“Oh?” Elzer raised an eyebrow. “Aren’t you exaggerating just a bit?”
Auberson sagged back into his chair. He looked around the mahogany-lined room at the other members of the Board. “Would somebody please tell this… this high-priced bookkeeper just what the HARLIE project is all about?”
The other Directors stared back, impassive. Auberson had committed a serious breach of courtesy — he had insulted one of them. White-haired Griff, the oldest member of the Board, coughed and looked at the ceiling. Hudson-Smith, down the table, made a show of refilling his pipe. Next to him, young Clintwood took off his glasses and examined them for dust If Aubie was going down the tubes, he was going to go alone. The only one in the room not appreciably cool to Auberson was Miss Stimson, the executive secretary.
After a bit, after he had let the silence make its point, Dome, the Chairman of the Board, took his thick cigar out of his mouth and grunted, “I’m sure you can do it, Auberson. You know more about this piece of hardware than any of the rest of us.” He replaced his cigar and settled himself in his chair.
Auberson didn’t like the emphasis on “piece of hardware.” Didn’t they understand? HARLIE was more than that, much more. “All right,” he said. “I will. The HARLIE project is the logical extension of Digby’s work with the variable brain path—”
“The variable brain path?” asked one.
“The Mark IV judgment unit. Instead of base two, it uses base twelve. With compaction we can increase its precision by a power of twelve for each stage. First stage compaction is twelve squared, second stage is twelve cubed. Third stage compaction gives us twelve to the fourth power, or 20,736 possible choices.”
“You’ve lost me,” said Elzer. “Now tell it in English.”
Auberson suppressed an impulse. He forced himself to be calm. “I assume you mean one-syllable words?” He didn’t wait for an answer. “Binary code means your machine can make only two possible decisions — on or off, ‘yes’ or ‘no.’ There’s no possibility for ‘mostly yes,’ ‘somewhat yes,’ ‘slightly yes,’ ‘maybe yes,’ ‘maybe yes and maybe no,’ ‘maybe no,’ ‘slightly no,’ ‘somewhat no,’ ‘mostly no’ — there’s no selectivity. It’s either/or. By increasing the number of choices you increase the range of the machine’s judgment. Base three gives you ‘yes,’ ‘no,’ and ‘maybe.’ Base five adds ‘slightly yes’ and ‘slightly no.’ Give it base ten to work with and it’s a pretty selective system. Base ten,” he explained, “is the system most people use.” He held up his hands, spread his fingers and wiggled them. “See? Ten fingers. That’s if you count on them.” Elzer ignored it.
He continued. “We use base twelve in the judgment units for mathematical reasons. It eliminates some of the problems inherent in using tens. The nearest way I can explain it is that twelve divides into neater pieces. Ask a mathematician sometime about the advantages of base twelve over base ten.”
“Got that,” said Clintwood. “How do you do it with computers?”
“You mean the circuitry? I’m not sure I can answer that. I don’t know enough about it.”
“Can you give me an idea?” the younger man asked.
“Well, are you familiar with fluidics?”
“Sort of.”
For the rest of the Board, Auberson explained: “Fluidics is a term used to describe computers or computer circuits based on the flow of a liquid or gas, rather than on the flow of electricity. Just as a transistor uses a small current to modify a large one, a fluidic circuit can use a small flow of liquid to modify a bigger one. There’s an important difference, though. An electric circuit is either/or; either the circuit is on or it’s off. With fluidics, however, you can vary the force of the modifying flow and vary the modification of the bigger. You can push the ‘current’ to be modified all the way over to the ‘yes’ side or to any notch in between. Because your major flow responds in proportion to the force of the modifying flow, you can have your full range of ‘yes’ to ‘no’ responses.”
“How does it do that?”
“It’s the simplest thing. The major flow, the one to be modified, is forced down a channel, which splits into several different directions. The modifying flow is directed into or against the major flow and deflects it into the desired channel. The pressure of the modifying flow is the variable thing. The harder it pushes at the major flow, the farther over it’s deflected. If the major flow is fast enough, you can vary its response several hundred times a second. What you have is a system that responds with surprising accuracy to the pressure of a fluid in a pipe. They’ve been using fluidics arrangements in industry for several years now, and also in the fuel feed systems of jets.
“The judgment circuit is the electronic equivalent of a fluidic unit. It measures the voltage, or pressure, of an electrical current and responds in degree to it. It’s very much like the way the human nervous system works. If a nerve cell releases a strong enough charge, it’s enough to set off the nerve cell next to it. Our judgment units do the same kind of thing; that’s how we can duplicate the action of a fluidic unit — or more importantly, of the human brain. With hyper-state layering, we can compress the circuitry into a size comparable to that of an equivalent piece of brain tissue.”
There were one or two nods around the table. Clintwood looked up from his notepad. “You used another term. Compaction?”
“Right,” said Auberson. “Compaction is the term we use for giving the unit a second level of judgment circuits. It increases the number of choices by one power of the base number — twelve times twelve gives one hundred and forty-four choices in any given situation. One hundred and forty-four degrees between ‘yes’ and ‘no.’ Want more precision, increase the number of levels. Each level increases the number of choices by twelve times.”
“Doesn’t that run into an awful lot of circuitry?”
“No. We can use the same circuits for almost any level of judgment. All the machine has to do is keep straight which is which. The machine makes a choice, decides it isn’t precise enough, shifts down one level and runs the thing through the same circuitry again. That’s compaction. It allows us to get a high degree of precision with a lot less circuitry. If Handley were here, he could explain it. Don Handley is the design engineer on the HARLIE project.”
