GIGO stands for Garbage In Garbage Out, an expression used by programmers to indicate that the quality of a program's outputs depend on its inputs. In terms of AI systems, this can refer to the nature of the model or the training data. Taken more literally, it could refer to content on the Internet, or in the real world, the processed food some people ingest, and the mountains of waste we all produce.

That waste is the subject of Nikolaus Geyrhalter's film Matter Out of Place which was awarded at the 2022 Festival dei Popoli. I recognised the title as a quote from Mary Douglas' classic 1966 anthropology book Purity and Danger.

GIGO in this experiment is more literal. I took the promotional still from the film as a starting point. Kaiber automatically generated a description: 'a group of people standing next to a dump truck filled with trash and other trash and other trash items'. This served as a prompt to generate new content, along with a description of the film from the Festival program: 'a film that oscillates between the hyperrealist and the visionary, the grotesque and sci-fi'.

Image generated using Kaiber

On the surface, these two images are faithful re-creations of the film still – the dumptrucks at right and centre, the pile of trash in the centre, the surrounding hillside are all preserved. The instruction 'a group of people standing' becomes unsettling when they are not obviously workers, when they stare blankly at out of the frame, when they all mysteriously wear hats. In the second image, the worker in the left foreground has been transformed into a bin. Trees beyond the hillside have become highrise buildings.

Runway wins for realism, but not faithful re-creation. The sky has become a cloudless California blue, the colours generally brighter. Trees remain, making the scene less post-apocalypctic, the trash has not accumulated on the hillside as in the other images.

Yet a sense of the unreal remains - the unnatural joins between surfaces, for example, echoes the way Large Language Models can only attempt to replicate true human speech. 'Building models is very different from proclaiming truths,' according to MIT's Neil Gershenfeld. 'It's a never-ending process of discovery and refinement, not a war to win or destination to reach.... Making sense of anything means making models that can predict outcomes and accommodate observations. Truth is a model.' As a machine learning model ‘learns’, 'it is simply learning features at each layer (edges, angles, etc.) and attributing a combination of features to a specific output.' [source]

Putting this process in motion exposes more of the process, its limitations, and perhaps some insights. You can see how the model was trained to faithfully reproduce certain categorisable objects - trucks, power poles, clouds. The essence of truck-ness, pole-ness, cloud-ness that makes such objects recognisable to the human observer. And after all, that is the end result of this model and this process - to produce a simulacrum (cue Baudrillard) of something, which is convincing, believable. If we ignored the essentialising process of categorisation, we would have to regard the mere existence of things, each on its own terms, ineffable and indescribable. The morphed, mis-joined seams and mysterious objects at least hint at this possibility. (Please read Technic and Magic by Federico Campagna)

The process of animation is exposed as a series of images, here at a low rate of 12 frames per second. The scene from Geyrhalter's film becomes a hazy Los Angeles flood control channel, where mysterious figures appear and disappear. Proportions shift. Everyone stands passively, facing away from the 'camera', and when they do face the camera their features are indistinct. An individual becomes a position in a series, a succession of morphing, replacable individuals without identity. We move from the qualities and existence of things to quantities and essences. The same for corporate logos, trash bags of different colours - their contents formerly composed of objects once used and perhaps loved, now they are forms without content.

Here in the part of Florence where I stay, large, carefully designed trash bins reside on every street, larger objects accumulate around them, trucks not unlike the ones seen here arrive regularly and lift them to empty their contents from the bottom. I found it an interesting exercise to watch one of these videos all the way through - you can choose the three or eight minute version - then to go outside and observe the world. Seeing like a machine? Maybe. But observing rhythms, processes, morphings, indistinct objects and perhaps people. Power poles in the distance remind us of grids and movement of resources. Our social and environmental reality of inputs and outputs. When and where do they take on meaning, cease to become 'garbage' or ephemeral forms without content?