Thursday, August 31, 2023

A Photo Testifies

A photograph which looks like a photo of something or someone, as well as anything else which isn't a photo but which looks like such a thing, mainly does one thing: it testifies to that-which-was.

This has been my thesis for a little while now, and it's recognizably lifted directly from Barthes, so if I'm a crank, at least my crankery has a pedigree! What I mean is that a photo, or something that looks like a photo, which also looks like it's of something (not an abstract, obvious collage, or what have you) mainly asserts that something existed, and it looked like that at a moment in time.

There's other shit these things do, of course. They're a mass of tone and color in pleasing, or less pleasing, arrangements, and so on. Paintings do all those things, but paintings do not testify in the same way.

Let us now turn our attention to AI-generated photo-realistic imagery.

It functions in the same way a photo does, if it is sufficiently photo-realistic. It cannot do otherwise. It testifies to that-which-was.

The point is not that it's functioning differently but that its testimony is false.

An unaltered actual photograph cannot be false in the same way. Within the limits of its capacity, its testimony is completely, utterly, true.

The attentive reader might notice here that I am introducing the idea of index in a way that sidesteps the traditional analysis of that concept (light particles physically induced a blah blah blah therefore it's a direct blah blah index index) in order to include digital imagery or whatever. The point is that the testimony is 100% truthful, within the extremely narrow limits of the medium.

To be clear, I am perfectly aware of the many ways a straight photo can misrepresent reality. My point here is that there is a core of visual facts about which no straight photo lies. It looked like that. That thing was in that visual relationship to that other thing. Those two forms overlapped thus. And so on. It is this core of truth that is the testimony of the photo, no more, but also no less.

It is this core of truth that begins to erode the moment we modify the photo (yes, including burning and dodging, contrast adjustments, etc, so yes the core truth of the testimony begins to erode immediately, I am also aware that digital cameras do image processing, thank you.)

An AI generated "photo" testifies in the same way, but its testimony is a complete fabrication.

A perjurer and a priest testify in exactly the same way. The former, however, lies, and we like to imagine that the latter does not.

What is the value of any testimony? Most photos testify as indicated, but nobody cares. Oh, what a nice bowl of tomatos. The light falls just so. Who gives a shit? The aesthetics might be nice, and maybe you even want to decorate your kitchen with a copy of it. But, it doesn't matter if it's real, photoshop, or AI then. So what if the tomatos never existed? Or did? It simply doesn't matter.

Most real photos testify to facts that almost nobody cares about and that don't matter even slightly, to anyone. If we're talking about aesthetics, and if aesthetics is all we care about, then it doesn't matter how the dumb thing got made. Its nature as a piece of testimony doesn't matter a fig, although the fact that it adheres to a photographic aesthetic may.

That said, most real photos testify to something that someone cares about, at least a little. You and I don't care, but to whomever went to the trouble of hauling out her phone, it matters, at least enough to take a photo. It's trivial, but it's real. The photo testifies, and to the photographer, that is in fact what matters. My kid did a cute thing. What a pretty flower. Look at my latte. AI imagery has no place here.

AI imagery only applies to circumstances where we either don't care about the testimony of the image (i.e. Fine Art and Fucking Around, ok maybe Stock) or in places where we explicitly want false testimony. Everyone is focused on the "where we explicitly want false testimony" case because they're worried about things. Let's look at that in a moment.

The point though is that in almost all uses for photography it is the testimony which matters to whoever is taking the photo, albeit to almost nobody else. Nobody looking at an especially pretty flower wants an AI to make an even prettier one, they want to record the one they're looking at. That's literally the point. It's my flower, my child, my town, whatever. If you just want to make a pretty picture of a flower or a child, you could take up painting, and nobody paints.

Almost all uses for AI image-generators that I observe today consist of fucking around and discovering the limits of AI image-generators. The only use case is to post the result online and say "wow, check out what this AI image generator did." This is already starting to get worn out.

As for the case where someone wants false testimony, well. The trouble with false testimony is that as a rule it doesn't work. Nobody accepts any testimony of any kind by itself. Whether we mean to or not, we place testimony in the context of our own world-view, we place it next to other testimony. Even photos, perhaps especially photos: we don't believe testimony unless it supports a larger, more or less coherent, picture of the world.

The only actual use cases for AI imagery that strike me as having any legs at all are basically variations of I wish I could paint, but I can't which honestly seems a bit thin. Not sure there's a big market here.

Tuesday, August 29, 2023

Guest Post: David Smith reviews Someone for Everything

David is a regular commenter here, and a friend. Normally he's substantially more acerbic, which I suspect means something. The book sounds fascinating.

Someone for Everything by Michael LeBlanc — a review
(and a speculative digression)
by David Smith

Michael Leblanc is a visual artist, and a professor of digital design who I studied under in the early '90s. We stayed in touch, trading news and projects. I recently emailed him to ask if he or his colleagues had any truck with AI (he’s unaware of anything specific), and learned of his latest work, Someone for Everything 100 Sequential Drawings


Why would this book of drawings be of interest to aficionados of photography? It manifestly isn't a photobook, but it is a book that includes many photographs altered in ways familiar to those of us who have experimented with collage and other graphical devices. Working with stock photos, Michael has turned these experiments up to eleven.

