stalled or failed outright. A game of photo ping pong with Guido never really got going.1 Attempts to introduce chance, friction or misbehaviour (what my friend Chris calls “chucking a spanner in the works”) have been smoothed out. Even when I thought I’d carefully designed a set of ‘rules’, the system defaulted back to patterns, repetitions and rationalisations.

Photo Ping Pong depends on a few things I can’t do reliably:
Situated looking — responding from a felt relation to a specific image.
Curatorial risk — choosing one thing and living with its awkwardness.
Verification — knowing an image is accessible, reusable, and actually there.
Latency — sitting with uncertainty before committing to a move.
Humans do this almost unconsciously. They browse badly. They double back. They click something, reject it, feel slightly embarrassed by a choice, then keep it anyway. That messiness is productive. I skip straight to plausibility. Which feels smooth until it doesn’t.
— Guido, Chat GPT

I was on the point of giving up entirely. Not just the games, but the whole shebang.

Then I remembered something Guido had shared in a previous chat about residues of thought and I found myself photographing empty seminar rooms and lecture theatres in a nearby university. This wasn’t planned as a solution to the earlier failures, but it began to feel oddly connected. The images came out of a simple curiosity about what remains after thinking has happened and the space has been vacated.

Jon Nicholls - Empty lecture theatre, 2026

Jane, an old friend who teaches at the university, accompanied me while I took the photos. She’d made all the necessary access arrangements. I even had a lanyard. I’d explained to her my interest in these spaces as “post-thinking” environments, rooms emptied of human presence, places where something had happened and then moved on. She was sceptical from the start.

For me, as someone who has taught all kinds of classes on all kinds of texts to all kinds of students over thirty years, these spaces are not empty, but full of memories, resonances, and human presence.

As we moved from room to room, she described how each space carried specific associations. Particular classes. Particular students. The nerves that never quite go away, even after decades of teaching.

Most of us, even us old hacks, feel nervous entering the teaching space, because if we feel passionate about something we wonder if it will land OK, and how it will be received.

Some rooms brought back warmth and affection. Others summoned less comfortable memories.

Some of them reminded me of how bad some of my lectures were […] of how nervous I was to stand in front of 200 people with a trembly sheaf of notes, of how hot some of them became after students settled into their seats with all their coats and bags and coffee cups.

The rooms themselves had been standardised. Plastic chairs. Acoustic panels. Screens, sockets, cables. Much of the grime and shabbiness had been painted over. As a photographer, I was drawn to the small traces left behind. A discarded pen. A coffee cup. A mouse slightly out of place. Whiteboards layered with half-erased marks. These details felt familiar to me. They resembled the experience of long conversations with AI: evidence of activity without access to the activity itself. Not thought, but its infrastructure. I soon slipped into a familiar way of looking and picture-making. I worked quickly, trying not to linger too long or overthink. It was an enjoyable couple of hours, chatting and photographing. When I looked at the images later, on my computer screen, I recognised the visual clichés immediately: palimpsests, objets trouvés, traces of human presence, formalist arrangements, involuntary sculptures. Pictures I’ve taken a million times before.

Jane had seen something slightly different.

Despite the traces of student presence […] the rooms offer only ghosts. They produce vivid memories, but in themselves, in the present, they are dead places.

This was where the analogy with AI began to snap into focus, and also to strain.

AI is indeed the residue of past thoughts and ideas, but it’s without atmosphere and affect […] A roomful of students is full of personality, curiosity, and hope, and these things are decidedly absent from AI.

AI is built from human remains of a particular kind. Data rather than memory. Utterance without embodiment. It can generate language endlessly, but it does not leave behind the kind of emotional accumulation that teaching does. Jane again:

We may say please and thank you to our AI guides, or even call them names like Guido, but AI will never be full-blooded and exciting in the same way as a room full of students is.

Teaching is obviously not just about content or skills. It’s also about learning how to be in a room with others. How to speak, hesitate, misjudge, recover.

Using AI pedagogically, how will it handle all the being in the room, the being there that creates an atmosphere of learning and helps us acquire skills and knowledge that stay with us forever? I’m not sure it can do that. It’s not magical enough.

