The Ouroboros Wave (4 page)

Humanity had been living in space for several generations, but the history of that settlement was still limited. Chandrasekhar Station had been named after a Nobel Prize–winning physicist, but there remained a preference for ancient names like Ouroboros and Amphisbaena—perhaps a sign of humanity’s desire to feel that they had planted their feet firmly in space long ago.

“
HEY, KUROKAWA—GET THIS.
”

Tatsuya had finished his rice and was now holding his tapered chopsticks up horizontally, butted end to end. “It’s Amphisbaena!”

“Stop trying to change the subject and start acting like the head of this station!”

“All right, okay…”

“If you’re finished eating, I’d like to go over the cargo capture schedule.”

Tatsuya cleared away his dishes, somewhat sorry that lunch was over. When he looked up, there was an augmented-reality display board floating in midair a few feet away. He activated an agent program from his web, and the same opaque white rectangle appeared throughout the station where his team was at work.

The board displayed the current positions of Ouroboros, the station, the nearby planets, and an approaching fleet of cargo containers. Kali’s orbit was highly inclined to the ecliptic plane, approaching aphelion. Once Kali reached its furthest point from the Sun, AADD’s goal was to redirect it along a path that would lead to eventual capture by Uranus.

Reducing relative speed with the ring to zero was not the only reason for Amphisbaena’s rotation. It also facilitated the recovery of cargo containers coming from other locations throughout the solar system. Kali was already more than ten astronomical units from the Sun. Because its orbit was highly inclined, the distance to Mars and Jupiter was about the same. But since the speed of the containers varied depending on their origin, speeds had to be equalized by Amphisbaena’s rotation before the containers could
be sent on to their destination.

The tips of the station moved at several kilometers per second. With proper timing there was scarcely any need for an approaching cargo container to decelerate before recovery. This meant less propellant to make the journey, even as other navigational elements including angular motion had to be adjusted. Still, the huge inertia of Amphisbaena’s millions of tons of mass acted as a kinetic energy
sink for incoming containers.

“The accident on East Platform threw our construction schedule off. I’ve configured a new timeline. Suppliers on Titan and Mars
are now working to this schedule.”

There were advantages and disadvantages to sending cargo containers across space. One advantage was that space itself could be used for large volumes of cargo storage. One disadvantage was that emergencies and changes in plan were naturally difficult to
accommodate.

To solve this problem, AADD used containers with limited propulsion capability. While the containers coasted on an inertial trajectory after launch, a certain amount of acceleration or deceleration allowed time of arrival to be adjusted for unforeseen problems or changes in plan. Kurokawa’s plan, displayed on the board, detailed
a raft of changes in container trajectory speeds.

“I think this is the optimal strategy, given our present status. What do you think, Chief?” Kurokawa was clearly trying to minimize any impact on the current program, and there were no conspicuous changes in the arrival sequence. “This should let us stick as close
as possible to schedule.”

Tatsuya accessed the ring’s computer system via his web to verify the new timeline. The schedule did indeed minimize changes, but this was not what he had in mind. In a few seconds, he formulated an alternative based on different parameters. “Your proposal keeps us close to program, Kurokawa. But under the circumstances it’s not what I’m looking for.”

“I don’t understand.”

“Our logistics strategy is to keep construction going as efficiently as possible, not necessarily to stick to the program. Right now our first priority is to get Sati up and running. That means we’ve got to expedite the materials we need to get that done.”

“How are we going to do that?”

“The other divisions won’t like it, but we’ll have to change their allocations. SysCon has first call on network equipment.” Tatsuya uploaded his plan to the board. First, all equipment on Amphisbaena needed to activate Sati would be offloaded at the next rendezvous with West Platform. The effect of this change would be offset by rejiggering the velocity correction sequence for the containers.

Amphisbaena’s recovery schedule now looked very different. The board used different color densities to indicate relative differences in the correction sequence for each container in the fleet. Some would be parked in orbit around Kali to form a recovery queue, again using space as a temporary staging area. The overall plan was feasible but clearly more complicated than what Kurokawa was proposing.

“But, Chief, why are we prioritizing SysCon’s needs when the schedule is already out of whack because of the accident?”

“The accident is the reason we want to get Sati up and running as soon as possible. Listen, Kurokawa—why do you think Dr. Chapman was killed?”

“It’s obvious. He shouldn’t have disconnected the speed limiter.”

“Wrong. The cause was whatever modifications he made to Shiva before he was killed. When he realized that, he was forced to depart from standard operating procedure. Right now Shiva is our only high-level AI looking after resonance control. And this same AI violated its operating protocol by not gradually slowing the truck down. If the safety and integrity of the ring take priority, so does solving this problem.”

“What does activating Sati ahead of schedule have to do with it?”

“A lot. Whatever Chapman did, it was like an operation on the frontal lobes of a human brain. Analyzing the logic chains of a high-level AI is no mean feat, even for human specialists. There’s just too much to look at. AI reasoning is very different from ours. Even the people who write the software can’t predict with complete certainty how an AI will behave in a given situation. In a way, that was the whole point of trying to model intelligence. Sati is like Shiva but also different, so we need her help to get to the bottom of this.”

“I didn’t know you knew so much about AIs.”

“I have a good teacher.”

