Evelyn had arranged to take the next day off, and she collected him after breakfast the next morning. She was pleased to see him, even if somewhat awed at his having made the time; she was nervous that she might be misreading more into things than reality warranted, then relieved when he seemed to show as much enthusiasm as she felt.
They went first to the AI Laboratory at MIT and visited some of Corrigan’s former colleagues from his first years in the States. His postdoctoral work at that time had been on the emerging subject of “psychotectonics”: unraveling the roles and dynamics of the sometimes competing, sometimes cooperating hierarchies of functional agencies that make up the phenomenon called “mind.” Although it was Corrigan’s work here on the simulation of evolving neural networks that had earned him his initial recognition, he had moved later, as he had told Evelyn when she was in Pittsburgh, to join Carnegie Mellon’s group working on “Trunk Motor Intercept” technology, which eventually produced MIMIC.
The aim of one project that he showed her at MIT was to expand a machine intelligence’s everyday world-knowledge by getting it to solve detective mysteries. In another, devoted to speech interpretation, they watched a computer creating a cartoon on the fly in response to a narrative being read by Evelyn. On the floor below, a supercomputer from Thinking Machines Corporation in nearby Cambridge was generating admittedly not very good critiques of literature texts. Finally, in yet another room filled with screens, racks, and tangles of cable, Corrigan introduced the department head, Jenny Leddel. She was graying, entering middle age, and wearing a woolen cardigan with a tweed skirt.
“This is Evelyn, from Harvard, who I told you about on the phone,” he said. “She’s going to be joining us down in Pittsburgh.”
“Stealing our talent now, eh?” Jenny said, nodding knowingly. Her eyes sparkled with a mischievous light, young for her years. “It figures.”
“It hasn’t been confirmed yet, Joe,” Evelyn reminded him.
“Ah, don’t be worrying yourself about that at all.”
“How are things going with Pinocchio down there?” Jenny asked Corrigan. “I’ve been following the reports. It sounds exciting.”
“Going well. We’ll have to get you down there sometime to see for yourself,” he said.
“I’d like that.”
“We’re all set for P-Two: going up to the pons. That’s what Evelyn will be working on.”
“You achieved a full two-way integration, yes?”
“DINS with MIMIC. We’ve had it running for about three months now.”
“Complete internal haptics?”
“Total. It works. Uncanny. Evelyn tried it a few days ago.”
“What about the secondary instabilities that Goodman’s people at Chapel Hill kept running into? You didn’t have a problem with them?”
“Our DINS expert came up with a C-mode suppression filter that cured it. A character called Eric Shipley. Do you know him?”
“I’m afraid I don’t.”
“He’s good—the old-school type. Infuriatingly plodding at times, but he gets it right in the end.”
“We could use a few like that here,” Jenny said. “Too many these days trying to fly before they’ve grown feathers.” She gave Corrigan a pointed look as she said this, but he missed it. Jenny didn’t make an issue of it, but turned to Evelyn. “Anyhow, enough of that. We’re ignoring you. Joe says you want to talk to Perseus.”
“Sure. If he wants to talk to me.”
This was the latest to come out of the learning systems based on goal—directed, self-adaptive, neural-net analogs that Corrigan had worked on during his time with MIT: systems that experimented with problem-solving strategies. They devised new variations of what seemed to work best, and forgot about what didn’t—the process known in nature as “evolution.” An ideal strategy-testing environment—full of clearly defined challenges and yielding easily measured results—was the classical dragons-and-dungeons type of adventure world. Perseus, accordingly, was a computer-created character who explored such mythical realms, with similar goals to achieve and obstacles to be overcome. Half of AI research, it seemed, was wrestling with the problem of trying to impart world-knowledge.
Jenny tapped commands into a console to activate the system. A simplified image appeared on a screen of a typical D & D setting of a large room, assorted objects, with passages, stairways, and tunnels going off in various directions.