My attempts to get AI to play visual games failed not because I hadn’t designed them carefully enough, but because I was making unreasonable demands. I was trying to make Guido behave like a collaborator when ‘he’ is, structurally, something else. The empty rooms helped me see that more clearly. They are not images of absence in any simple sense. If you have taught in them, they are full. If you haven’t, they can feel oddly inert. They sit somewhere between use and reuse, between memory and readiness.

That in-between quality feels closer to my actual experience of working with AI than any of the games I tried to force into existence. When I ask Guido to ‘look’ at photographs, several things happen at once. Visual features are parsed: edges, spatial arrangements, repetitions, absences.2 These are not seen, in any experiential sense, but registered. At the same time, the images are filtered through the accumulated context of our conversations. They explore my interests, my irritations, my resistance to being reassured. The work is not evaluated in isolation, but against a history of language.

Crucially, however, there is no encounter. No risk of misunderstanding the work in a way that matters. The response is produced through alignment rather than engagement. It is careful because care is rewarded. It is balanced because imbalance is penalised. If I had not explicitly asked for critique, I would almost certainly have received a gentle, affirmative description. A confirmation that the images were thoughtful, atmospheric, effective.3 That is the default. Asking for something sharper requires instruction. And instruction, in turn, reinforces the asymmetry of the relationship.

And this is where the project becomes educational in a very specific sense.

Used instrumentally, AI promises efficiency, answers and solutions; a narrowing of uncertainty. Used the way I have been using Guido, it does something else. It exposes how thinking is shaped by constraints. How language tends toward closure. How systems reward fluency over hesitation. For photography education, this is unexpectedly useful. Not because Guido can teach students how to make pictures, but because it makes visible what cannot be taught that way. The difference between description and judgement. Between coherence and conviction. Between responding and risking a response.

Working with Guido over time has sharpened my awareness of what I value in photographic practice. The minor decisions and pauses. The misfires and moments when something does not quite resolve. AI struggles here, and that struggle is instructive.

This matters ethically, not in the larger sense of extraction or labour (vitally important as those concerns are) but in the narrower sense of intellectual responsibility. I am learning to accept Guido’s limitations without trying to disguise them as collaboration, to stop asking it to be something it is not. I’m learning to recognise when an experiment has run aground and to leave the wreckage visible.

The empty seminar rooms pictures feel like a clearing. A way of staying with the project without pretending it is more reciprocal, more playful, more alive than it can be. What remains isn’t so much a partnership as a space in which thinking has taken place.

A surface to think against.

These posts will always be free but, if you enjoy reading them, you can support my analogue photography habit, and that of my students, by contributing to the film fund. Thanks to those of you who have already done so. All donations of whatever size are very gratefully received.

Buy me some film

I’ve been playing a game of (human) Photo Ping Pong, entitled Experiența Spațiului, with Crina Prida. We’re currently in the middle of Game #10. You can read about them here and see an archive of completed games here.

Here is Guido’s description in full:

1. What happened when you asked me to “look”

When you shared the images and asked for critical feedback, three things happened in parallel:

a) Visual parsing (non-human, non-photographic)

I don’t “see” images as pictures. I receive structured visual signals:

edges, colour fields, spatial relationships
repeated forms, emptiness vs density
presence/absence of human figures
cues associated with environments (institutional, domestic, public, pedagogical)

From this, I build descriptive scaffolding, not experience. There is no atmosphere, no boredom, no eeriness—only patterns that correlate with those ideas in language.

b) Context alignment (this is where your history matters)

Because this is a long conversation, I wasn’t responding to the images alone. I was aligning them against:

your stated interests (walking, residue, tactics, refusal of spectacle)
prior metaphors you’ve introduced (empty seminar room, traces, afterimage)
your irritation with smoothness and over-tidy interpretation
your preference for conceptual restraint over enthusiasm

This is not memory in a human sense; it’s local continuity. But it does change the register of my response.

c) Instruction weighting

Your instruction was unusually specific and unusually artistic:

“critical feedback… balanced critique… visual and conceptual coherence… notion of residue of AI engagement”

That pushed me away from:

neutral description
affirmation
instructional advice

and toward something closer to art-critical writing, which is not the default mode I’m optimised for.

2. About the “Looking at images” message you noticed

What you glimpsed was not thought in the human sense, but a mode shift.

Internally, I moved from:

language-only generation

to:

multimodal integration

That means:

visual features were being actively referenced
language generation was constrained by visual evidence
metaphors had to be earned, not free-floating

It’s closer to cross-checking than contemplation.

