“Is it stuffy in here after the food?” Lilly asked suddenly. “I can’t tell. I’ve got a sinus problem that stops me smelling things.”
“It’s okay,” Corrigan said. “I used to be with one of the big companies here: Cybernetic Logic Corporation—I worked at their corporate research center out at Blawnox. They were big in Artificial Intelligence-based systems. Still are, for that matter. The aim of the AI field had always been true, human-level intelligence, one day. But around the turn of the century, the technology was plateauing out. After some progress and mixed results, there didn’t seem to be any obvious way to advance things further.”
“Yes, I know CLC,” Lilly said. “They’ve got a building downtown, near Westinghouse.”
Corrigan nodded. “Well, about twelve years ago, CLC set up a big research project to try a new way of achieving AI. It might come as a surprise, but I practically invented it.” He paused, but Lilly merely returned a stare that could have meant anything. Corrigan went on: “You see, traditionally there had been two approaches to AI: top-down and bottom-up. Top-down meant trying to understand all the complexity of this thing we call `mind’ in sufficient detail to code it into programs.” He waved a hand in front of his face. “Forget it. The immensity of the task would make it intractable, even if we knew what to code.”
A strange half-smile was playing on Lilly’s lips, but in his soliloquizing Corrigan failed to notice. He continued: “The other way, bottom-up, meant trying to create simple neuronlike configurations that could be made to evolve, the same as we did. The problem you run into there is that you don’t realize how efficient animal nervous systems are until you try imitating them. You can spend ten years, fifty million dollars, and the best brains in the business putting TV cameras and legs on a computer to make it walk, and the average twelve-month-old will run rings around it—literally. The simple fact is, computers don’t interact very well with the real world outside. They haven’t had a billion years of evolution optimizing them for it. They operate better on their own, internal worlds.”
Lilly nodded, finally, and raised a hand. “It’s okay, Joe. You don’t have to go on. The project was called Oz—set up under a new CLC division called Xylog, across the river, along Carson Street—yes? The idea was to let an AI evolve by interacting with a virtual world.”
Corrigan stared at her in astonishment. “Xylog! That’s right. Some of the buildings are still there . . . I don’t know what they’re used for today. How in heaven would you know about that?”
Instead of answering immediately, Lilly continued, “But Oz was shut down in the preliminary test phase. Before that, were you working on the program that led up to it: a project called EVIE?”
Corrigan shook his head bemusedly. “How in God’s name—”
“I’ve got one more,” Lilly said. “Then you’ll get your answers. What happened? Why was the Oz project abandoned, and what did it have to do with your winding up in a place like the Camelot?”
It wasn’t something that Corrigan normally talked about, especially to people he hardly knew. But these were hardly normal circumstances. “How much do you know about how the interaction was going to be implemented?” he asked, to avoid launching off into needless explanation.
“Enough,” Lilly answered. “The idea was that the system would learn by manipulating humanoid animations to emulate real-person surrogates projected in from the outside.”
The AI would evolve by controlling artificial characters in a virtual world. As a substitute for the directional thrust of biological evolution, the system would endeavor to shape the behavior of its creations closer to that of surrogate representations of volunteer participants coupled in from the outside. Thus, the virtual world would contain two kinds of inhabitants: humanoid “animations,” manipulated by the computer; and “surrogates,” controlled by real people, represented as themselves. The test would be to see if the machine could make the behavior of the animations indistinguishable. From her reply, Lilly was aware of all this.
“And am I right in supposing that you know how the surrogates were coupled in?” Corrigan said. “VIV? DIVAC? You’ve heard of them?” He meant the latest developments at that time in direct-coupled neural I/O, which had appeared on the scene after the earlier VR paraphernalia of head-mounted displays, bodysuits, and so forth. It had come out of work going on at places like Carnegie Mellon and MIT, certain government departments, and Advanced Telecomms at Kyoto, that involved interaction directly with the neural structures of the brain.