“Oh, you’re just passing on something you got from Kitty.”

“Why not? We’re logistics. Passing things on is our job.”

“I hope you won’t mind if I go to the source, then. Kitty?” Kurokawa initiated a link between himself, Tatsuya, and Catherine.

“How many times do I have to tell you? Stop calling me that,” said Catherine. “Have you disabled your agent’s learning function?”

“Of course not. Why do you think it always calls you Kitty?”

“Great. So what’s going on? You know I’ve got my hands full.”

“Listen, about your procurement schedule—the chief tells me we’re supposed to prioritize Sati. What I want to understand is, why are we using one machine to debug another machine? Chief says our reasoning processes are different from an AI’s. But I mean, one plus one is going to equal two whether a human or an AI is involved.”

“Are you sure? What does apple plus sugar equal?”

“That’s not addition. What am I supposed to say—apple pie?”

“I don’t know. But an AI might come up with an answer to that question, if it was required to do so. Maybe it would say apple pie, or maybe it would decide that ‘apple’ and ‘sugar’ are the same thing.”

“But where’s the correlation? Is that even reasoning?”

“Of course. When correlation is obvious we don’t need to reason about it. If we decide that the answer the AI comes up with is meaningless, then we reject its reasoning. So here’s the problem—when we send a human to debug an AI, how do they evaluate reasoning
processes that appear meaningless to them?”

“They could just ignore the processes.” Kurokawa wouldn’t give in,
but Catherine was used to dealing with this kind of resistance.

“You know, it would be so nice if we could,” she said, a hint of enjoyment in her voice. “But what if the AI rejects two as an answer to one plus one? Listen carefully—the AI’s conclusion isn’t important. What’s important is the reasoning chain that led to that conclusion. If Shiva had behaved as expected, he would’ve slowed the truck
gradually. But that’s not what he did. And now Graham’s dead.”

“For reasons unknown,” said Kurokawa. “So that’s where Sati
comes in?”

“Yep. But what makes things sticky is that Shiva isn’t malfunctioning. Shiva—as a machine—is functioning normally even as we speak. Yet he’s demonstrated that he’s capable of reasoning to an incorrect
conclusion. It’s not easy to repair a machine that isn’t broken.”

“And that’s why we’re supposed to upend the schedule to prioritize System Control’s equipment needs. So let me ask you this: what makes you think Shiva will cooperate?” Tatsuya had already asked Catherine this very question. Her look of pity came back to
him vividly. “After all, why wouldn’t he rebel?” Kurokawa continued.
“He’s supposed to be smarter than we are in some ways. According
to you, he’s already killed once.”

“Kurokawa, you need to get this through your head. Shiva doesn’t know we exist. He doesn’t even understand the world in terms of
relations between objects, at least not the way we do.”

“I can prove you wrong.” Kurokawa touched his wrist, stood, and went to the coffeemaker in his quarters. A few seconds later he held a fresh cup of coffee. “Behold. I give the order, here’s my cup of coffee. Shiva knows which human I am, he knows what coffee is. Otherwise how could he do any of this?”

“You’re obviously suffering from interface illusion. Shiva doesn’t ‘understand’ coffee. Mr. Black River knows that every item on Ouroboros or the station has an item code. Well, Shiva understands the relationship between the code for coffee and the code for Black
River, and that is precisely all he knows.”

“I’m not Black River. I’m Kurokawa.”

“And I’m not Kitty, I’m Catherine Sinclaire. Still, when I called you Black River, you knew I was referring to you. That’s because you grasp what it means for there to be an object behind a label. For an AI, the universe is symbols and abstractions, nothing more.”

“But there’s no item code for Kurokawa.”

“You used your web,” said Tatsuya.

“Oh, right. My ID.”

“Yes,” said Catherine. “Shiva has no concept that a human named Kurokawa ordered a stimulant drink. What he ‘knows’ is that web ID THX-1138 or whatever is in a certain location, and that’s where item XYZ needs to be delivered. When a machine does this, humans
think it ‘understands’ what it’s doing.”

“So what’s all this about high-level intelligence?”

“It
looks
like intelligence to us. It’s one of the problems we face in artificial intelligence research. Most of the work being done these days is about the interface, not about teaching AIs to grasp meaning. Basically, most AI work is designed to fool humans into
thinking that real intelligence is present.”

The discussion shifted to the container recovery process. After some tweaking based on input from the other specialists online,
they voted to proceed with Tatsuya’s plan.

“Happy now, Cath?” said Tatsuya.

“Totally. We ran the prelims on Sati and she seems fine. I would like to commandeer one more thing—the temporary habitats that
were allocated to East Platform. Can you swing it?”

“Why do you need more shelters than we already have out there?”

“A lot of people are going to be working on this. After reviewing the timeline, I think we need to increase the living space so we can stay on top of this round the clock.”

“I assume Sati’s like Shiva? No remote access?”

“Exactly.”

“Hold on a sec, let me take a look.” Tatsuya viewed Catherine’s incoming wish list on his web. Everything that AADD moved around the solar system was bar-coded and tagged electronically. It was an old technology that had proven its worth; so far no one had come up with anything better. Tying the solar system together with a vast, smoothly functioning logistics system was one of AADD’s main long-term goals.