OPTIMIZED

James Bellinger

The small disk spun blindingly fast as it flew from metal hand to hand to hand around the tripod robot. The Iinterference patterns started to give Dasu a headache. The scan rate for the camera in the lab didn’t match that of the monitor the robot watched.

“This is your star?” Dasu asked. “The triumph of the AI program?” The NSF had been funding the group for six years now.

“Yes indeed,” said Lars. “That monitor—you can’t see it clearly here, of course—he re-built it himself. We gave him the latest Davidson-Harley learning algorithms, and the fastest Minuit-Millipede optimization, and then prompted him with an irritant stimulus–—a monitor with a resolution and scan rate that didn’t match his sensors.”

“He had just the basic electronics learning module, but it took less than an hour for him to figure out how the monitor and the bench tools worked, and in another half hour he rewired it and fiddled somehow with the LEDs. We don’t know what he did yet, but that monitor now runs at five times the speed of the industry’s best!”

“We already knew how to set up learning—this is creativity. Real artificial creativity! Proof of artificial intelligence! No, proof of superhuman intelligence!” Lars warmed to his subject.

“Think of it: a breakthrough in monitor technology in less than two hours! And technology isn’t the only possibility. The sociology department thinks a superhuman intelligence could revolutionize their field, and maybe even solve intractable political problems. The sky’s the limit!”

Dasu looked back and forth from the robot to the colorful press release. “I can’t make head or tail of what it’s looking at. What exactly is that?”

Lars turned to the wall behind them and clicked a control. The large screen glowed on and then displayed a complicated mandala which morphed into spinning tesseracts and then to patterns Dasu had no name for. “We slowed the displays down to study them,” said Lars. “Nothing repeats. Badgett’s team thinks these represent a scan of the electron cloud quantum states in the LED array. If so, we might have a brand new tool for solid-state research! It may take us a while to figure it out—after all, he’s much smarter than we are.”

Eleven years of reviewing research projects had left Dasu with an allergy to “possible exciting breakthroughs”—a fixture in the conclusions of every single report. Almost none of them ever materialized. This, though, looked promising. For years he had been scrupulously fair, and squelched his skepticism about AI and its engineers trying to code descriptions of how they— _think—_ they think. Shortly he’d stand beside these men in front of the cameras and no one would know he’d been wrong.

“Where do those patterns come from?”

Lars said, “They seem to come from an interaction of something the robot beams to it and something generated internally byto the monitor. We have over 500 hours of uninterrupted signal to study, and we’re confident we can find—”

“Wait. 500 hours? Uninterrupted? The robot built that monitor—let’s see,” Dasu flipped pages. “About three weeks ago. About 500 hours ago.” He turned to look at the robot cam. “Has that thing been standing there the whole time? Not doing anything else?”

“Well, we can’t say for sure what he is doing. I assume learning something from the patterns. He’s smarter than we are.”

“But the robot generates the patterns itself.” Dasu stared at the robot. “Just standing there, watching a robot’s movie—–That’s that’s what the breakthrough means?”

Lars urged something about “local optimum” and assured him that they were doing their best to understand the superhumanly sophisticated video, but Dasu barely listened as the robot stood watching and watching and flipping a coin.

Food for Thought

We apply intelligence in domains ranging from designing rockets and how to harvest the best crops, to solving properties of prime numbers and crossword puzzles: very applied to very abstract. Forget for the moment that some prime number studies have had important applications—most don’t.

Is there a proper domain for intellect? Or put another way, can it be inappropriate or unworthy? The Greeks were accused (probably incorrectly, see _The Forgotten Revolution_ by Lucio Russo) of believing that applied research, or anything hands-on, was déclassé. I have met people who on the contrary believe that if there is no application, the study is not worthwhile.

If there are such boundaries shaping the proper operation of intelligence, on what grounds can you figure them out? Are those grounds universal?

About the Author

James Bellinger is a physicist at the University of Wisconsin, where he has worked since April Fool’s Day of 1985. He is happily married with five children.