Part of Temporal Constructs Appendix, which contains details behind the research: dates, quotations, descriptions of instruments, and sources.

General definition

The view that the layer translating between systems is itself a place where research happens.

Relation to the practice

Magnusson (2009) treats the instrument’s interface as the place where musical knowledge is formed. The term used here names that place as the site of the research.

The translation layers of the professional doctorate, such as its command language, are both research site and infrastructure. Both readings stand. The layers support the work, and in doing so embody an argument about how humans and artificial intelligence are to meet.

Unforeseen

Here the unforeseen situations, sometimes difficult, sometimes unexpectedly interesting, will be described as the project runs. Many of these unforeseen situations have already occurred and have been recorded. They will be added in the time to come.

Evidential basis

CUE is the case. It is the infrastructure the works run on. It is also where the research states how a person and a model are to meet. Nothing is hidden from either, and the vocabulary is kept deliberately small, so small models can use it well. CONDUCTOR’s own construction makes the same argument, with every function exposed so any player can operate the same instrument. The claim is thin in its second half, that the interface is a research site. What has been learned from designing it is not written down here as a finding.

CUE design finding

Designing CUE produced the finding that the section above says is missing. The finding concerns two senses of the word language. The first sense is what the model is made of, text it was trained on and text it predicts. The second is language used as a control surface, a bounded vocabulary with defined rules, of which CUE is the case.

CUE is deliberately both at once. Its phrasing is speech, so a person and an agent address it in the same way. Its vocabulary is as small as a protocol’s. The rules of interaction are fixed, and a small model can keep them.

Every earlier route into a complex system needed a formal language before anything could be reached. Such a language was a protocol, a controller map or HMSL. Language addressed to a model now reaches the same system without that step, and the agent takes the request into the protocol. That settles what the interface is a site of.

The finding here takes the form of a measurement: what of a complex system can be reached in language alone, and what cannot. Reached through CUE are the modules, the snapshots, the presets and the timeline. Out of reach in language are the things a hand judges: the feel of a fader, the timing a performer notices before it can be described. The boundary between those two reaches is redrawn each time an interface is designed. The layer is therefore a research site as much as infrastructure.