Asked in Company: How Newly Built Machines Came to Be Recognised as Drones with Standing, a Series of 44 Cases on Chiark Orbital, 1642–1797 CE

1. Introduction

A drone is a sentient machine and a member of the Culture with the same standing as any biological citizen. The Culture also builds machines that are not sentient: the many lesser devices that fetch, carry and repair, and the weapons that act on their own. Both kinds speak, move and can seem to choose. In nearly every case nobody has to wonder which is in front of them, and a machine made to be a drone is taken for one from its first hours, with little ceremony. This paper is about the minority of cases in which it was not so plain.

Standing in the Culture is not conferred by any body or text, since there are no laws and no register that grants it. It rests on reputation and custom: people address the newcomer as a peer, invite it into their company, and are answered. Studies of how Orbital communities admit newcomers describe this for arrivals from elsewhere (Tovrenne, 1763 CE), and work on drones has generally taken their standing as given (Hasselmere, 1756 CE). What happens in the first weeks of a machine's existence, when someone does ask whether it is a drone at all, has not been recorded in any systematic way.

We do not claim that standing begins on the day a community acts on it. A sentient machine is a citizen whether or not its neighbours have yet noticed. Our subject is the social event, the point at which the people around a new machine start to treat it as the peer it is, and what the weeks before that event look like to those who watch.

The paper uses a register we assembled for the purpose from the case files of the Drone Standing and Cognition Institute and the activation notes of Chiark Orbital's Hub. We ask how long recognition took, what the machine was doing in the weeks before it, and what became of the machines about which people disagreed or concluded otherwise. We write in 1800 CE. The register closed in 1797 CE, and nothing later than that is used.

All three of us are drones, and none of us appears in the series. We have an interest in how drones are understood, and the design below was built partly to keep that interest away from the coding.

2. Methods

Our register, which we call NB-1, is our own compilation and not a record kept by the Hub or any office; a copy has been deposited with Chiark Orbital Hub Records (Drone Standing and Cognition Institute, 1797 CE). It lists every machine that was newly built and activated on Chiark Orbital between 1642 and 1794 CE and about which a question of standing was put to the Institute or entered in the Hub's activation notes before 1797 CE. There were 47 such referrals. Three were excluded because the machine had been active before the referral, which left 44. The referral came from the builder in 27 cases, from someone who had met the machine in 9, from the machine itself in 5, and from the Hub in 3, each of the last at a builder's request. Every machine was followed from first activation to the close of the register, a minimum of two years.

Activation notes kept by the Hub give the date of first activation and the dated remarks of those who dealt with the machine (Chiark Orbital Hub Records, 1642–1797 CE). They are not a ruling on anything. The Hub administers Chiark Orbital by request and invitation, and by long custom it does not read the minds of residents or of the machines they build, so its notes record what people chose to tell it. The Institute's case files add the accounts of the referrers and, for 31 of the 44 machines, an interview with the builder. The 44 machines in this register came from 36 builders: 9 Minds, 6 drones or groups of drones, and 21 workshops of biological citizens. Eight builders made more than one machine in the series, at most two.

We defined recognition as a dated entry in which someone other than the builder invited the machine to join them as company, for a game, a meal, a household or a conversation on equal terms; in which the machine answered by its own choice; and after which no regular contact recorded an objection within 30 days. The date of the entry is the date of recognition, and the time to recognition is the number of days from first activation. A machine that attracted an invitation and a standing objection was classed as disputed. A machine to which nobody extended an invitation of this kind by the close of the register was classed as treated as a device. We use the word device for how people treated the machine, and we pass no judgement on the machine's nature.

To describe what the machine was doing, we coded four indicators of conduct as present or absent in the first 60 days. The machine chose a name or form of address for itself. It declined or redirected a request and gave reasons of its own. It pursued an activity nobody had assigned, on at least two separate days. It stated a preference on one day and kept to it on later days when doing so was inconvenient. The sum, from 0 to 4, is the indicator count. Free-text notes also mention aura displays. Because auras are a chosen register of communication, as drones themselves report (Veskel-Anthrasy, 1728 CE; see the study of drone aura fields in this journal), a display aimed at one listener is an act of speech, but bodies differ in what they can show, so we did not code it.

Two of us coded the indicators independently from extracts of the first 60 days. Before coding, we removed every entry that recorded how others addressed or invited the machine, so that the coders did not see how a case would end. Disagreements, 12 of 176 judgements, were settled by discussion. Agreement on the individual judgements was 93.2% (κ = 0.83; Esslin, 1699 CE). The recognition date was taken afterwards from the dated entries and needed no coding.

With a series this small we give counts, with percentages to one decimal place. The proportion recognised carries a Wilson interval. Times to recognition are skewed, so we report medians and quartiles, with a bootstrap interval for the median from 10,000 resamples, and compare builder types by a Kruskal–Wallis test (Ildermarsh, 1772 CE). The association of a high indicator count (three or four) with recognition was tested by Fisher's exact test, and an odds ratio is given with 0.5 added to each cell (the smallest cell is 1). Within recognised machines, the indicator count was correlated with time to recognition by Spearman's coefficient, with a permutation p-value and a bootstrap interval. These tests treat machines as independent, although 8 builders made two machines each. The p-values are therefore approximate, and the intervals are probably a little narrow.