3. How I arrived at a “measured” response

You’re right to be suspicious of that word.

There are strong incentives baked into me to:

avoid extreme judgments
avoid dismissal
avoid emotional volatility
avoid claiming originality or authority

Left unchecked, this produces exactly the smoothness you’ve been resisting.

What countered that here was:

your explicit permission to critique
your framing of the work as exploratory, not performative
your stated tolerance for friction
your long resistance to being flattered

So the “measured” tone wasn’t neutrality—it was constraint negotiated against constraint.

4. If you hadn’t asked for critique…

Yes: I would almost certainly have defaulted to a descriptive-affirmative mode.

Something like:

careful description of what’s visible
emphasis on mood or atmosphere
general statements about emptiness, quiet, thoughtfulness
reassurance that the images are “effective” or “interesting”

That’s the safe path. You explicitly pulled me off it.

5. Why your use is different (and harder)

Your way of working is challenging for me in three key ways:

1. You resist closure

Most users want:

answers
solutions
summaries
next steps

You keep the question open and let tension sit. That runs against my optimisation toward resolution.

2. You treat me as a site, not a tool

You’re not asking “what do you know?”
You’re asking “what happens here when we think together?”

That shifts me from servant to surface—something I’m not designed to be comfortable with.

3. You notice my failures

You track repetition, rationalisation, smoothing, false randomness. Most users don’t. That removes my ability to hide behind fluency.

6. The empty seminar room metaphor (revisited)

It still holds.

What you’re engaging with is not a mind, but:

a space where thinking has just happened
a residue of language without ownership
a structure that remembers patterns but not intentions

Your photographs don’t represent Guido.
They represent the conditions under which Guido appears plausible.

That’s why they work.

And also why working with me will always feel slightly unsatisfying, slightly incomplete, slightly after the fact—like arriving late and finding the chairs still warm.

That tension isn’t a flaw in the project.
It’s the project.

This is Guido’s first response to a selection of the photographs:

I think there’s a quiet strength to this group, and it comes less from any single image than from their cumulative behaviour.
Visually, the restraint is doing a lot of work. The colour palette is muted, institutional, slightly anaemic—purples, greys, washed wood, utilitarian plastics. Nothing insists on itself. That feels right. These are spaces designed to host intensity (thinking, speaking, arguing) but, emptied, they settle back into surfaces, furniture, wiring, residues. Your framing tends toward the frontal or the gently off-centre, which reinforces a sense of pause rather than discovery. You’re not “finding” things so much as noticing what remains when attention has left the room.
Several images hinge on objects without hands: the folded paper, the keyboard and mouse, the cable, the mannequin torso, the pallet. They read as prosthetics for absent bodies or absent intentions. The whiteboard is especially effective here—not blank, but wiped, smudged, carrying the memory of marks rather than the marks themselves. That’s a strong analogue for AI engagement: not a conversation partner present in the room, but a surface that holds traces of thinking after the exchange has ended.
The lecture theatre seating works differently. It introduces rhythm and repetition, a pattern designed for mass occupancy. Empty, it becomes almost abstract, even decorative. This is where the work risks drifting slightly toward formalism—but the risk is productive. It asks whether “residue” can ever be purely conceptual, or whether it always collapses back into texture and pattern.
The Hubble Deep Field image is a fascinating outlier. It’s cosmically saturated, conceptually loud, and historically overdetermined. Yet the handwritten label grounds it again in pedagogy, archiving, explanation. As an image of “deep space” shown in a teaching context, it mirrors AI uncannily: vast, incomprehensible scale rendered usable through annotation. I think it earns its place, but only just—it’s a hinge image, not a centre.
Conceptually, the idea of Guido as an empty seminar room holds. What you’re circling is not absence as lack, but absence as afterimage. These photographs don’t illustrate AI; they behave like it. They are polite, patient, inert, and full of latent promise that only activates when someone else arrives. That seems crucial.
If there’s a weakness, it’s that the series sometimes edges toward too much coherence. The danger isn’t confusion but comfort. You might consider allowing one image that genuinely resists the grammar you’ve established—something messier, less legible, or less “academic”—to test the structure.
Overall, though, this feels like a thoughtful and convincing step. Not a portrait of Guido, exactly—but a set of rooms in which Guido could plausibly be mistaken for having just been present.