Murray Leinster
Page 26
The process outlined here for a robot mind is exactly the way a human mind works — as far as it goes. But something new has been subtracted. It just happens to be the fact that one can know the shape of a thing, and how big it is, and what it’s made of, and its color and the way its atoms are linked together — and still not have the ghost of a notion what it happens to be or do. One can know everything about an object that the most imaginable senses can tell us, and know no more about it than a baby. Which is the point. After the first few months a baby’s eyes are pretty good. It sees things clearly, but it doesn’t know what they are. Its eyes don’t tell that. Its mind has to do the job.
Well... We’re working on a mind.
I went to an auction some years ago to bid on some books because there was a copy of Parson Weems’ Life of George Washington in one lot. (I got the book and it’s ghastly. I do not believe the cherry-tree story.) The auctioneer put up a small, varnished box and pulled a brass object out of it. I bought it out of curiosity for a quarter. Nobody knew what it was. It was brass and it was made to do something, but it was completely cryptic, and the printed instructions were printed in some completely unknown language. Now, the thing had no telescope or plumb-bob or degree-circle or anything resembling any of those things. And I’d never heard of such an instrument, but I made up my mind that it was a sort of eccentric, patented, impractical level for running levels for ditches. Eventually I found someone who could read the directions. They were Swedish and I’d guessed right. I’m very vain of that achievement though I haven’t the slightest use for the gadget.
Now how the devil would a thinking machine work out a problem like that?
How do we work out such problems? We do, and by a very simple system. We know that a hammer is made to hit, a saw to saw, a knife to cut, a boat to sail, a gun to shoot, and so on. When a new object comes to our attention, we look for its purpose, its function, its use — what a philosopher would call its “act.” We don’t know what a thing is until we know what it does. When we can build that conscious ignorance into a robot mind, it will look for the same things a baby’s mind looks for, and accomplish probably more.
It’s worth thinking over. It’s quite a simple problem, after all. We can make a machine that will inspect something and learn and record that it is a one-inch-192
A P P E N D I X B
in-diameter thing, an iron thing, a flat thing, and a pierced thing — that it has a hole in it. It’s a round, flat, pierced, one-inch object. The machine can discover and apply all those adjectives to it. But it can’t discover a verb or a noun so it can make a really intelligent statement about it.
A human being, looking at that object will discover that it can go under the nut of a bolt to spread the pressure when the nut is tightened. That’s the verb.
The noun follows. It’s a washer.
The key to a robot’s mind’s construction would seem to be simply the discovery and/or recording of the equivalent of verbs and nouns so the robot can know what a thing is by what it does — or, of course, the other way about. If we make a machine do that very simple little trick, performed by human babies the first time they grab for a bottle, we can add as much to the effective intelligence at our disposal, as computers have added to our ability to do sums.
Such a trivial thing! But if you want to acquire a really fine case of pure intellectual frustration, just work on it for a while. Just try to invent a way to key a machine so it will recognize what a thing is, and never miss what it is, and never mistake it for something else. Accomplish that very minor feat, and all generations henceforth will revere you. You will have cracked the problem whose solution will give us thinking machines. A machine, which can do that, will think straighter than a man — and if it can’t do that it can’t think. But it seems that it ought to be so easy!
The difficulty is as idiotically simple as the thing to be done. Think, say, of a boat. When you do, a picture comes in your mind. Maybe your picture is of a Star Class racer. (Mine is.) But an outboard motorboat is a boat too, and so is a sea-sled and a canoe and a wind-up toy and a ship with a mast and a note-paper sailboat and the liner United States and a destroyer and a catamaran and a pirogue. Upon Lake Titicaca in the Andes they make boats out of bundles of straw. They are all boats. But there is not one single thing abut them — as objects, aside from what they do — which is applicable to describe all of them and which a mechanical device can detect or record.
One more. You know what a timepiece is. A clock is a timepiece. So is a wristwatch and a grandfather’s clock and a time-clock in a factory. But also a sundial is a timepiece, and an hourglass and a clepsydra — water clock — and a chronograph and the ammonia clocks that have no moving mechanical parts and are the most accurate clocks we possess. Even the Carbon 14 in organic matter makes any organic matter a timepiece of sorts!
