TRANSCRIPT — Technology segment aired 9 September 2172, 21:30 GMT. Hosted by Dr. Caitlin Turner (Kate) and Dr. Leonard Forbes (Lenny), with guest Dr. Nadia Orlova, Instrumentation and Fabrication Group, Luna Research Dome 1. Produced from Luna University campus broadcast facilities.
[THEME MUSIC — FADES UNDER]
TURNER: Welcome back. I’m Caitlin Turner.
FORBES: And I’m Leonard Forbes. This is The Peer Review, from Luna University.
TURNER: Most of the people listening to this programme have a terminal on them right now. In a pocket, on a belt, bolted to a suit. If you’re in the belt you call it a cuff. If you learned your trade from a Soviet crew you call it a KT. Everybody else says “the terminal” and doesn’t think about it. Last week Dome 1 published the results of a six-month trial in which two hundred and forty of those terminals were given a direct line to Oracle. Not a relay to a governor’s office. Not a ticket in a queue at Mare Frigoris. A query from the hand, and an answer back to the hand. Our guest led the hardware. Dr. Orlova, welcome.
ORLOVA: Thank you. And I’ll say at the start, so nobody writes in: no relation to anyone in the story I’m about to tell. Everyone at Dome 1 asks.
Where the Terminal Came From
FORBES: Let’s do the history, because most of our listeners have never been told it. Why does a twenty-second-century civilian carry a Soviet EVA tool?
ORLOVA: Because in the 2030s nothing else worked outside. The Soviet lunar programme tried commercial tablets and they failed in every way a thing can fail. The screens went dark in the cold. The batteries died on the terminator. Lunar dust got into everything with a seam. Radiation flipped bits in the memory until the operating system forgot what it was. A cosmonaut on the surface needed to read suit power, oxygen reserve, base life support, and the habitat systems bus, and do it in gloves, in vacuum, with fourteen days of night ahead of them.
So the programme specified its own: the Vnekorabelny Diagnostichesky Instrument, the Extravehicular Diagnostic Instrument. Внекорабельный диагностический инструмент. Nobody ever called it that out loud.
TURNER: What did they call it?
ORLOVA: Karmanny terminal. Pocket terminal. KT. Which is a joke, because the first ones did not fit in any pocket made for a human being. But it fit in the thigh pouch of an Orlan-L suit, and that was the pocket that mattered.
FORBES: And the man everyone in your field knows.
ORLOVA: Yuri Kuznetsov. A systems engineer on the lunar programme. In 2032 he took the internals of a Chinese civilian tablet, reinforced them, shielded them, unlocked the firmware, and wrote an operating system for them on the Linux kernel. Hardened against radiation faults, rebuilt for the suit’s diagnostic interfaces. That part was good engineering, and other people were doing good engineering too. The Chinese programme had its own unit, and by most accounts it was better hardware.
TURNER: So why did his win?
ORLOVA: Because he gave it away. He published the whole thing. The kernel modifications, the drivers, the diagnostic stack, everything, open source. Any crew, any agency, any machinist with a soldering iron could build a KT or fix one. When yours broke on the surface you didn’t wait for a part from Moscow; you opened it. The Chinese unit was better. The KT was everyone’s. In engineering those are not the same thing, and the second one wins.
FORBES: And the names followed it outward.
ORLOVA: In English it’s “the terminal,” or “pocket terminal,” PT, which is just the Russian translated. In the belt it became the cuff, because the miners bolted it to the forearm of the suit, armoured, with a second battery, and it stopped being a pocket anything. Same lineage, same kernel underneath. One hundred and forty years of people adding to Kuznetsov’s code. The terminal in my lab is the fourteenth hardware generation and I can still find his comments in it.
What Dome 1 Built
TURNER: So, the trial. What did you actually do?
ORLOVA: We added a transceiver module, the KT-14C, that lets a terminal submit queries to the CAI network and receive answers. Two hundred and forty units, issued to engineering and science staff at Dome 1, Dome 3, and Dome 5, from February to August.
FORBES: And the obvious objection, which I’ve heard from three people in this building since the paper came out: won’t you drown Oracle?
ORLOVA: No, and this is the part people get wrong. Compute was never the constraint. The core at Mare Frigoris runs on a quantum co-processor, and it runs non-governmental queries in parallel without strain. It has been doing more and more of that since the 2140s, after Liang’s Spacer paper showed what it was and wasn’t good for in research. The constraint is the door. Until this year there were two ways to ask Oracle anything: the CAI room at Dome 3, and a governor’s office. If you were a materials engineer at Tsiolkovsky with a question about a weld, you filed it, it went through a liaison, it waited behind every other filing, and the answer came back in a day.
TURNER: And with the terminal?
ORLOVA: Median response on the Luna network was one point four seconds. Over the six months the trial units submitted one point nine million queries.
FORBES: And the queue? Governmental queries always go first. That’s not a technical rule, it’s a Council rule.