I was particularly interested in Michael's take on AI, because he's a 'skeptical enthusiast' who introduced Luddite me to the uses of digital technology in visual art. But what, exactly is meant by "AI"? Well before the ChatGPT and Stable Diffusion (etc.) marketing blitz, AI encompassed many, less hyped applications that don't depend on the extravagant resources of deep fake manufacture.

"Before we go any further, let’s get one thing out of the way."

In his text, Michael takes considerable pains to make clear the 100 drawings were not AI-generated. Based on his emailed comments, I think it possible they were, at least partly, AI-inspired — if true, an interesting conceptual switch! Indeed, the methodology he lays out in detail sounds a lot like the AI process of assembling and modifying fragments of found imagery into a composition — a process historically known as collage.

To recap what is widely understood about the collage medium: one starts with source images clipped from newspapers, magazines, and other printed ephemera (other materials may be introduced). The images may be close-cropped to particular subjects/details, or (less commonly) they are included 'as is,' and glued down in a arrangement. In the digital era, many artists have adopted the collage method with scanned source and/or digicam images. There have been bitter accusations of plagiarism leveled against some AI image-generators, but this is a core precedent for the process itself.

If one takes the narrow position that a drawing is mainly comprised of brush or stylus marks made by human hand, the works in Michael's book look more like collage than drawing at first glance. Closer inspection reveals another possibility: the collage elements are a matrix on which drawing is overlaid. This doesn't take anything away from the book’s expressive power, but I feel it is important for understanding its conceptual provenance.

In his text, Michael describes his intentions for the series, reaching back to his practice of traditional printmaking (intaglio, lithography, etc.): maximal tonal range, 'richness,' and the idea of successive proofs in the development of a print. I'm going to say here I wish I had looked through the drawings first (my bad, I should have just skipped ahead). I've lately become interested in 'prompts' (titles, captions, and texts) — how they affect our readings of visual art. While I think such texts may ideally provide some insights into what a visual work might mean and why, I much prefer to absorb the information visually, and make up my own mind first. I strongly feel the proper role of visual art is to mystify and delight, rather than instruct (Michael is a professor, so…).

The 100 drawings

The sequence begins with Someone for Everything I, a charcoal drawing of a central figure delineated in, and surrounded by blocky tones. The scanned drawing is a scaffold for ensuing digital overlays of collage and drawn elements that shift, morph, are replicated and replaced. It may also be taken as a signal the work is to be interpreted as a series of drawings — something that might otherwise be overlooked! Compared to what soon follows, the composition is relatively simple and stylized, which makes it easier to spot superimposed drawing in later iterations.

The next few drawings in the sequence build up complexity and depth ("richness"). By the fourth (Someone for Everything IV), Michael has got his vocabulary, but the best is yet to come.

Some of the details feel 'off,' in a way that is uncannily similar to how AI image generators conjoin source fragments by attempting to conjure up the missing bits from the sparser reaches of a database. One obtains the strong sense that Michael isn't working to cross an uncanny valley to a seamless realism, but he is deeply interested in the surreal glitches and artifacts AI throws up in near-misses and abject failures. This tracks to his previous works on technology. I have to say again, that flipping this phenomenon into a basis for human invention is an impressive conceptual feat (cf. Francis Picabia’s Réveil Matin).

Michael repaints the collaged photographs, masking and fading elements, and sketching in new details in a kind of loose impressionism, as he builds each drawing.

Paging back and forth through the drawings, I am immediately struck by how they function together as a flip-book animation. This bridges the variable 'completeness' or polish of the individual drawings. While many of the drawings work very well as stand alone visual statements (and I'd like to see a tighter edit with these in), the sequence benefits from the inclusion of intermediary states in comprehending the greater whole.

Michael's collage sources include early Soviet photographs, and I think this is a good place to bring up what I see as possible conscious or unconscious influences on the overall direction of the series: Alexander Rodchenko and other Constructivists worked extensively with dynamic collage compositions during this era, along with Germans in the Dada movement (Hannah Hoch, Raoul Housmann).

Most of Michael's figures here are similarly poised for action, and are often rotated or flipped to an unstable angle for heightened dynamic tension. They are combined and recombined with other figures, whom they seem to be dancing with, fighting, or otherwise spatially interfering. Many of the figures have acquired exaggerated expressions through Michael’s overlaid drawing. Others appear as bland, anonymous ciphers. Drawn elements keep things in the frame, by setting up pauses in the action, and routing our attention around the composition.

The narrowly vertical (6:10) aspect ratio of the drawings perfectly frames a straight-on shot of a standing person with ample margins, and many of the drawings are of this configuration, often with limbs and other visual elements flying out at odd angles. The format, and placement on the page spread support both the 'book of drawings' identity in terms of traditional presentation, and the flip-book browsing experience — variations (and there are many) are locked into this layout scheme.

Michael alludes in his text to the possibility of the drawings printed large scale for an exhibition of the white-cube persuasion. It is easy to imagine LXIV, LXV, LXXV and LXXXI (among others) making for a very dramatic show indeed. He also states that he now considers the project a sketchbook of studies for (e.g.) large paintings.

I feel fortunate to have learned of this very interesting project, which has great depth and many facets of interest to me personally. I will be returning to this book as I try to unravel its meanings, and absorb its lessons.