Chosen Colours: Drone Aura Fields as a Conversational Register, from 1,120 Observed Interactions on Chiark Orbital and GSV Quiet Enthusiasm, 1742–1744 CE

1. Introduction

Most drones have no face. A drone shows its feelings, when it chooses to, through a field of coloured light around its body, the aura. Humans and drones in the Culture learn to read auras early and read them constantly, and a drone's colours are part of how its friends know it. Yet the aura has usually been described, in drone studies as in ordinary speech, as a display of state: a drone is annoyed, and its field shows it (Kerrow, 1690 CE).

That description carries an assumption. If an aura shows state, it resembles a human blush, which appears whether or not its owner wants it to. Drones themselves have long objected to this picture (Veskel-Anthrasy, 1728 CE), and many report that they can dim, alter or suppress their fields; the generators in current drone bodies are under the drone's own direction (Olk-Vaskett, 1710 CE). How far that control is exercised in ordinary company is a separate question, and it is the one this study tests.

The question matters beyond semiotics. Drones are members of the Culture with the same standing as biological citizens. A reading of auras as involuntary readouts tends, however politely, to treat drones as machines whose insides are on show. A reading of auras as a register of communication treats them as speakers.

One of us is a drone, and has argued in print that auras are chosen (Veskel-Anthrasy, 1728 CE). We regard that as an advantage in gaining the trust of participants and a risk in interpretation, and the design below separates coding from interviewing for that reason.

2. Field Setting and Method

Fieldwork took place between 1742 and 1744 CE in two settings: public spaces and households on Chiark Orbital, and the common areas of the GSV Quiet Enthusiasm. We approached 81 drones known to the authors or introduced by participants, and 64 agreed to take part (79.0%). Consent covered recording, coding and a short interview after each recorded episode (Chiark Orbital Hub Records, 1742–1744 CE). Humans present in recorded episodes consented separately or were excluded.

An episode was any interaction in which a participating drone produced a visible change of aura while addressing a human or another drone. We recorded 1,120 such episodes: 686 addressed to humans and 434 to drones. Hue was coded into 14 perceptual families fixed in a pilot of 90 episodes that are not part of the main data; the families name colours, not meanings. Brightness was coded on a five-point scale, and extent as even or narrowed toward someone.

Coding was done from the recordings, independently, by the first author and a human research assistant, before any interview had been read. Each episode then received a consensus reading, settled by discussion where the coders differed. Interviews were held by the second author, as soon as practical after each episode, and never by the drone coder. Each interview had three parts: what the display had shown, what the drone had been feeling, with its intensity on a five-point scale, and whether the display had been chosen or had happened without a decision. We call the answers glosses.

Agreement between coders is reported as percentage agreement with Cohen's κ, weighted for the ordinal brightness scale (Esslin, 1699 CE). The coders' readings were compared with what drones said the display had shown, and brightness with reported intensity by Spearman correlation. The contrast between displays to humans and to drones was tested by a logistic model with a random intercept for each drone, since episodes are nested within drones, and checked drone by drone with a sign test. We did not start from any glossary of colour meanings; community glossaries circulate on many Orbitals and do not fully agree (Kerrow, 1690 CE), and a study that assumed one would have found it.

3. Findings

Hue carries categories that the community shares. The coders agreed on hue family in 91.4% of episodes (κ = 0.86). Their consensus reading matched the drone's own account of what the display had shown in 1,043 of 1,120 episodes (93.1%, 95% CI 91.5–94.5%); taken separately, the drone coder matched in 1,051 (93.8%) and the human coder in 1,029 (91.9%). The 77 mismatches were readings the coders got wrong or found ambiguous. Participants used between six and eleven hue families each, with a common core and individual additions, as speakers share a vocabulary and keep private words.

Brightness tracked intensity. Weighted agreement on brightness was substantial (κ = 0.71), and brighter fields went with stronger feelings as the drones reported them (ρ = 0.62, 95% CI 0.58–0.65).

Extent carries address. Coders agreed on extent with κ = 0.79. Of 318 displays coded as narrowed, 291 (91.5%, 87.9–94.1%) were described in the gloss as meant for a single listener. Drones used narrowing in company to say something to one person that the others could see but were not being told, and several compared it to lowering the voice.

Most displays were chosen. Drones described 812 of the 1,120 displays (72.5%, 95% CI 69.8–75.0%) as deliberate and the rest as having happened without a decision. Deliberate displays were more common toward humans (541 of 686, 78.9%) than toward other drones (271 of 434, 62.4%). With a random intercept for each drone, the odds ratio was 2.31 (95% CI 1.63–3.27, p < .001), close to the pooled figure of 2.24, and it was similar on the Orbital (2.19) and aboard the ship (2.40). Drone by drone, 49 of the 58 drones observed with both kinds of addressee made a higher share of deliberate displays to humans, 7 a lower share, and 2 the same (sign test, p < .001). The share of deliberate displays did not drift over the study: 71.9% in each drone's first third of episodes and 73.0% in its last.

Two explanations for the difference came from the drones. Many said other drones read them easily, so they took less care. In 188 of the 434 drone-to-drone episodes, the drone also reported a silent exchange with the other drone by other means at the same time, and in those episodes the aura was deliberate less often (57.4%) than in the rest (66.3%). Some of the gap, then, may reflect a second channel carrying the message.

Some displays were chosen to differ from the feeling behind them. In 17 episodes (1.5%) a drone showed something its account of its own feeling contradicted. In 13 the reason was courtesy: a bored drone showed polite interest, or an irritated one showed calm to spare a human's feelings. In the other 4 the purpose was irony, a display meant to be recognised as false. Culture manners prize tact, and drones apply it to their auras as humans apply it to their words.

4. Discussion

The aura behaves less like a face and more like tone of voice. It uses shared, conventional categories. It is mostly deliberate, and more often deliberate when the listener is human. It marks its addressee. And it can be used to be polite or ironic, which an involuntary signal cannot. These are the marks of a register of communication, and we propose that drone studies treat the aura as one.

A comparison with Marain is useful. Spoken Marain has registers that speakers move between according to company and purpose (Zeddaline, 1731 CE), and the aura shows the same sensitivity to company. Drones, who also speak Marain, can use their voices, their auras and, with each other, silent exchanges, and the aura can confirm, soften or contradict what the other channels carry.

The finding bears on how drones are regarded. If auras were readouts, a human who read one would be learning a fact about a machine. On our evidence, a human who reads an aura is usually being told something by a person, who has decided what to say. That fits the Culture's settled view of drones as citizens, not devices, better than the older description does.

5. Limitations

The deliberate-or-not split rests entirely on the drones' own reports, which nothing in the design can check. Glosses were given after the event, and a drone may describe a display as chosen when it was not. The close match between blind coding and the drones' accounts of what was shown suggests that meanings are not being invented afterwards, but it does not bear on deliberateness.

Being recorded and questioned after every display, sometimes over two years, may itself have taught participants to display more deliberately. The share of deliberate displays did not rise over the study, which argues against a strong effect, but it may have been present from the start. Recruiting drones known to the authors, or introduced by participants, also favours sociable drones. Drones who avoid company, or who serve in Contact far from any Orbital, may use their fields differently.