“You can’t explain it?” asked Elzer acidly.
“I can explain what I know,” Auberson said, suddenly cautious.
“I thought you knew what HARLIE was. You are the chief of the project aren’t you?”
“I’m a research psychologist not an engineer. Anything I’ve picked up about computers, I’ve had to learn specifically on this project I—” He stopped himself. Justifications wouldn’t do any good here. He’d have to try something else. “Elzer, do you drive a car?”
The little man was startled. “Yes, of course.”
“What kind?”
“A Continental.”
“This year’s, I suppose?”
“That’s right.” He said it proudly.
“You knew that its Thorsen Auto-Pilot was one of our units, didn’t you?” He didn’t wait for an answer — it was a rhetorical question. “It was made possible by the variable-path circuits that we’ve been producing for the past four years and marketing as the Mark IV. Basically, that’s a simplified version of one type of HARLIE function module.”
“You mean HARLIE’s a giant judgment circuit?”
“HARLIE is a human brain — with solid-state circuitry instead of organic nerves. We use the judgment circuits to duplicate the human functions. The important part of the human brain is actually a series of very complex judgment paths. They don’t work exactly the same as HARLIE’s, but close enough. The difference is in mechanisms, not basic principles. If a nerve impulse is strong enough, it can trigger other nerves around it; the number of nerves reporting allows the brain to interpret the strength of the original stimulus. HARLIE’s circuits work the same way. The strength of the ‘yes’ impulses (or ‘on’ circuits) determines the interpretation. Just for HARLIE to complete one thought involves several thousand compacted judgment boxes.”
“Uh, what stage of compaction are HARLIE’s judgment boxes?” Clintwood again.
“It’s adjustable, depending on the precision HARLIE wants to bring to any one problem. Or needs to. It’s a matter of how many times a decision can be subdivided before such precision becomes redundant. He has a judgment unit to control it.”
Clintwood nodded and scratched something on his notepad.
Elzer remained unimpressed. “It’s still a computer, isn’t it?”
Auberson looked at him, frustrated by the man’s inability to understand. “Yes, in the same sense that your brain is equivalent to a toad’s.”
The reaction was immediate, a chorus of disapproving remarks. One voice, Dome’s, louder than the rest, kept insisting, “Here now!-Here now! We’ll have quiet.” As the noise subsided, he continued. “Auberson, if you can’t keep your personal opinions out of this—”
“Mr. Dome — Chairman Dome — I did not mean the comment as an insult to Mr. Elzer. I was assuming that Mr. Elzer’s brain was better, more complex than a toad’s. Assuming that he has an average human brain, he is as far above a toad as HARLIE is above a simplified autopilot judgment circuit.”
The room quieted somewhat. “However,” Auberson went on, “if Mr. Elzer feels that there is not enough difference between his brain and that of a toad, I’ll have to use some other comparison — hopefully one not so open to misinterpretation. Did you get all that, Miss Stimson?”
Miss Stimson, the Executive Secretary, looked up at him, eyes twinkling. She had gotten it.
“There is a significant difference that I might note,” he added, spacing out his words carefully. “HARLIE uses all of his brain…” Auberson waited to see if Elzer would rise to this; he didn’t. “Estimates vary, but we figure that the average human being uses only ten to fifteen percent of his available brain cells. We couldn’t afford that kind of luxury with HARLIE, so we built him to use his total brain capacity. He’s not as complex as a human brain — he has nowhere near the same number of “cells,” — but he can still function quite well at human levels. Building HARLIE taught us quite a bit about the workings of the human brain. In fact, we were surprised to find out that in many ways it’s simpler than we thought it was.
“HARLIE’s the result of a very foresighted decision made several years ago to explore the possibilities of judgment circuitry as thoroughly as possible. I’m sure I don’t have to comment on the wisdom of that decision. An on-off circuit can’t do the things a variable pattern can. It’s only the Mark IV unit that’s given us a serious piece of the computer market. That’s why we have to keep pushing. If we ever want to catch up with IBM — and such a thing is not impossible — if we ever want to catch up, we need to be the front-runner in judgment circuits. We have to continue with the HARLIE project.”
“Why?” asked Elzer. “Certainly we can continue producing judgment circuits without HARLIE.”
“We can — but that’s the sure and certain road to corporate oblivion. Look, the Thorsen Auto-Pilot is a fine little unit; it can’t be disparaged. But it’s only the equivalent of an IBM Pixie Desktop Calculator. It isn’t any more complex than that. If we want to catch up, we have to go after their JuggerNaut Series. That’s what HARLIE was originally supposed to be — the ultimate in self-programming computers.
“When Handley came on the project, though, its direction changed; the goal became even more lofty. Or maybe I should say, the way to achieve the goal involved an even greater challenge than we had originally thought. Look, you have to understand what Don was up to before he came here. He’d been doing research with a neuro-psychology team down in Houston; they’d been diagramming the basic pattern structures of the human brain. Have you ever seen the schematic of a thought? Don has. Do you know how to program a human brain? Don does. That’s what he was working on before he came here. Anyway, when they started to design HARLIE — he was called JudgNaut One then — Handley was struck by the similarity of the schematics to those of the human brain. The basic judgment paths were too much alike for the thought patterns not to be similar.