The armchair created by the hallucinations of AI

Moments is an object, an armchair, designed with the help of Artificial Intelligence. However, the creative process behind its creation by Lashup, a Florence-based studio founded by Claudio Granato and Enrico Pieraccioli, does not employ AI in the conventional sense. Instead, it leverages the limitations of AI, specifically its “hallucinations.” To fully understand what this means, we need to take a step back.

The Mona Lisa with the face of a dog. Pufferfish floating in Van Gogh’s Starry Night. Cats merging together echoing Dali’s style. Exactly ten years ago, DeepDream was created: one of the first artificial intelligence programs specifically designed for artistic purposes, capable of reworking images it is fed in a style that blends the dreamlike with the surreal.

As British artist and researcher Anna Ridler writes, DeepDream − developed by Google in 2014 − “transforms the image to optimize what it sees, enhancing the similarity between what exists and what it perceives: a paradigmatic demonstration of reading what has never been written.”

Therefore, the dreamlike effect produced by DeepDream’s reprocessed images are the result of specific instructions given to the algorithm by its programmers. For example, it might “see” fish in an image where there are none, forcing it to “increase the similarity” between the actual content and what the algorithm wants to see.

Sound familiar? Artificial intelligence’s ability to “enhance similarity” is the digital equivalent of pareidolia − from the Greek èidōlon (“image”) and parà (“beside”) − the illusion that leads humans to perceive familiar shapes in random forms. Pareidolia occurs whenever we see images in clouds, notice faces in wood grain, or animals in damp patches on walls. As Anna Ridler frequently notes, Walter Benjamin refers to this as our mimetic faculty: the human ability to “reproduce and recognize similarity.”

DeepDream was one of the earliest and most significant instances where we began to explore the creativity of artificial intelligence, drawing parallels between its illusory perceptions and our own. Since then, we’ve never stopped. On the contrary, we have continued to investigate this phenomenon and have learned how to leverage it to our advantage.

Just as with pareidolia, we identified phenomena similar to visual hallucinations in machines. According to a study from the University of Padua, visual hallucinations are “phenomenologically very similar to pareidolia” and represent a state of consciousness in which individuals perceive things that have no objective basis.

However, “Artificial intelligence hallucination,” generally refers to something much more specific: instances when a language model like ChatGPT presents incorrect or completely fabricated information as if it were fact. In other words, during the process of statistically reprocessing the information in its dataset, the AI system gets lost, losing sight – so to speak − of the relationship between the initial query, the context, its sources, and the appropriate way to combine all of this. Much like human hallucinations, artificial intelligence “sees” − or at least textually reports − something that does not actually exist, something that is the product of its imagination.

Identifying phenomena like pareidolia or hallucination in machines may be an illusion itself of our own making: an anthropomorphization of mechanisms that have logical, mathematical, and computer-based explanations. It is short-circuit that doesn’t say much about artificial intelligence itself but tells a lot about humanity’s fascination with these systems which leads us to seek out, even within the computations − or simply the errors − of a computer code, the mysterious phenomena of our own minds.

This intricate relationship between humans and machines has been extensively explored for artistic purposes, yielding far more refined results than the intriguing yet conceptually underdeveloped output of DeepDream. This is the case with the works of Refik Anadol, a Turkish-American artist, who, starting 2016, has been creating a series of art projects aptly named “Machine Hallucinations.” His goal is to craft immersive experiences that mimic the perspective of a brain-machine undergoing hallucinations, rooted in the spontaneous and uncontrolled processing of data within its dataset.

The relationship between human beings, machines and the concept of hallucination has also been exploited in the field of architecture and design. One of the most recent cases is precisely the project Moments by Lashup.

Moments is not merely an object, but an armchair that is the result of the collaboration of the designer, the machine, and the hallucinations induced within the image generation system through keywords such as “body,” “narrative,” “relationship,” and “feel,” which are core to the studio’s research. As Lashup explains, “these words, when combined, generate designs that do not always have an immediate place in reality, generating a useful extempore that creates an unexpected element, especially when contextualized in the real dimension.”

Moments “challenges our perception of reality, inviting us to explore the boundary between what is real, what is generated by the human mind and what is produced by the machine.” The outcome is an object that assumes the form of an armchair, or perhaps merely mimics it, evoking once more the phenomenon of pareidolia. It presents itself as a soft, inviting seat to leisurely sink into, yet it may ultimately deceive our expectations. Its surface reflects our distorted image, adding yet another layer of significance.

Moments, recently featured in Milan in an exhibition titled – indeed − “Allucinazione” (Hallucination), highlights and expands on the similarities and differences that emerge in both human and machine creative processes. This underscores how the limitations of artificial intelligence can become not only an area of study, but a creative possibility, to bring the unexpected to life and ultimately enhance our perceptual and imaginative abilities.