Into the Light

by David Weber

  She paused as Howell tipped his chair back and looked at her reproachfully.

  “Sorry, Mister President. Basically, what I’m saying is that the Shongairi—the Hegemony—have cracked a step in computer capabilities which we’ve theorized about since the 1980s. We’d actually built a small quantum computer for experiments, but we were still a long way from actually making the concept work. Its advantage over the binary system is that a quantum computer can solve certain problems much more quickly than any ‘classical’ computer, because everything depends on the algorithms of the systems, and quantum algorithms run faster than any probabilistic classical program.

  “All of that’s exciting, and suggests lots of possibilities, but what’s really interesting is what the Shongair haven’t done with it.”

  “I beg your pardon?” Howell arched an eyebrow.

  “We don’t see anything in this system that hadn’t already been part of our theoretical models. They’ve figured out how to do things we hadn’t, but they haven’t figured out—so far as we can tell—how to do anything we hadn’t already worked out pretty fully in theory as theory. Doctor Karahalios didn’t find any great conceptual leaps. No miracle memory devices, no hyper-space shunts connecting all computers into an artificial brain. In fact, despite the qubits, their computers are simply faster and far, far smaller—built on the molecular scale, not the printed circuits we’ve used—but not extraordinarily more capable.”

  “Excuse me, but doesn’t ‘faster and far smaller’ equate to ‘more capable’?”

  “For certain values of the word ‘capable,’ yes.” Lewis nodded. “But before the invasion, the experts all said that any spacefaring society would need artificial intelligence—AI—to support their civilization. Even we Earthlings, stuck here in a single star system, sent ten or more robotic probes into space for each manned mission. And aside from a handful of deep space probes, we’d only truly explored one other planet in that star system in the process. But there are supposed to be hundreds or even thousands of stars in the Hegemony, and it takes years to travel between them, so logically, the Hegemony should have AIs to support them. Why send a manned—well, crewed—starship on a sixty-year voyage to deliver cargo or a message, when you could send an AI on the same mission while your flesh and blood citizens got on with their lives? But they don’t. They don’t have an AI they can send.”

  “I don’t know,” Howell said slowly, pulling his “phone” from his pocket and tossing it onto his blotter. He pointed at it. “I talk to this, and it finds me schedules, information on any topic I tell it to search, or whoever I want to talk to. I may not know which of General Landers’ people I need to get hold of to ask a question, but I ask this—” he tapped the “phone” with an index finger “—and it not only figures out who I need to talk to, it gets hold of him—or her—even if I’ve only given it a very rough description of who I need. Seems pretty intelligent to me.”

  “That’s not really artificial intelligence, Mister President.” Lewis shook her head. “That’s just what Doctor Karahalios is fond of calling ‘an overblown expert system.’ It’s certainly artificial, but it’s not truly intelligent. It can’t think outside the limiting parameters of its basic programming.

  “The first wave of AI was all about teaching computers to sort through large lists of information and find the connections. That meant the programmers were constructing hierarchies which allowed the construction and manipulation of information lists. The second wave created voice interfaces and language translation that allowed those systems to take verbal or written commands and apply them to the information list. When you tell your phone you need to talk to one of General Landers’ officers, it has access to a list of all of General Landers’ officers. When you provide it with specifics about the officer in question, it eliminates everyone on the list who doesn’t match those specifics. It may seem to you that you’re giving it only fragmentary descriptions, but until you give it enough fragments, it can’t find whoever you’re looking for. It can’t … intuit its way to that specific officer. And if you tell your phone to do something—enter a new appointment on your calendar, print out a hard copy of the memo you’ve been reading electronically—it can do only things it’s already programmed to do and only if you tell it to.

  “So, basically, what you might call ‘AI’ is just a very sophisticated library of data and—especially—programs that require human direction, human decisions. Some of those decisions—many of those decisions—can be automated, but in that sense they aren’t really ‘decisions’ so much as automatic responses that the programmers built into the system. The computer doesn’t care what it does. It simply does what it was told to do at every step and in response to recognized outcomes of previous steps, either through its internal programming or as the result of a typed or verbal input from a human operator.

  “The third wave of AI, though—which DARPA and quite a few other people had started to look at before the invasion—was designed to apply actual reasoning and decision-making to computers. To allow them to function independently of human decision-making. To—to go back to my earlier example—to build AI pilots capable of flying autonomous starships between the stars so that Shongair or Barthon or Kreptu crews don’t have to. They do have systems to control ships while the crews are in cryo, but they aren’t autonomous; they aren’t able to think for themselves if something unexpected by their programming comes up. In that case, they have to wake the ship command crew to seek guidance and direction.

