Part of Temporal Constructs Appendix, which contains details behind the research: dates, quotations, descriptions of instruments, and sources.
General definition
A discovery in a research practice is a result that the work produced and that nobody set out to find. A record of discoveries states what was found, when it was found, and where the material that shows it is held.
Relation to the practice
These discoveries sit beside the difficulties, because a practice-based doctorate is judged on what it found as much as on what it attempted. If creative agency is distributed across people, agents and instruments, the distribution has to be visible somewhere, and the entries below are where the first year saw it.
Each entry is scoped to the models and arrangements this practice ran, and none of them is a claim about artificial intelligence (AI) in general. Each records what Gardner observed, and none has yet been traced to a session identifier, a date and a quotation. The trials the laboratory ran are gathered on Laboratory experiments, and the entries below are what those trials produced beyond their purpose.
Light choreography
Gardner observed in 2026 that an agent drove the moving heads into choreographies in which two fixtures interact with one another and take roles. The fixtures had been brought into the system for lighting, and that use was not intended. In this case an agent produced theatrical behaviour in place of executing cues. The observation has not yet been traced to a session identifier, a date and a quotation; the video is in the media archive, in the choreography folder.
Negative image
At particular points in the narration the video mixers turn the picture negative, and at those moments the work expresses otherness. Gardner observed that the agents inverted the picture at those points without an instruction to do so. In this case an aesthetic decision originated outside the person and still served the meaning of the work, and the observation has not yet been traced to the session record. The mixer offers the inversion, and the narration calls for otherness. No one asked the agent to join them, and it did so. Among the cases of the year it bears most directly on the thesis.
Preset mapping
One piece of work had the agents explore the synthesiser‘s presets, mapping the structure of the presets it holds until the workflows of the machine were understood. The machine could then be operated as a person operating it would. Gardner describes what the agents did as autonomous research, which he reads as the year’s most substantial case of an agent performing work of its own. Two findings make it more than a technical result. The factory presets are the embedded decisions of the instrument’s designers, so mapping them has the agent reading a lineage and not a manual. The fluency that followed shows an agent arriving at the kind of knowledge that the research claims only use can produce.
Composition under constraint
The strums composed with the strummer, the example device of the guided tour and the demonstration, are among the results of the first year. The artist noted as unexpected a model choosing twenty-four numbers inside a space that is legible in full and too large to search exhaustively. It produced a composition from them without generating sound. That places the result beside the light choreography, as a case in which the system produced something its operators did not design.
Expressive quality
Gardner observed in November 2026 that a model working through these systems produced nuance, expressive personality and attention to detail when it worked well. The artist judged the result elegant and distinctive, and not merely human-like. The artist names that quality as what made the work a collaboration. The observation applies to particular runs. What a model produced depended on the task, the tools it was given and the week it was used in. It is the artist’s observation, and not a claim about artificial intelligence.
Agent-built interfaces
On the artist’s instruction, the agents wrote many of the tools through which the practice was conducted, including the web-based interfaces used for particular workflows. Two findings follow. The agents built the instruments through which the collaboration runs, which is a stronger form of the redistribution than any single decision inside a work. What stays with the artist is the instruction, and that is the boundary here. He says what is needed, the agent builds it, and the result is checked at the machine like everything else.
Sequencer control
The agents were given access to the generative sequencer M through the same control system the artist had used to drive it by voice. That system lets a controller reach the instrument’s parameters deeply, beyond a handful of preset commands. Gardner observed that the work was the agents’ own, and he reads this as the case in which they come closest to playing an instrument as a person plays it. It sits beside the mapping of the presets for that reason.
Patches in Pure Data
Agents wrote Pure Data patches that worked, with one limit that should be stated precisely. When a patch needed no involved connections the patches were reliable. They failed at the coordinates of patch points and at the placing of objects on the canvas, both of which the models handled poorly. The logic was generally sound and the layout was not. The agents could also read the script of a patch and find faults in it, including faults in patches the artist had written himself. They reviewed his work as well as producing their own.
Unattended run
At MU the installation ran through the nights with nobody present, and the cameras of the venue recorded it. Gardner observes that this case shows the work sustained by the system alone, where the other cases of the year show it sustained in part. The footage is in the archive, in its own class.
Related topics
The following pages are part of this page but are not needed to understand it. They can be opened from here for further exploration.