The Turing Option

“Hardly! We’re just tracing input and feedback. If you look at all those finger extensions under a glass you will see they are clustered in regular bundles. Each bundle contains a tripartite subbundle made up of two optical pickups and a single light source. The pickups are mounted at fixed distances from each other. Does that give you any ideas?”

“Yes — binocular vision.”

“Bang on. In addition to what you might call the eyes in every bundle there are four mechanical manipulators. Three blunt-ended ones for grabbing, the fourth with a knife edge for dismembering. This carves off the insect’s head just before the thing is dropped into the hopper. The bundles work independently — almost.”

“What do you mean?”

“Let me run a film for you and you’ll see for yourself.”

Brian put a cassette into the video, ran it forward to the right spot. “We shot this at very high speed, then slowed it down. Take a look.”

The image was sharp and clear and magnified many times. Rounded metal bars reached out slowly to embrace a foot-long fly. Its wings flapped slowly and ineffectively as it was drawn out of sight off the screen. The same process was happening to an aphid located off to one side.

“I’ll run it again,” Brian said. “This time keep your eye on the second bug. Watch. See the bundle above it? First it’s motionless — there, now it is operating. But the fly didn’t move until it had been grabbed. Do you see what that means?”

“I saw it — but I’m being dumb today. What’s the significance?”

“The hand didn’t try to use brute force and speed to try to catch the fly in flight. Instead, this robot uses real knowledge to anticipate the behavior of each particular kind of insect! When it goes for the housefly, Bug-Off contracts its grasping-bundle as it approaches the fly, making it look to the housefly as though it were moving away from it — until it’s too late for the insect to escape. And we’re sure that was no accident. Bug-Off seems to know the behavior of every insect described in this book.”

Brian handed Ben a large volume entitled Handbook of Insect Ethology, 2018 Edition.

“But how can Bug-Off tell which insect it is dealing with? They all look the same to me.”

“A good question — since pattern recognition has been the bane of AI from the very first day that research began. Industrial robots were never very good at recognizing and assembling parts if they weren’t presented in a certain way. There are thousands of different signals involved in seeing a human face, then recognizing who it is. If you wrote a program for picking bugs off bushes you would have to program in every bug in the world, and size and rotation position and everything else. A very big and difficult program—”

“And hard to debug?”

“Funny — but too true! But you or I — or a really humanlike AI would be very good at bug grabbing. All the identification and reaching out and grabbing operations are hideously complex — but invisible to us. They are one of the attributes, one of the functions of intelligence. Just reach out and grab. Without putting in any complex program. And that’s what is happening here — we think. If there is an AI in there it is reaching out one bundle at a time and grabbing a bug. As soon as the insect is held it turns the grabbing bundle over to a subprogram that plucks it off, brings it to the container, chops it dead and dumps it, then returns to operating position ready to be controlled again. Meanwhile the AI has controlled another bundle to make a grab, another and then another, changing control faster than we can see at normal speeds. You or I could do that just as well.”

“Speak for yourself, Brian. Sounds pretty boring to me.”

“Machines don’t get bored — at least not yet. But so far this is all inferred evidence. Now I’m going to show you something a good deal better. Do you see how Sven is plugged into Bug-Off’s operating system? It is reading every bit of input from the detectors as well as getting all the return control messages. I am sure that you know that the society of the mind, human or artificial, is made of very small subunits, none of them intelligent in themselves. The aggregate of their operation is what we call intelligence. If we could pull out one of the subunits and look at it we might be able to understand just how it operates.”

“In a human brain?”

“Pretty impossible. But in an AI, at an early stage of construction, these subunits can be identified. After analyzing some of the feedback loops in Bug-Off we found a pattern, a bit of a program that could be identified. Here it is — let me show it to you.”

Brian punched up the program on the screen, a series of instructions. Brian rubbed his hands together and smiled happily.