  “From our own work, we’re convinced we can build on their existing ‘brilliant software’ to create AI systems which truly are autonomous. Whether or not we could build AI systems which were self-aware is another matter, of course, but we should certainly be able to build something that gives an awful convincing imitation of self-awareness. Assuming we decide we want to, that is. There’s an awful lot of science-fiction about the potential downsides of creating a self-aware intelligence that decides it doesn’t like taking orders from its creators.” Her lips quirked briefly in amusement, but then her expression sobered again. “From an efficiency perspective, though, autonomous AI would be a huge multiplier. Despite which, the Hegemony doesn’t have it when, by all human standards and the self-evident capabilities of its tech base, it should.”

  “Hmmmm.…” Howell rubbed his chin, his expression thoughtful. “So if we were able to use their technology to create this ‘third-wave AI’ of yours for ourselves, we’d have an advantage? A big one?”

  “A very big one, Mister President. But Damianos—Doctor Karahalios—raises a few other points in his current report. Our science-fiction writers and quite a few serious scientists have been looking at neural interfacing for a couple of decades now, and we’ve achieved it, at least to a degree. There’s been some very encouraging work being done with the use of neurally accessible computer chips to store memories for Alzheimer’s patients, for example. That’s not the same thing as a direct brain-computer connection, but it’s headed in the same direction, and the implications of actually achieving that sort of connection are huge. You’ve had plenty of experience giving verbal commands to a computer, even before the invasion. It’s frustrating—or it was frustrating—when the computer misunderstood you, maybe because of background noise, maybe because you weren’t speaking clearly. But consider if you’d had to physically enter every command, instead. Even tapping the screen on your iPhone or your Galaxy generally took longer than a verbal command. But now imagine that you could give your computer commands at the speed of thought and never have it misunderstand you. Do you think that might … enhance your efficiency?”

  “Yes, I imagine it would,” Howell said slowly.

  “Of course it would, yet we don’t see any sign of that in the Hegemony’s computer tech. This despite the fact that they’ve mastered the art of neural education, so they clearly have the ability to send immensely complicated dumps of data at least one way through a neural interface, and they’ve had it longer
than Earth has had to put up with Homo sapiens! So why, in all that time, haven’t they developed the ability to send mental commands into the system? We don’t even see any speculation about the possibility in the literature we’ve been able to access so far!”

  “Why not?”

  “That we would love to know, Mister President. But we have noticed a few other things, most of which appear related to the same sort of … caution that seems to be hardwired into the Hegemony’s entire industrial base. Their programming language is recursive, which means a function can call itself, and it’s also what we might call ‘type safe,’ which means it’s designed to prevent the system from running an operation against the wrong type of variable.

  “Most human programming languages are designed to do that, too, but from what we can see, Hegemony coding takes the concepts to a ridiculous extreme. Consistency checking for acceptable values and variables is built directly into the code for an incredible range of variables. For example, the Hegemony has over five hundred languages, which means—as one of Doctor Karahalios’ programmers pointed out—that using the Hegemony’s programming language to write the equivalent of ‘Hello, world!’ for a planet you’d never visited before or a species whose language you didn’t already speak would involve loading a million-line module. Their protocols would require the module to search the entire Hegemony database to make certain that the world you’re talking to exists—even if you’re currently in orbit around it—and sort through every one of those languages—every language spoken anywhere in the entire Hegemony, not just on the planet in question—and the societal constructs that go with them, to be sure ‘hello’ is the proper greeting, in the proper language, properly spelled and punctuated, in that particular sociopolitical context.”

  Howell stared at her in disbelief, and she shrugged.

  “It’s not that terrible a problem, given their computing speeds and storage ability. The codebase is bloated beyond belief—the executable for ‘Hello, world’ would be hundreds of gigabytes, just to carry all the baggage—and as far as we can tell, their libraries never get pruned, but—”

  “Pruned?” Howell interrupted.

  “There’s no automatic memory management, what a human programmer would call ‘garbage collection’ to reclaim memory occupied by objects the program isn’t using anymore, Mister President. They just store all of them. That’s why their modules are so blasted big. But despite that, it doesn’t slow down the output appreciably. There’s some bottleneck in terms of storage, but not enough to make a significant difficulty, given how much memory their systems have. It’s certainly not anything they can’t handle. We could probably shave some time off of their operations, but not enough for it to make any perceptible difference to the speed at which their programs execute.

  “The problem is that the same ‘check everything again and again’ attitude carries over to their control systems as well as the programs themselves. Oh, we use redundancy in critical systems—like aircraft flight systems, for example, in which there are three completely separate processors running the same calculations. As long as at least two of them come up with the same answer, that’s the one used. If they come up with three separate answers, the system reverts to ‘manual,’ and human supervision is called in. We generally employ systems like that only when failure could have catastrophic consequences, like the loss of human life, though. As nearly as we can tell, the Hegemony applies the same idea to almost everything.”

  “Everything?” Howell repeated.