Name a means by which any possible device, examining a Nuremburg egg, would identify it as a pocket watch and therefore in a class of time pieces along with alarm clocks, sundials, sextants — which in one use determines local time —
and a wax taper of King Alfred’s time with hours and quarter hours marked on it to tell the time by its rate of burning.
Looked at that way, it appears that thinking is simply impossible, not only for a robot mind, but for a human one. But we do think. We can identify things with ideas. The process, even, is perfectly clear. It’s simply one that nobody has been able to duplicate. If you work on it, you may hit the jackpot.
It works like this: When I think “timepiece” a picture comes in my mind. It’s rather fuzzy, but in my individual case it does have a clock face. It is not the idea of a timepiece, however, but simply a sort of filing-envelope to contain the idea of a timepiece for use. When I want to think about timepieces I drag that out
“To Build a Robot Brain”
193
and use the idea — not that picture — in my thinking. In the same way, when I think of “water,” I usually think of water as contained in a glass. I use “glass-of-water,” as a file clue, as an index-symbol, as a container for the idea of water, without pretending for a moment that it is actually what I think water is. In the same way I’m apt to think of a bluebird when I think of happiness, because it’s a good symbolic container for an idea. It suffices to hold the idea for use. But I know a clock-face is not essential to a timepiece, or a glass to water, and since I live in the country, any day all summer I can go outdoors and see bluebirds fighting like hell on the lawn. (But, of course, that may be their idea of happiness.)
I’m trying to establish that the idea of a thing — the notion of what something really is; the thingumbob that we use in thinking about things and that a robot mind needs to be able to handle — is not itself a picture. We can store ideas in pictures for convenience, but we know that the containers aren’t the things. And that’s the trouble. A robot mind or a thinking machine is going to have to handle knowledge of what things are, if only because it has to take account of what they do. At its baldest, simplest, barest statement — how could you write down the idea of anything at all? How could you note down the idea “food” so a machine could identify a substance as food? How, again, would you arrange for them to be hunted for in the robot’s brain? (Don’t ask me how we associate ideas! Or find them! But just for simplicity’s sake...) If we knew that ideas range through this variation in size or shape or weight or volume, or if we could detect something about two ideas that a mechanical or electronic device so it could distinguish between them ... well ... we might get started. Maybe you can work something out. But as of now nobody seems able to detect anything about ideas at all. There is definite evidence that they exist, of course, We know them and live with them and dream them and think with them every second we’re alive.
But what kind of gizmo is an idea, anyway.
There are some very definite details — mostly of what they aren’t and do not contain or possess. For one thing, an idea does not contain anything that �
� as a matter of perception or by our senses — we have seen or heard or tasted or smelled. But ideas are contained in things we see and notice — even such unsub-stantial things as magnetic fields. And ideas do not contain specifications of material, but they may be contained in specific material. A ball is an idea, and we can find that idea in round objects of any imaginable solid or liquid, and even some stars are said to be made of mere balls of gas. An idea does not contain a design, though it can be obtained in designs. The idea of a house or a cabin cruiser doesn’t include the idea of a blueprint, to most of us, but anybody can see the idea of a house or a cruiser in a blueprint. Perhaps the most baffling of all facts about ideas considered as things is that they do not even seem to have any parts. Each idea is simply itself. Which is an item you can check with your inner consciousness.
With these rather depressing details to go on, it’s clear why we have trouble imagining a robot mind — a mechanical or electronic device — to deal with ideas.
We can imagine the robot mind, all right, but all imagining is simply a shuffling and reshuffling of things we have in some fashion perceived. If we haven’t perceived something with our senses, in one way or another, we can’t imagine it.
Thus, a blind man who has never seen light can’t imagine it. He can think about it, but not have a picture of it. In the same way we can think about ideas, but 194
A P P E N D I X B
we can’t picture them. So when we try to think about a robot brain to handle ideas, we are trying to imagine something we can imagine, dealing with things we can’t imagine.
Naturally, we have some trouble with the details. But it is certainly possible in theory that somebody might concoct some gadget to perform with ideas what can’t be done with numbers.
