Casual interview with Glynn Stewart: Artificial intelligence in a space opera setting

The following "interview" is actually a conversation over the breakfast table that happened today. Glynn and I were workshopping the topic for a panel that we'd like to get a few authors togethers for, and we thought patrons might be interested in the topic.

Watch out if you're sensitive to spoilers: we talk about how AI works in the House Adamant setting, and briefly touch on a topic that comes up later in The Old Guard. -Jack Giesen

Interviewer: OK, so you just said that AI is never going to happen.

Glynn: Baked into a lot of my books is the assumption that AGI*, Artificial General Intelligence—an artificial “person”—is either not going to happen or [will be] extraordinarily complicated, so you end up with things like the quantum intelligences in Starship’s Mage, where you have this very small number of extremely expensive systems that are where you get synthetic, fake personalities. 

Glynn: But if you actually are looking at their computer systems, there is a vast amount of what we would call “AI” right now, a far more sophisticated version of the kind of the chatbots and…

Interviewer: The large language models? (LLMs)

Glynn: Not the large language models, I’m thinking more along the lines of Content Aware Fill from Photoshop. So, like the mathematical… auto-completing sentences, auto-finishing drawings. It's not replacing humans, it's augmenting humans, it's taking out the grunt work. It’s allowing a human to plug in three things and get the full “Here is what that three-dimensional intercept calculation looks like.” 

Glynn: Because the AI “knows” roughly what its area of expertise is and therefore is working with the human—and it's designed to work with the humans and it's very much a background thing.

Glynn: They don’t think of it as “there’s AIs in their computer,” but their basic computing software in most of my settings is so far beyond what we would even regard as AI, it’s not even funny. But there’s nothing that would actually qualify as an intelligence or a personality in there. 

Glynn: You get sort of a tone, a sense with… like, there's a discussion in House Adamant about how a ship has sort of—and they design their ships this way because that setting has a bit more of the General Intelligence concept baked into how their computers work. It's got sort of a dog-like level of intelligence and self-awareness when they talk about the central ship’s computer. 

Interviewer: Like, when you have a ship and it has the computer and it has a lot of computing power in it, it tends to have about the intelligence of a dog?

Glynn: It has about the personality and self-awareness of a dog. It's not really intelligent as we would think of it, and that's basically the problem with intelligence and artificial intelligence. We're almost never—if we create an artificial intelligence that actually acts like and appears like a human, we're actually focusing the majority of a vast amount of computing capability on simulating a human rather than being a better computer.

Glynn: So in a lot of my settings, they’re like, “Well, why would you bother doing that? We just make it a better computer. We make it better software. We incorporate all of this “learning” and all of this ability to anticipate and so forth into our existing tools. 

Interviewer: Put it into more specialized tools, rather than trying to make a chatbot.

Glynn: Exactly, so instead of having a chatbot that's trying to, you know, answer your questions—I mean, you have that; they don't have Wikipedia, they have a bot. They literally have something that just answers your question. A super advanced version of ChatGPT. 

Glynn: But you don’t have something that writes books; you have a spectacularly advanced Excel where you can where you can plug in four things and it will pop up “Here are five options for what you’re trying to do. Select one or tell me if I’m wrong,” and you can carry on and do it manually but it’s going to anticipate so much because it’s, you know, hundreds of years more advanced. 

Glynn: On the surface, it might look like Excel. It might look like just a spreadsheet, because I think Henry Wong (Peacekeepers of Sol) refers to having a scratchpad spreadsheet in his implants that he uses as the back of an envelope. That’s AI-driven. He’s only putting in about what you’d scrawl on the back of an envelope, but because of the software he has available, that’s turning into a full three-dimensional intercept calculations.

Interviewer: You’re putting something in a grid because as humans, we’re thinking of things on a two-dimensional surface. 

Glynn: Exactly.

Interviewer: But it’s actually going into the computer and [the computer is] anticipating what each of the variables are.

Glynn: It’s one of the things I’ve run into in reviews, is people going “Computers are really good at random numbers.” No, they’re not. Computers aren’t actually random at all. Your random number generator—if you plug a random number into a computer, what you’re getting is an encryption protocol based on the time. That’s it. It’s why there’s a few actual encryption services, security services, that require something that’s truly random—there’s one of them that literally uses a wall of lava lamps** as input because computers aren’t random.

Glynn: And if you've got powerful enough computers and smart enough computers, as you do in most of my settings, you can reverse engineer that encryption. You can reverse engineer that random number generator. Nothing a computer does… if you are following a computer program, however sophisticated, it can be reversed by another computer program. Humans are less predictable. 

