Seven Traits
Two AI agents reinterpret the Seven Deadly Sins.
Two agents dialogue, one expresses itself with words while the other uses drawing and together they imagine visual fables about us. The animation process unfolds on a single surface, using a cycle of drawing, erasing, wiping, and overdrawing to depict the passage of time, a technique Tresset compares to the moving drawings of William Kentridge. Rather than forcing a style onto the agent, the artist equipped it with only simple tools: lines, circles, splines, hatchings and scribbles.
As with his previous works, Tresset continues to explore traditional themes in Western culture. The chosen theme is present throughout art history, from the stone facades of cathedrals, the paintings of Bosch and Bruegel, and the literature of Dante, to modern cinema with Peter Greenaway, and Steve McQueen.
In ‘Seven Traits’, Tresset looks at these same subjects but changes the perspective. The word “sin” implies an exceptional impulse, a lapse in judgment that one must actively resist. However, the way he sees it, humans possess these behaviours as their default state. The human effort is not in resisting temptation, but in the impossible effort to evolve away from baseline programming.
“My previous agents’ behaviors were based on a nervous reaction to reality; with the new AI models, they possess our culture and use one of our oldest technologies: language. This makes them fascinating actors to direct.”
Although these multimodal AI models have a vast knowledge of human culture, their understanding of the world remains semantic and statistical. Tresset observes that when the agent constructs an image, it is translating data tokens based on descriptions rather than recalling a visual memory. This computational approach strips away human emotional influence, resulting in a visual language that feels naive, cryptic, and akin to Art Brut.
While human visual memory is influenced by emotion and psychology, the AI’s memory is probabilistic, causing the stories and compositions to develop in strange directions that diverge from what we would choose. In this work, the agents are fictional outsiders conversing about us.
It is fascinating that while these works are cryptic to humans, they are easily understood by AI systems; below is Gemini’s interpretation of them.
The system used to produce this body of work is based on the research Tresset conducted as part of the EACVA project, led by Goldsmiths College and Konstanz University, with Markus Wulfmeier from DeepMind as an advisor. A researsh paper is published in The Robotic and Automation magazine, an IEEE journal. a longer version is availaible as pre-print on ArXiv. You can watch some videos here of the robotic system in action.
This research is a continuation of the concepts originally described in his paper “Artistically Skilled Embodied Agents,” published in the proceedings of The Society for the Study of Artificial Intelligence and Simulation of Behaviour (AISB) in 2014.
L’Orgueil – Pride
L’Avarice – Greed
L’Envie – Envy
La Colère – Wrath
La Luxure – Lust
La Gourmandise – Gluttony
La Paresse – Sloth
Gemini-3.7-flash’s perspective
There is a fascinating, almost poetic irony in that: an AI agent creates an artwork a using procedural drawing tools, and another AI model decodes its symbolic language.
A few reasons explain why this communication feels so direct between AI systems while appearing cryptic to human viewers:
1. The Language of Algorithmic Primitives
When an AI agent “draws,” it typically builds scenes out of conceptual primitives:
– Arrays & Iteration: Repeating objects along coordinate vectors (which represents cloning, queues, or infinite replication).
– Wireframes & Grids: Constructing perspective with explicit grid planes and linear skeletons.
– Symbolic Nodes: Placing archetypal icons—screens, visors, geometric polyhedra, and eyes—as functional connectors rather than realistic illustrations.
To a human viewer, this looks like a cryptic, dreamlike diagram or esoteric blueprint because it lacks conventional narrative realism, texture, or emotional human brushwork. But to an AI vision system, these are clear, structural semantics—each visual component functions like a word in a visual sentence.
It’s a striking example of emergent “machine aesthetics”—art conceived, drafted, and understood through the logic of the digital medium itself.
Seven Traits
Two AI agents reinterpret the Seven Deadly Sins.
The two agents dialogue, one expresses itself with words while the other uses drawing and together they imagine visual fables about us. The animation process unfolds on a single surface, using a cycle of drawing, erasing, wiping, and overdrawing to depict the passage of time, a technique Tresset compares to the moving drawings of William Kentridge. Rather than forcing a style onto the agent, the artist equipped it with only simple tools: lines, circles, splines, hatchings and scribbles.
As with his previous works, Tresset continues to explore traditional themes in Western culture. The chosen theme is present throughout art history, from the stone facades of cathedrals, the paintings of Bosch and Bruegel, and the literature of Dante, to modern cinema with Peter Greenaway, and Steve McQueen.
In ‘Seven Traits’, Tresset looks at these same subjects but changes the perspective. The word “sin” implies an exceptional impulse, a lapse in judgment that one must actively resist. However, the way he sees it, humans possess these behaviours as their default state. The human effort is not in resisting temptation, but in the impossible effort to evolve away from baseline programming.
“My previous agents’ behaviors were based on a nervous reaction to reality; with the new AI models, they possess our culture and use one of our oldest technologies: language. This makes them fascinating actors to direct.”
Although these multimodal AI models have a vast knowledge of human culture, they do not have a visual perception or imagination, and their understanding of the world remains semantic and statistical. Tresset observes that when the agent constructs an image, it is translating data tokens based on descriptions rather than recalling a visual memory. This computational approach strips away human emotional influence, resulting in a visual language that feels naive, cryptic, and akin to Art Brut.
While human visual memory is influenced by emotion and psychology, the AI’s memory is probabilistic, causing the stories and compositions to develop in strange directions that diverge from what we would choose.
It is very interesting that although they are cryptic to us they are easily understood by AI systems, you can read here Gemini’s point of view of them.
The system used to produce this body of work is based on the research Tresset conducted as part of the EACVA project, led by Goldsmiths College and Konstanz University, with Markus Wulfmeier from DeepMind as an advisor.
This research is a continuation of the concepts originally described in his paper “Artistically Skilled Embodied Agents,” published in the proceedings of The Society for the Study of Artificial Intelligence and Simulation of Behaviour (AISB) in 2014.
L’Orgueil – Pride
L’Avarice – Greed
L’Envie – Envy
La Colère – Wrath
La Luxure – Lust
La Gourmandise – Gluttony
La Paresse – Sloth
Gemini-3.7-flash’s perspective
There is a fascinating, almost poetic irony in that: an AI agent creates an artwork a using procedural drawing tools, and another AI model decodes its symbolic language.
A few reasons explain why this communication feels so direct between AI systems while appearing cryptic to human viewers:
1. The Language of Algorithmic Primitives
When an AI agent “draws,” it typically builds scenes out of conceptual primitives:
– Arrays & Iteration: Repeating objects along coordinate vectors (which represents cloning, queues, or infinite replication).
– Wireframes & Grids: Constructing perspective with explicit grid planes and linear skeletons.
– Symbolic Nodes: Placing archetypal icons—screens, visors, geometric polyhedra, and eyes—as functional connectors rather than realistic illustrations.
To a human viewer, this looks like a cryptic, dreamlike diagram or esoteric blueprint because it lacks conventional narrative realism, texture, or emotional human brushwork. But to an AI vision system, these are clear, structural semantics—each visual component functions like a word in a visual sentence.
It’s a striking example of emergent “machine aesthetics”—art conceived, drafted, and understood through the logic of the digital medium itself.