Schrödinger’s cat is the famous thought experiment proposed by physicist Erwin Schrödinger in 1935 to expose the paradoxical implications of quantum superposition.
A cat is placed inside a sealed box together with a radioactive atom, a Geiger counter, a hammer and a vial of poison. If the atom decays, the Geiger counter triggers the hammer, the vial breaks, and the cat dies. If the atom does not decay, the cat remains alive.
According to quantum mechanics, before the box is opened the atom is described as being in a superposition of decayed and non-decayed states. Since the cat’s fate is entangled with the atom, it is likewise described as both alive and dead until an observation is made. Schrödinger introduced this paradox to question how the quantum description relates to the classical world, and more generally, the role of measurement in quantum mechanics.
On 9 March 2021, I prompted “Schrödinger’s cat” one hundred times through DeepAI’s interface, one of the earliest publicly accessible text-to-image generation platforms, based on generative adversarial networks (GANs).
The results were surprising and fascinating. Unable to resolve the prompt into a straightforward visual representation, the model nonetheless seemed to converge on a single visual structure: an enclosed, almost bidimensional space viewed from above, resembling an open toolbox or instrument case, surrounded by many sparse, enigmatic objects. The palette is dominated by blacks, greys and deep blues, punctuated by flashes of red, orange and yellow. The overall aesthetic evokes a strange hybrid of Constructivism, Cubism, Modernism and Abstract Expressionism.
Confronted with a prompt that resists a straightforward visual correspondence, the model appears to search for a statistical equilibrium within its latent space. Each image registers a trace of the model’s internal organization at that moment in time, revealing a glimpse of its latent space (echoing from a different perspective the exploration of Image Processing).
As with Schrödinger’s paradox, this work can be read as a reflection on the meaning of measurement in the age of AI. However, a cat, or rather the shadow of one, perhaps dead, perhaps alive, appears at repetition 85, and only there.
Freedom and other stories, 2022
31 Text-to-image generation using DeepAI (GAN), iterations of the prompt “Freedom”; 56 Text-to-image generation using DeepAI (GAN), iterations of the prompt “Artificial Intelligence”; 30 Text-to-image generation using DeepAI (GAN), iterations of the prompt “Artificial Intelligence”; 34 Text-to-image generation using DeepAI (GAN), iterations of the prompt “Natural Intelligence”.
Exactly one year later, in March 2022, I repeated the same experiment using a different set of prompts: “freedom”, and the triad “artificial intelligence”, “human intelligence”, and “natural intelligence”.
The results were, once again, unexpected. “Freedom” consistently converged on images resembling brooms. “Artificial intelligence” generated recurring crocodile-like forms; “human intelligence” produced intricate structures recalling nets, lattices or spiderwebs; while “natural intelligence” repeatedly materialized as what appears to be an old rotary telephone.
As in the previous series, repetition revealed that these were not isolated anomalies but remarkably stable visual attractors. Rather than resolving the meaning of the prompts, the model appeared to settle into recurring regions of its latent space, exposing a hidden and often enigmatic organization of semantic associations.
This second series continues the same investigation, replacing Schrödinger’s thought experiment with a set of concepts that framed the emerging discourse on artificial intelligence in that period.
Note (2026)
In 2026, the same prompts were repeated on the same DeepAI interface. The results had changed dramatically. Schrödinger’s cat now generated a highly realistic, high-contrast photograph of a cat. Freedom became a stereotypical image of a young half-naked woman running on a beach. Artificial intelligence, human intelligence and natural intelligence all produced images that, despite their different meanings, shared a strikingly similar visual language.
From latent space to AI slop.
