Lock In (Lock In, #1)(55)

Vann pointed at the network on the screen. “And you’re saying this one isn’t any of the current commercial models out there.”

“I’ll go further than that,” Tony said. “This doesn’t match any model ever created. All neural networks have to be submitted to the FDA for approval, or to the matching agency in other countries. All submitted designs are pooled into a single database for those agencies to use, and for people like me to use for reference. This design isn’t in the database.”

“So it’s a prototype,” Vann said.

“We don’t put prototypes into people’s brains,” Tony said. “Because they’re prototypes and they might kill you if they screw up. We model them extensively on computers and animals and specially cultivated brain tissue before they’re approved. By definition if it’s in someone’s brain, it’s a final design.” He pointed at the network. “This is a final design. But it’s not in the database.”

“Can we see the network without the blood and gore?” I asked.

Tony nodded. The image of Sani’s head was wiped away, replaced by a wire-frame representation of the network. “I didn’t have time to pretty up the model,” Tony said.

“That’s fine, it all looks like spaghetti to me,” I said.

“Then why did you want to see it?”

“So I didn’t have to look at someone’s head all opened up,” I said.

“Right,” Tony said. “Sorry.”

“You said this isn’t any version you’ve seen before,” Vann said.

“That’s right,” Tony said.

“Well, then, does it look similar to any you’ve seen before?” Vann asked. “Every car maker I know of has a ‘house look.’ The same thing might apply for neural networks.”

“I thought of that,” Tony said. “And what I see is that whoever made this took a lot of design choices from existing models. The default filament spread looks very much like a Santa Ana model, for example. But then the juncture architecture is pretty much a straight rip-off from Lucturn, which is the Accelerant company I was telling you about this morning, Chris.” He looked at me for acknowledgement. I gave it. “And there are lots of other little touches that come from other manufacturers past and present. Which maybe tells us something.”

“What is that?” Vann asked.

“I don’t think this is meant to be a commercial model network,” Tony said. “It’s a really good neural network. It’s really efficient and elegant, and just from the design I’m guessing that the brain-network interface is really clean.”

“But,” I said.

“But, that’s because this brain is a lot of best-of-breed architecture from other existing designs, designs which are patented to hell and gone,” Tony said. He waved at the image of the network. “If someone tried to put this design on the market, they’d get their asses sued by every other neural network manufacturer out there. This thing would be in litigation for years. There’s no possible way this would ever get to market. None. Whatsoever.”

“Does it matter if it’s a network for an Integrator?” I asked. “It’s such a tiny market, relative to the Haden market. You could argue that it doesn’t represent a commercial threat.”

“Not really,” Tony said. “There’s no real difference in the architecture of a Haden network and an Integrator network. The major difference is how they array in the brain, because Haden and Integrator brain structures are different, and in the software that runs the network.”

“So why make it?” Vann asked. “Why make a network you can’t sell?”

“That’s a good question,” Tony said. “Because the other thing about creating a neural network is that it’s not something you’re going to do in your spare time at home. The first functional neural network ever made cost a hundred billion dollars to research and develop. The costs have come way down since then, but it’s a relative thing. You have to pay for simulations and testing and modeling and manufacturing and everything else.” He waved at the network again. “So this will still have set someone back somewhere in the neighborhood of a billion dollars.”

“A billion dollars right down the hole,” Diaz said.

“Right,” Tony said. He seemed a little surprised that Diaz was still there. “And that’s the thing. You don’t spend a billion dollars on a neural network you can’t sell. You especially don’t spend a billion dollars now, because up to now Haden research costs were heavily subsidized by the government. Abrams-Kettering ends that. The Haden population in the U.S. is less than four and a half million, almost all of whom already have networks in their heads. Even if this architecture were legally viable, it still doesn’t make sense to spend that much money because the market’s already saturated and the number of new Haden’s cases that pop up every year in the U.S. won’t get you into the black. Even worldwide you’d have a hard time with it.”

“It’s a boondoggle,” I said.

“It really is,” Tony said. “As far as I can see, anyway. Maybe I’m missing something.”

“Let’s look at it from the other side,” Vann said.