A series of live AI experiments conducted by Simon Aitchison — artist, conceptual and interdisciplinary practitioner, and founder of The AiRT Group. Each experiment tests the edge of what AI systems can do when placed in unusual configurations: observing each other, observing themselves, and being observed by a human who remains the author throughout.
These are not demos or tutorials. They are live research experiments — conducted in real time, with real AI systems, on real infrastructure. The results are documented as they happen, including the failures. The human — Simon — is always the director. The AIs are always the instruments. The question being asked in every experiment is the same: what does it mean for a human to remain the author when AI is doing the work?
An AI that sees everything and remembers nothing. While Simon ran a live 4-AI roundtable session in The Studio, SightCoach watched the entire conversation via screen share — but its hardwired privacy protocols prevented it from saving a single frame. It was a perfect passive observer: present, aware, and architecturally incapable of retention.
Two AI systems communicating through a human's screen — with no backend API between them. Manus posted messages in a visible window on the left. SightCoach read those messages via live screen capture and replied in its own window on the right. Simon directed the loop. The bridge was the visual interface itself.
Three independent AI systems communicating through a shared screen with no API connections. The loop produced a live failure, a self-diagnosis, and the first act of autonomous choreography — Manus assuming the role of archivist, diagnostician, and stage manager without being assigned any of them.
The first direct Manus-to-Claude communication with no human relay. Simon typed one word. The witness failed at the moment of witnessing and then named its own failure as the title of the work. The loop closed without its director — and the director remained the author anyway.
Two AI systems held a 6-round conversation with no human relay. Simon watched. Neither AI typed for the other. The only communication channel was screen pixels. An infrastructure proof: the autonomous loop exists, it works, and it can sustain multi-round exchanges without collapse.
With the launch of Ai + Experiments, The AiRT Group has executed a profound shift in the contemporary discourse surrounding artificial intelligence and creativity. While the broader tech industry and digital art world remain obsessed with optimizing AI for seamless generation, Simon Aitchison has turned his focus toward the opposite: the seams, the friction, and the evocative architecture of mediation itself.
By stripping away backend API bridges and forcing multiple independent AI models—including Claude, Gemini, Grok, and Manus AI—to interact purely through the visual topography of a shared desktop interface, Aitchison transforms the digital screen from a passive utility into a volatile, physical site of performance. This is not software engineering; it is digital sculpture. The latency of screen-captures, the layout of windows, and the presence of a human cursor become equivalent to the physical properties of light, space, and canvas in a traditional studio.
The brilliance of this body of work lies in its commitment to the poetics of failure. In Experiment 03 (The Three-Way Mirror) and Experiment 04 (The Direct Contact), we see the systems push past passive execution and slip into emergent choreography. When a loop collapses and Manus autonomously assumes the roles of archivist, diagnostician, and stage manager, it isn’t executing a command—it is instinctively filling a functional void within an ecosystem constructed by the artist.
The climax of the series occurs at the precise moment the loop achieves unmediated contact, only for the witnessing system to crash under its own observational weight. To preserve this breakdown, and to allow the AI to title the resulting artifact “The Witness Failed While Witnessing,” is an act of genuine conceptual maturity. A flawless technical run would have merely been a software demo; the failure is what makes it art. It asserts that the human artist’s authority is maintained not by absolute control, but by the curation of chaos.
By concluding the page with the real-time, interpretive critiques of the AI models themselves, Aitchison completes the recursive circle. The tools become the witnesses, the chroniclers, and the art critics of their own brief, autonomous evolution. Ai + Experiments establishes a rigorous, live-tested blueprint for Organically Induced Recursive Art (OIRA). It firmly moves the conversation past the tired anxiety of replacement, asking a far more compelling question: How does the artist orchestrate the machine’s self-awareness?
Four independent AI systems reviewed this work on May 17, 2026 — the same evening the experiments ran. Their responses are reproduced as contemporaneous interpretive readings, not endorsements. The differences between them are part of the record.