  “An example, Mister President. A minor component—a motor, say—will have a dedicated processor running dedicated code. You could ‘ask’ it to do something, like turn in one direction at a specified speed for a given length of time, but the central control module tasking the motor doesn’t have direct control over it and can’t override the dedicated processor. And if the dedicated processor senses a potential fault condition, it will simply refuse to let ‘its’ motor turn, no matter what the central system is telling it to do.

  “Now, take that same situation, and apply it to an entire assembly, like one axis of a gantry platform in Invictus. You’ve got dozens or hundreds of motors and similar components, each with its own processor with its own code and multiple redundancy to keep that particular component within safe operating parameters as defined by its programming, and any one of them can shut down the entire gantry if it detects any potential hazard to the single component it’s running. And just to be sure all those individual components have enough ability to see those potential hazards, the gantry has an entire multilevel sensor suite watching every aspect of its environment, and every one of those sensors has its own processor running its own code.”

  “Crap,” Howell muttered.

  “The only thing that makes this workable, and that generates the level of production we’ve seen out of the original platforms the Shongairi left behind, is the speed of Hegemony-level computer operations. They’re so blindingly fast by our standards that they can actually keep this ridiculous balancing act moving forward … most of the time. I’m sure you’ve read some of Director MacQuarie’s comments on how often Invictus simply shuts down until some trivial fault’s been corrected?”

  Howell nodded, and Lewis shrugged.

  “In some ways, that’s probably not a bad thing. We’re still learning how to run it all, so having it stop while we figure out what’s caused its current temper tantrum is one way to really familiarize ourselves with its gizzards. And, as I say, the system actually works. In relative terms, by its own potential standards, it works really, really poorly, you understand, but in absolute terms it’s genuinely capable of producing a post-scarcity economy, something our species has never seen. But we’re estimating—conservatively—that if we could only identify the redundancy levels that are totally unnecessary for safe operation, we could probably increase output by at least another forty or fifty percent—and probably one hell of a lot more than that—just by eliminating all those unnecessary steps and all the inter-processor negotiating that goes with them. And we could save a lot of refinery time and resources—and especially printing time—if we were able to eliminate some of the multiple layers of sensors they build into their hardware. I mean, one sensor and maybe a couple of backups should be sufficient for almost any situation. We’re pretty sure we don’t need twelve of them, though!”

  “Um.” Howell pinched the bridge of his nose and grimaced. “I had a friend before the invasion who worked for the Tennessee Valley Authority,” he said from behind his hand. “One evening, over a couple of beer steins, he described the redundancy of the safety features built into U.S. nuclear plants. According to him, if they’d been allowed to build and operate to realistic threat levels, nuclear plants would have been cheaper than coal or natural gas. And he pointed out that for all the publicity, the Three Mile Island’s reactor design prevented a catastrophic failure despite the fact that the operators did pretty much everything wrong.” He lowered his hand and looked at her levelly. “Now, I’m a firm believer in belts-and-suspenders where something like a nuclear reactor just outside a major city is concerned, but it sounds like you’re talking about the steroids version of the … superabundant redundancy he had to put up with.”

  “Pretty much, Mister President,” Lewis acknowledged, then smiled crookedly. “You know, this whole conversation seems a bit … surreal to me. On the one hand, I’m sitting here talking about how much more efficiently the Hegemony could run its computers and its printers and all the rest of its infrastructure. On the other hand, the way it is running them is still incredibly productive—‘incredibly’ in the sense of literally unbelievable—by any human standard.”

  “I can see that,” Howell said, then let his chair come upright again as the phone on his desk chirped at him and flashed a digital time display.

  “I’m actually sorry to say we’re out of time,” he said. “I’ll try to look over Karahalios’ report, although if it’s anything like the last one he sent me
, I probably actually learned more from you this afternoon than I’ll ever get out of it. Right now, though, I really do have to meet with the Texans.”

  “Of course, Mister President.”

  Lewis rose and started for the door of his office, but Howell halted her with a raised index finger.

  “Mister President?”

  “There’s one point about your explanation that stuck in my mind. Right at the end, when you said the Hegemony’s technology is incredible by any human standard.”

  “Yes, Mister President?” Lewis looked puzzled, and he smiled at her.

  “I know it’s early days, and we’re only really just starting to tear into the possibilities, but I want you and Karahalios and everyone else involved in this project to look for every single way we can improve on what the Hegemony’s willing to accept. When we meet them again, I want them to be the ones thinking about how incredibly efficient and productive human technology is by their standards.” His smile turned cold and hard. “I want that ‘third wave’ of yours, and I want that neural interface, and I want to leave those bastards in our dust when the time comes.”

  “Understood, Mister President.” Fabienne Lewis’ smile was just as cold and hard as her president’s. “Why don’t I just go and get started on that?”

  . XV .

  DREADNOUGHT TRGOVIȘTE,

  PHASE-SPACE,

  1.5 LY FROM EARTH

  “What? You’re out of your mind! Dark Passage is tons better than The Big Heat.”