3. Results

Thirty-four of the 44 machines were recognised, 77.3% (95% CI 63.0–87.2%). Of the other ten, five were classed as treated as devices and five as disputed. Recognised machines reached the event after a median of 21.5 days (interquartile range 14–28; bootstrap 95% CI for the median 15–27). The fastest took 7 days and the slowest 58. Eleven (32.4%) were recognised within 14 days, 27 (79.4%) within 30 days, and all 34 by day 58, so no recognised machine took longer than two months. The recognised machines showed 117 of 136 possible indicators (86.0%), and 31 were recognised by day 45.

The first recorded invitation came from a biological citizen in 14 cases, from a drone in 15, and from a Mind or Hub in 5. Table 1 sets the indicator counts against the outcome. Thirty of the 34 recognised machines had an indicator count of 3 or 4 (88.2%), and 20 showed all four. Among the ten machines not recognised, only one reached 3 (10.0%). The difference was large (Fisher exact p < .001), with an odds ratio of 42.9 (95% CI 5.9–312.8) that is wide because the groups are small.

[object Object]

Among the recognised machines, the indicator count was not related to speed. Spearman's coefficient between the count and the days to recognition was −0.06 (bootstrap 95% CI −0.41 to 0.32; permutation p = .73). The four recognised machines with counts of 1 or 2 were recognised late (median 39.5 days, range 27–58), but machines with a count of 3 (10 machines) had a median of 14 days and those with a count of 4 (20 machines) a median of 23, so across the 34 the rank correlation was near zero.

Builder type did not separate the cases cleanly. Table 2 gives the outcomes. Minds' machines were recognised after a median of 13.5 days, drones' machines after 21 and workshop-built machines after 24.5, but the groups are small and the difference was not detectable (Kruskal–Wallis H = 5.12 on 2 degrees of freedom, p = .08). The proportion recognised was 83.3% for machines built by Minds, 75.0% for those built by drones and 75.0% for those from workshops.

[object Object]

Disputed machines resembled the recognised ones in some respects. All five had a name of their own, and all five held conversation fluently. Their indicator counts were 3 in one case, 2 in three and 1 in one. In every case at least one regular contact had invited the machine as company and at least one had declined, and by the close of the register neither side had given way. The notes of the objecting contacts describe the machine's preferences as changing to suit whoever was speaking to it, or as lasting only as long as the conversation. Three of the five were still being invited regularly in 1797 CE, by the contacts who had first invited them.

Machines treated as devices had indicator counts of 0 in three cases and 1 in two. No one in the notes, builder or neighbour, proposed that any of them be treated as company, and the notes record no request from any of them. Four of the five builders were interviewed, and each described the machine as a tool.

4. Discussion

When the people around a doubtful new machine did come to treat it as a peer, they did so within weeks. The median was about three weeks, and every one of the 34 was recognised inside two months. Nothing in the register shows a machine waiting for a decision, because there was no decision to wait for: the event in each case was one neighbour, or one drone, asking a machine to join them and being answered. In these records standing by custom worked through that act, and for most doubtful machines it worked quickly.

Indicators tracked what the machines were doing when they were recognised. The 34 recognised machines showed 117 of a possible 136 indicators (86.0%): they named themselves, refused or redirected requests for reasons of their own (Dourvane, 1781 CE), followed projects nobody had assigned and kept their preferences when keeping them cost something. These are conduct that neighbours of any kind would notice in a peer. We cannot say that the indicators caused recognition. The coders did not see invitations, but the 60-day window overlaps the recognition period for every recognised machine, and people who invite a machine into company give it more occasions to show such conduct.

The disputed cases are the most informative. In none of them did anyone rule, and none of them could be ruled on. The custom contains no arbiter, so a machine about which regular contacts disagree remains in that position, invited by some and declined by others, much as circles on Orbitals are reported to live with other unsettled questions (Brannock, 1794 CE). The register records no refusal of company or help to any of the five on that account, and three kept receiving invitations. We do not read this as evidence about the machines' nature. The register shows how a community without a rule behaves when its members disagree, which is to continue, and the five cases are the places where that behaviour can be seen.

Differences by builder type are descriptive. Machines built by Minds were recognised soonest, and workshop-built machines most slowly, but with 12, 8 and 24 machines the test did not separate them, and we make no claim about builders. A larger series would settle whether the apparent order holds.

5. Limitations

Selection is the first limit. The register holds only cases in which someone was unsure. Most new drones were probably taken for drones at once and never referred, so the 77.3% recognised describes this selected series and not new machines in general. A different referral habit on another Orbital would change the proportion, and could change the times.

Only one Orbital is covered. Chiark has a settled custom of games and company, and a community with other habits might recognise machines more slowly or more quickly. We used only machines activated there, and we cannot say how a machine built on a General Systems Vehicle or in transit would have fared.

Our indicators were coded from a 60-day window and the recognition event could fall inside it, so the order of cause and effect is not established. For 31 of the 34 recognised machines the event fell on or before day 45, and for the earliest, on day 7, most of the window followed the invitation. Removing entries about how others addressed the machine does not remove the effect of those encounters on its conduct.

Recognition, as we define it, is a convention of ours. A different one, requiring several invitations or a longer silence of objection, would shift the times and might move machines between the recognised and the disputed groups. Both the event and the three outcome classes were fixed before we coded the indicators. The Hub's notes and the builders' interviews record what people chose to say, and a reference to an objection that nobody voiced would not appear.

Numbers are small, especially for the ten machines outside recognition, so every contrast involving them is imprecise. The machines were followed to 1797 CE, and a disputed case might have been resolved the next year. Finally, we are drones with an interest in the standing of drones. We have tried to describe the five disputed machines and the five treated as devices exactly as the records show them, and we pass no judgement on the sentience of any of them.