ORLOVA: It still applies, and the terminals obey it. When the Council is in session, or a governor files something heavy, terminal queries are pushed back. Our worst delay in six months was eleven minutes, during the March budget sitting. Nobody died waiting for a weld answer. But it’s there in the logs, and it’s honest: the hand gets an answer when the state doesn’t need the time.
The Numbers
FORBES: Let me do the part of the paper that’s going to be quoted badly. Dome 3’s audit of non-governmental queries since 2145 puts outcome agreement for scientific and engineering predictions at ninety-nine point nine percent. For the social sciences, ninety-one to ninety-six.
TURNER: I have got the same curiosity.
FORBES: Right? Because “ninety-nine point nine percent” is going to be on a poster within a week, and it’s ninety-nine point nine percent of something specific. It’s the rate at which Oracle’s predicted outcome matched the measured outcome, for experiments and engineering questions that researchers chose to ask it. Researchers ask it the questions it’s good at. Liang said this thirty years ago on the record: near-total accuracy where it has prior data to model, and nothing at all where it doesn’t. She asked it for the gradient field equations and it returned a null. Not a wrong answer. No answer. Ninety-nine point nine percent is the score of a system that knows when it doesn’t know, being asked things scientists already suspected it knew.
ORLOVA: I agree with every word of that, and it’s in our limitations section. The terminal doesn’t make Oracle smarter. It makes Oracle reachable.
TURNER: And the social science figure? Ninety-one to ninety-six sounds like a different kind of number.
FORBES: It is. People are less predictable than welds, and the model says so. A four-in-a-hundred miss on a question about a crew’s morale is a very good forecast and a very bad thing to treat as certain.
What the Terminal Is Allowed to Do
TURNER: Here’s the question I actually wanted to ask. The terminal is a diagnostic tool. It reads the suit. It reads the habitat bus. On a lot of stations it can also write to them: trip a breaker, vent a line, reroute power. Oracle, by charter, has no action pathway. It describes outcomes; it does not act. Did you just give it hands?
ORLOVA: That was the condition of approval, and it’s the part of the design I’m proudest of. The CAI channel on the KT-14C is display-only, in hardware. The terminal can read the suit. Oracle can read what the terminal sends it. Oracle cannot touch the suit. The answer comes back to the screen, a human reads it, and the human acts or doesn’t. There is no software path from the transceiver to the systems bus, because there is no wire. We built it the same way Dome 3 built the Ethics gate: one direction, one path.
FORBES: So it’s an adviser in your pocket. Not a pilot.
ORLOVA: It’s the speech interface from 2117, in your hand. And it carries the same warning. What comes back to a terminal is a summary-class output. It does not represent the full analysis. We print that on the screen with every answer.
TURNER: Will people read it?
ORLOVA: No. [laughter] That’s a human factors problem and I’m an instrument engineer.
TURNER: One more thing in the logs, and I’m surprised it wasn’t the headline. The trial was restricted to engineering, scientific, and operational queries. About three percent of queries fell outside those classes and were declined.
ORLOVA: People asked it personal questions. Whether to take a posting. Whether to renew a contract. One asked whether to end a relationship.
FORBES: And that’s the thing nobody has thought through. Oracle ranks outcomes for the collective. That’s what it’s for. Ask it about a reactor and the collective and you want the same answer. Ask it about your life and you’ll get the answer that’s best for everyone, and it will sound like it’s for you.
ORLOVA: Which is why those queries were declined. For now.
Where It Goes
TURNER: The paper’s last section is about ships.
ORLOVA: Every Strategic Fleet class above a corvette carries a CAI advisory core, and most of the time those cores have nothing to do. They exist for engagement and navigation decisions, and a ship is not in an engagement most days of its life. Meanwhile its crew does EVA constantly: hull inspections, sensor swaps, radiator repairs. Put a link on their terminals and the ship’s own core gives them real-time diagnostics while they’re on the hull. No light lag, because the core is forty metres away, not on Luna. That’s the application I care about.
FORBES: And the place it won’t work.
ORLOVA: Anywhere there isn’t a core. A UN ship, yes. A UN station, yes. Dome 1, obviously.
FORBES: A miner on a claim forty light-minutes from Hygiea, with a cuff bolted to her arm, gets nothing. The tool came from the cosmonauts and went out to the belt because a Soviet engineer gave it away. The judgment you’re adding to it only goes where the UN has put a core. That’s not a criticism of the engineering. It’s a description of who it’s for.
ORLOVA: That’s fair. I’d only say that Kuznetsov’s terminal took forty years to reach the belt too.
TURNER: Last question. You built this at Dome 1, at Tsiolkovsky, on the far side. Dome 1 runs the power for the radio-quiet array. Where did you test a radio transmitter?
ORLOVA: [laughs] Not at Tsiolkovsky. Inside the domes the link runs optical. Every surface trial ran at Mare Frigoris, because the radio astronomers told us, in writing, that if a single KT-14C keyed up within two hundred kilometres of the array they would personally come and remove it. They meant it.
TURNER: Dr. Nadia Orlova, Dome 1. Thank you.
ORLOVA: Thank you. And look after your terminals. Somebody gave you that.
[THEME MUSIC — FADES IN, THEN OUT]