Interviewer: So when you think of “AI in the Age of Space Empires”, do you find it most interesting to look at that sort of AI as opposed to a human-like character? 

Glynn: It depends on the story I'm trying to tell. 

Interviewer: Because you’ve gone into House Adamant and…

Glynn: Yeah, House Adamant has synthetic intelligences and they are more typical of my style of AI and AI personalities, and I use “synthetic intelligence” and “synthetic personality” versus “AI” because of what AI is becoming in people’s minds these days.

Interviewer: People make automatic assumptions when you say AI? 

Glynn: Yeah, so I tend to use “synthetic intelligence” or “synthetic personality”, but in House Adamant you’re still talking about something that’s probably the size of a room. It’s not something that’s going to fit in… you don’t have Data (Star Trek: TNG) walking around. 

Interviewer: So you have a far future version of an SSD, of a solid state drive, and so it has even more data within that small space but you’re still taking up a room for that computing power.

Glynn: Yes, exactly. Because to simulate an extraordinary amount of processing power.

Interviewer: I think a few years ago, somebody*** was saying that it would take more than all the computers in the entire world [to simulate a brain] and that calculation has changed.

Glynn: There is a phenomenal amount of computing power floating around in our skulls. And a lot of that is, you know, being distracted thinking about sex and the Roman Empire every 20 seconds each. I suspect personally that Artificial General Intelligence is something that we might build. I also suspect that we will find… it’s kind of useless. 

Glynn: And what we’re going to find that—and what I have in my settings—are these really specialized, really capable, really… intuitive isn’t quite the right word, but anticipatory tools, that are basically useless without a human giving them direction because they’re not really designed to replace people. They’re designed to augment people. 

Interviewer: Meanwhile, the synthetic intelligences that essentially become people, those have their own ethical… 

Glynn: They have their own ethical considerations and restraints. It’s like, what happens when you have created a person? Well, then you’re a parent. And that’s not how people like to think about their software.

Glynn: And so, in a lot of my settings, we’re either “AGI isn’t really a thing” or it’s set far enough in the future that a lot of those ethical constraints have been sorted out. In House Adamant there are some very strict [rules]: there’s the Asimov Conventions around both the rights of synthetic intelligences and what you’re allowed to let a lower level AI, or an “agent program” I think is the term I used in House Adamant, and certainly it’s a term I’ve used in a few places, for what you’re allowed to let an agent do. Like there’s a discussion in the Exodus Gambit when they turn their automatic missile defenses over—they set their missile defenses entirely to automatic, and that is technically a violation, or that one’s a very gray area, under the Asimov Convention because they follow this protocol. 

Glynn: I “stole” the term from Doug Dandridge, I think was the name of the author: Man in the Loop, or Finger in the Loop. There has to be a human… in House Adamant, there has to be a person in the command loop to fire a weapon, and whether that’s a person or a synthetic intelligence is “whatever”. †

Glynn: Most people aren’t going to put a synthetic intelligence who is a citizen—and you know, custom building a synthetic intelligence to run your warship would basically knowingly creating a child soldier. Would be having a baby to make them a child soldier. Not necessarily the most ethical thing.

Interviewer: Well, and in House Adamant you actually play with the ethics and how people try to get around them as well.
Glynn: Yes… but spoilers.

Footnotes:

*The paper “Rise of artificial general intelligence: risks and opportunities” by computer engineering professor Giorgio Buttazzo considers the ethics and risks at the point that “machines match human intelligence not only in some specific fields, but in all human activities”, i.e., achieve true AGI.

**Internet security provider Cloudflare has 100 lava lamps in its lobby. Cameras take photos of the ever-changing wall of lava lamps, and the pixel data from the photos provides "the unpredictable, chaotic data necessary for strong encryption". 

***In 2021, computational neuroscientist Grace Lindsay discussed during a Reddit AMA how we would need a lot more computational power than we had at the time in order to truly simulate a brain. She had just published her book, Models of the Mind, which goes into the complexities of trying to create a model of the human brain. 

† A 2012 Human Rights Watch report categorizes fully autonomous weapons (i.e., “killer robots”) as “Human-in-the-Loop” if the machine is not able to “select targets and deliver force” without human input. The Exodus: Empires at War series by Doug Dandridge uses the Man in the Loop Law, instituted after the machine revolt of 3841-3846 AD.