After all, ideas are realities. They have effects, and unreal things do not produce effects. Ideas not only produce passes at girls who wear glasses, but cities, wars, and streptomycin packaged ready for use. Nations exist because of them, statues are carved because of ideas. This magazine, as a matter of fact, is printed for the express purpose of giving you perceptions from which you can abstract ideas. And while it is true that ideas are fundamentally in the universe around us, ideas are formally and specifically in our minds. They exist in a certain place — my ideas are in my skull, and your ideas are in your skull — and they have effects, but they haven’t any dimensions or any inside or outside, and they are utterly different from each other. To cut down the discovery need to make a robot brain possible, you can say quite truthfully that the basic need for a thinking machine is simply some way by which it can tell one idea from another.
We do it all the time. It must be quite simple, if only one can get the right approach.
All through this article I have been tackling the problems of a robot mind by comparing the needed process with the observed operation of a human mind.
The system does not work so well when one gets this far. But it could be changed a bit and tried further. Maybe the approach to understanding how a machine could be made to think would be possible if one understood how an animal which could not think became capable of it. I suggest that a machine, right now, can do just about everything an animal’s brain can do. We can make machines to perceive, to recall, and even to scramble recollections and arrive at imagination akin to dreaming — and no more packed with sense. We can make mechanical devices which actually learn by experience and acquire rudimentary conditioned response.
We humans are animals, in a sense. Only we can think, which is all the difference anybody needs. It might be that one could get a clue to building a robot mind if he worked out the process by which — to be respectable one has to say by evolution — an animal’s brain became capable of ideas, as it has done in our case. It is a singularly isolated phenomenon. There are hundreds of thousands of other species of creatures on Earth, but we are the only one capable of thinking in terms of ideas. If there were another creature capable of it, we’d have some keen competition.
The difference between our brains and those of other higher creatures is more of function than of structure. If you can work out the difference in operation, you may make robot minds immediately possible and deserve well of your fellow-citizens.
If the direct approach, of seeking to understand how we happen to be human, does not yield results, you might try still more. You might try to figure out why we are human.
There is only one theory that I know of. It does not offer a solution to the technical problem of making a robot brain, but it is pretty plausible.
You learned it in Sunday School.
Bibliography
In addition to writing as Murray Leinster, Will F. Jenkins often used his real name and, occasionally, the pseudonyms William Fitz gerald, Louisa Carter Lee, and Florinda Martel. Entries are chronological.
Books
SCIENCE FICTION
Jenkins, Will F. The Murder of the U.S.A. New York: Crown, 1946. (Also known as Destroy the U.S.A. )
_____. Destroy the U.S.A. Toronto, Canada: New Stand Library Pocket Edition, 1950.
(Also known as The Murder of the U.S.A. ) Leinster, Murray. Fight for Life. New York: Crestwood, 1949.
_____. Space Platform. New York: Pocket Books, 1953.
_____. Space Tug. Chicago: Shasta, 1953.
_____. The Black Galaxy. New York: Galaxy Novels, 1954.
_____. The Brain-Stealers. New York: Ace Books, 1954. (Ace Double D-79. Also known as “The Man in the Iron Cap,” published in Startling Stories, November 1947.) _____. The Forgotten Planet. New York: Gnome Press, 1954.
_____. Gateway to Elsewhere. New York: Ace Books, 1954. (Ace Double D-53. Part 1 published as “Journey to Barkut” in Fantasy Book 2, no. 1, 1950, complete story in Startling Stories, January 1950.)
_____. Operation: Outer Space. New York: Fantasy Press, 1954.
_____. The Other Side of Here. New York: Ace Books, 1955. (Also known as “The Incredible Invasion.” Published in five parts in Astounding Stories, August–December 1936.) _____. City on the Moon. New York: Avalon Books, 1957.
_____. War with the Gizmos. New York: Fawcett, 1958. (Also known as “Long Ago and Far Away.” Published in Amazing Stories, September 1959.) _____. The Duplicators. New York: Ace Books, 1959. (Ace Double D-403. Also known as
“The Lost Race.” Published in Thrilling Wonder Stories, April 1949.) _____. The Monster from Earth’s End. New York: Fawcett, 1959.
