"I can't make anything out of this," Ross said, frowning.

"It looks like he's dozing on and off, and he's gotten a couple of stimulations, but…" She shook her head. "Isn't there another display mode?"

As she spoke, the computer produced another report, adding it to the column of numbers:

"People," Gerhard said, in mock irritation. "They just can't handle machine data." It was true. Machines could handle column after column of numbers. People needed to see patterns. On the other hand, machines were very poor at recognizing patterns. The classic problem was trying to get a machine to differentiate between the letter "B" and the letter "D." A child could do it; it was almost impossible for a machine to look at the two patterns and discern the difference.

"I'll give you a graphic display," Gerhard said. He punched buttons, wiping the screen. After a moment, cross-hatching for a graph appeared, and the points began to blink on.

"Damn," she said when she saw the graph.

"What's the matter?" Gerhard said.

"He's getting more frequent stimulations. He had none for a long time, and then he began to have them every couple of hours. Now it looks like one an hour."

"So?" Gerhard said.

"What does that suggest to you?" she said.

"Nothing in particular."

"It should suggest something quite specific," she said.

"We know that Benson's brain will be interacting with the computer, right?"

"Yes…"

"And that interaction will be a learning pattern of some kind. It's just like a kid with a cookie jar. If you slap the kid's hand every time he reaches for the cookies, pretty soon he won't reach so often. Look." She drew a quick sketch.

"Now," she said, "that's negative reinforcement. The kid reaches, but he gets hurt. So he stops reaching. Eventually he'll quit altogether. Okay?"

"Sure," Gerhard said, "but- "

"Let me finish. If the kid is normal, it works that way. But if the kid is a masochist, it will be very different." She drew another curve.

"Here the kid is reaching more often for the cookies, because he likes getting hit. It should be negative reinforcement, but it's really positive reinforcement. Do you remember Cecil?"

"No," Gerhard said.

On the computer console, a new report appeared:

"Oh shit," she said. "It's happening."

"What's happening?"

"Benson is going into a positive progression cycle."

"I don't understand."

"It's just like Cecil. Cecil was the first monkey to be wired with electrodes to a computer. That was back in '65. The computer wasn't miniaturized then; it was a big clunky computer, and the monkey was wired up by actual wires. Okay. Cecil had epilepsy. The computer detected the start of a seizure, and delivered a counter-shock to stop it. Okay. Now the seizures should have come less and less frequently, like the hand reaching for the cookies less and less often. But instead the reverse happened. Cecil liked the shocks. And he began to initiate seizures in order to experience the pleasurable shocks."

"And that's what Benson is doing?"

"I think so."

Gerhard shook his head. "Listen, Jan, that's all interesting. But a person can't start and stop epileptic seizures at will. They can't control it. The seizures are- "

"Involuntary," she said. "That's right. You have no more control over them than you do over heart rate and blood pressure and sweating and all the other involuntary acts."

There was a long pause. Gerhard said, "You're going to tell me I'm wrong."

On the screen, the computer blinked:

"I'm going to tell you," she said, "that you've cut too many conferences. You know about autonomic learning?"

After a guilty pause: "No."

"It was a big mystery for a long time. Classically, it was believed that you could learn to control only voluntary acts. You could learn to drive a car, but you couldn't learn to lower your blood pressure. Of course, there were those yogis who supposedly could reduce oxygen requirements of their body and slow their heartbeats to near death. They could reverse intestinal peristalsis and drink liquids through the anus.

But that was all unproven - and theoretically impossible."

Gerhard nodded cautiously.

"Well, it turns out to be perfectly possible. You can teach a rat to blush in only one ear. Right ear or left ear, take your pick. You can teach it to lower or raise blood pressure or heartbeat. And you can do the same thing with people. It's not impossible. It can be done."

"How?" He asked the question with unabashed curiosity. Whatever embarrassment he had felt before was gone.

"Well, with people who have high blood pressure, for instance, all you do is put them in a room with a blood-pressure cuff on their arm. Whenever the blood pressure goes down, a bell rings. You tell the person to try to make the bell ring as often as possible. They work for that reward - a bell ringing. At first it happens by accident. Then pretty soon they learn how to make it happen more often. The bell rings more frequently. After a few hours, it's ringing a lot."

Gerhard scratched his head. "And you think Benson is producing more seizures to be rewarded with shocks?"

"Yes."

"Well, what's the difference? He still can't have any seizures. The computer always prevents them from happening."

"Not true," she said. "A couple of years ago, a Norwegian schizophrenic was wired up and allowed to stimulate a pleasure terminal as often as he wanted. He pushed himself into a convulsion by overstimulating himself."

Gerhard winced.

Richards, who had been watching the computer console, suddenly said, "Something's wrong."

"What is it?"

"We're not getting readings any more."

On the screen, they saw:

Ross looked and sighed. "See if you can get a computer extrapolation of that curve," she said. "See if he's really going into a learning cycle, and how fast." She started for the door. "I'm going to see what's happened to Benson."

The door slammed shut. Gerhard turned back to the computer.