tagCloud

architecture arduino art audio audiovisual city classes coding computational creative coding design digital digital media documentary domestication interaction interactive iot localization max media mqtt node p5 performance physical processing programme psychogeography publication robotics sensing sensors sonic arts sonification sound synthesis system tangible teaching technology ubiquity visualization wearable workshop

Let’s Work Together

stavros.didakis@nyu.edu

Logico-Fantastic Machine

A Computational Ecology for the Advancement of Creativity

Art & Technology Research, Interactive Installation

3 Robotic Interfaces, 3 Mobile Phones, 5 OLED Screens, Generative AI
2025

Logico-Fantastic Machine is a cybernetic ecology of robotic and machine learning agents that generate visual and textual narratives with creative characteristics. Three physical kinetic structures scan the interior space and search for original features (semantic, contextual, sentimental). The sensed data are transferred to the generative system, which distills and outputs audiovisual content via custom-trained models. After each generation is created, the system’s agents evaluate it and improve their own logic through repetition, gradually learning on how to better extrapolate on the creative outcomes.

The work is inspired by Italo Calvino’s lecture Cybernetics & Ghosts, where he imagines a machine that generates endless stories, yet remains haunted by the ingenuity of human creativity – echoes of literature, language, and memory. In Logico-Fantastic Machine, this tension between algorithmic logic and imaginative possibility plays out in real time, as the system navigates between computation and invention. Over time, it develops an emergent vocabulary shaped by the interplay of data, vision, and speculation. The output shifts between the recognizable and the uncanny, blending synthetic language, fragmented imagery, and poetic substance. With real-time environmental feedback, Logico-Fantastic Machine rejects static authorship and becomes an open-ended speculative entity, one that is always dreaming, producing and learning.

Koios-3

Koios-3 consists of three robotic interfaces, each featuring a mobile phone with a custom app that scans distinct properties of both physical and virtual environments. The first device, equipped with a computer vision (CV) system and a Large Language Model (LLM), focuses on identifying design features of the space—such as colors, textures, patterns, objects, and other observable physical elements. The second device, also using CV and LLM, targets emotional and sentimental aspects of the environment, aiming to recognize moods, atmospheres, social interactions, and affective states of people present. The third interface displays an endless depthmap that wraps on a virtual 360 degrees space. The rotation of the device triggers the phone to adjust its angle and position, revealing views of the depthmap that always varying spatial depth and complexity. The data that is gathered from these three interfaces – encompassing design characteristics, emotional cues, and compositional dimensionality – is transmitted to the prompt generator that integrates these inputs into instructions for producing the textual and visual narratives of the main engine.

Divergence-Creator

The Divergence-Creator is an autonomous generative AI system that creates high-resolution imagery and corresponding textual narratives depicting divergent architectural and urban environments. Its goal is achieved by continually generating new content that deviates as far as possible from previous outputs, using real-time sensory input from Koios-3 to guide its custom machine-learning models. Each newly generated image features innovative spatial and aesthetic elements, while the narrative text aims to accurately describe these visual features. Right after images and text are created they are automatically assessed by machine agents that measure qualitative and quantitative attributes, such as novelty or technical composition. A physical interface is available for the audience to evaluate the content themselves by adjusting ten sliders for scoring bands (five for text, five for image). The responses are added to the system’s database alongside the machine evaluations, where the remain accessible for further review. The engine’s model uses this growing responses to retrain itself, learning from and incorporating valuable properties especially those deemed most original or well-executed, so as to ensure that future generations display even greater divergent characteristics, while at the same time staying aligned with the system’s core objective.

Obsidian Aperture

The generated images from the Divergence-Creator are automatically assembled into a flowing visual sequence that interpolates between each architectural and urban design, forming a continuous journey through the system’s evolving cityscapes. As the animation is constructed in (almost) real time, it aims on opening a window into the model’s latent space, showcasing the underlying creative processes that drive its dynamic exploration on form, structure, and composition. Calvino envisioned a device capable of endlessly reinventing its own codes to craft new narratives and textual structures. This works, aims to align with this objective and showcasing an emergent system that turns its internal logic upside down, transcending limitations, and discovering creative avenues that extend far beyond its original scope. In doing so, it enlarges the sphere of human imagination – not only by generating unfamiliar stories and forms, but by revealing that even rigorous, rule-bound processes can lead to unknown worlds of possibility.