_____. The Mutant Weapon. New York: Ace Books, 1959. (Ace Double D-403, backed with Murray Leinster’s The Pirates of Zan. Also known as “Med Service.” Published in Astounding Science-Fiction, August 1957.) _____. The Pirates of Zan. New York: Ace Books, 1959. (Ace Double D-403, backed with Murray Leinster’s The Mutant Weapon. ) (Also known as “The Pirates of Erzatz.” Published in three parts in Astounding Science-Fiction, February–April 1959.) 195
196
B I B L I O G R A P H Y
_____. Men into Space. New York: Berkley, 1960.
_____. The Wailing Asteroid. New York: Avon Books, 1960.
_____. Creatures of the Abyss. New York: Berkley, 1961. (Also known as The Listeners.) _____. Operation Terror. New York: Berkley, 1962.
_____. Talents Incorporated. New York: Avon Books, 1962.
_____. The Greks Bring Gifts. New York: Macfadden, 1964.
_____. Invaders of Space. New York: Berkley, 1964.
_____. The Other Side of Nowhere. New York: Berkley, 1964. (Also known as “Spaceman.” Published in two parts in Analog, March–April 1964.) _____. Time Tunnel. New York: Pyramid Books, 1964.
_____. Checkpoint Lambda. New York: Berkley, 1966.
_____. Space Captain. New York: Ace Books, 1966. (Ace Double M-135. Also known as
“Killer Ship.” Published in three parts in Amazing Stories, October–December 1965.) _____. Tunnel Through Time. Philadelphia: Westminster Press, 1966.
_____. Miners in the Sky. New York: Avon Books, 1967.
_____. Space Gypsies. New York: Avon Boo
ks, 1967.
_____. Time Tunnel. New York: Pyramid Books, 1967. (Novelization of TV series.) _____. Timeslip! New York: Pyramid Books, 1967. (Novelization of TV series.) _____. The Listeners. London, UK: Sidgwick & Jackson, 1969. (Also known as Creatures of the Abyss.)
WESTERN AND ADVENTURE
Jenkins, Will F. The Gamblin’ Kid. New York: King, 1933.
_____. Mexican Trail. New York: King, 1933. (Also known as “Dead Man’s Shoes.” First published in three parts in West, March 4, 18, and April 1, 1931.) _____. Fighting Horse Valley. New York: King, 1934.
_____. Outlaw Sheriff. New York: King, 1934. (Also known as Rustlin’ Sheriff.) _____. Rustlin’ Sheriff. London, UK: Eldon Press, 1934. (Also known as Outlaw Sheriff. ) _____. Kid Deputy. New York: King, 1935. (Published in three parts in Triple X Western, February–April 1928.)
_____. Black Sheep. London, UK: Wright & Brown, 1936. (Published in Adventure, January 1, 1928.)
_____. Dallas. New York: Fawcett, 1950. (Novelization of screenplay by John Twist.) _____. Son of the Flying Y. New York: Fawcett, 1951.
_____. Cattle Rustlers. London, UK: Ward Lock, 1952.
Leinster, Murray. Guns for Achin. London, UK: Wright & Brown, 1936. (See Collections, below.)
_____. Two Gun Showdown. New York: Astro Distributors, West in Action, 1948 . (Also known as The Gamblin’ Kid.)
_____. Texas Gun-Law. New York: Quarter Books, 1949. (Abridgment of Black Sheep.) _____. Texas Gun Slinger. New York: Star Books, 1949. (Abridgment of Fighting Horse Valley.)
_____. Wanted Dead or Alive! New York: Quarter Books, 1949.
_____. Outlaw Deputy. New York: Star Guidance, 1950.
_____. Outlaw Guns. New York: Star Books, 1950.
MYSTERY
Jenkins, Will F. The Man Who Feared. New York: Gateway Books, 1942. (First published in four parts in Detective Fiction Weekly, August 9–30, 1930.) Leinster, Murray. Scalps. New York: Brewer & Warren, 1930. (Also known as Wings of Chance.) _____. Murder Madness. New York: Brewer & Warren, 1931. (First published in four parts in Astounding Stories, May–August 1930.) Bibliography