© 1975-1979, 2008 Robert A. Freitas Jr. All Rights Reserved.

Robert A. Freitas Jr., Xenology: An Introduction to the Scientific Study of Extraterrestrial Life, Intelligence, and Civilization, First Edition, Xenology Research Institute, Sacramento, CA, 1979; http://www.xenology.info/Xeno.htm

16.4.1  Artificial Intelligence

Intelligence may be broadly defined as any system capable of processing information -- a negentropic ordering process of higher capacity than life itself. It is of critical significance that the actual physical form of the system is of comparatively little importance to what is taking place. The ability to reason -- to think -- is not a property of biological organisms alone. Rather, it is a property of specialized, highly-organized, complex patterns. Whether these patterns find expression in biological, artificial, extraterrestrial or even nonmaterial form is largely irrelevant to the fact of intelligence.

Time does not permit a detailed discussion of the arguments favoring the proposition of intelligent thinking machines. The objections are many, but their refutations are sound.* Projecting the advances in human computer technology over the past century into the century ahead, many leading scientists are now predicting the creation of artificial machine intelligence of extremely high order. And if we can do it, so can ETs.

Many decades ago an English mathematician named Alan Mathison Turing defined a crucial test to prove the existence of machine intelligence.^2667,3698 According to the "Turing Test," a computer is intelligent if a man is unable to distinguish it from another human being by talking to it over a teletype communications link. This is the ultimate objective test: If an entity speaks and acts like a man, presumably it is just as intelligent.

To date, no machine on Earth has passed the generalized Turing Test. However, the Test has been passed by computers restricted to limited areas of discourse or to certain well-defined human emotional or psychiatric states.

For instance, in the early 1970’s Dr. Kenneth Mark Colby at Stanford University created a program called PARRY that imitates the verbal behavior of a human paranoid. The system was designed to give psychiatry students useful experience in talking to severely mentally ill patients. PARRY is supposedly 28 years old, unmarried, and a postal clerk. He is sensitive about his looks, education, religion and sex. He once attacked a bookie for not paying a bet at the racetrack, and he’s afraid that the underworld is out to get him. A sample conversation with PARRY follows (his words in capitals):

Are you connected with the Mafia?	YOU KNOW, THEY KNOW ME.
How do they know you?	THEY SPY ON ME.
Why do they?	THEY WANT TO GET FALSE EVIDENCE TO FRAME ME.
What do they have against you?	THEY ARE OUT TO GET ME.
What for?	I HAVE INCRIMINATING EVIDENCE AGAINST THE MAFIA.
Are you sure of all this?	YOU DON’T BELIEVE ME?
I am doing my best.	(SILENCE)
(Silence)	CAN’T YOU THINK OF A DECENT QUESTION TO ASK?
You seem hostile to me.	SOME OF YOUR REMARKS ANNOY ME.
I am sorry.	FORGET THE EXCUSES, YOU GUYS ARE ALL ALIKE.
Don’t you trust me?	I HAVE GOOD REASON TO BE THIS WAY.
Okay, tell me more about the Mafia.	THEY KNOW WHAT I’M DOING.

Can PARRY pass the Turing Test? Forty-eight psychiatrists were given paired transcripts of conversations with the computer and with a genuine human paranoid mental patient. Only half of the doctors could distinguish man from machine. Realizing that psychiatrists might not be able to recognize the subtle cues that might tip off a computer science expert, Dr. Colby sent similar transcripts to a hundred members of the Association for Computing Machinery. Of the 67 respondents, 32 guessed correctly and 35 guessed wrong.²⁶⁷⁰ On his own, very limited turf, PARRY passed the specialized Turing Test with flying colors.

Besides mimicking human psychoses, computers have been programmed to play intellectual games as well as men. Machines are now World Class checkers players. In 1965, A. L. Samuel estimated that only a dozen of the world’s checkers masters could defeat a program he’d written.²⁶⁷³ Far more challenging, however, is the complex game of chess.

Attempts to devise computer programs to play chess have been underway for decades.^2674,2675 In August of 1968, David Levy, then Chess Master of Scotland and later International Master, made a bet of $2100 with several computer scientists that no machine program could beat him at regulation chess within ten years. In the early 1970’s, Lawrence R. Atkin and David J. Slate of the Computation Center at Northwestern University wrote "Chess 4.5." This program held the United States computer chess championship title in 1977.

In the Spring of that year, Levy sat down with Chess 4.5 to play a regulation 40-moves-in-two-hours game. Levy won, but the machine made him work hard for it. Then came the real shocker. After the regular match was concluded, the two contestants played an off-the-record "blitz" game in which each side was allowed only 5 seconds per move. Chess 4.5 won handily.

Experts estimate the official ranking of Chess 4.5 in regulation play to be about 2000-2200 points on the standard scale, as compared to 2325 for Levy and 2690 for the current World Champion. It is believed that in "blitz" chess the computer improves its rating by 200-400 points over human opponents.²⁶⁷⁶ Thus, in blitz matches at least, Chess 4.5 is already at the grandmaster level.

Earthly computers may be quite clever, but how much can they really understand what is communicated to them? Dr. Terry Winograd at MIT has developed an amazing machine program, named SHRDLU. This complex system was designed for the express purpose of understanding English.²⁶⁹⁶

Like a small child secluded in a playroom, SHRDLU inhabits a world that includes a toy box, building blocks and pyramids of various sizes colored blue, green, and red. Mechanical arms and mobile camera eyes give SHRDLU the ability to stack or assemble the blocks in any desired configuration on the playroom floor. Most important, the computer responds to commands and gives answers to questions in an intelligent manner.** Extracts from conversations between Winograd and SHRDLU (in capitals) follow.

SHRDLU can understand spatial relationships:

How many blocks are not in the box?	FOUR OF THEM.
Is at least one of them narrower than the one which I told you to pick up?	YES, THE RED CUBE.
Is it supported?	YES, BY THE TABLE.
Can the table pick up blocks?	NO.
Can a pyramid be supported by a block?	YES.
Stack up two pyramids.	I CAN’T.

SHRDLU can identify its own motives:

Had you touched any pyramid before you put the green one on the little cube?	YES, THE GREEN ONE.
When did you pick it up?	WHILE I WAS STACKING UP THE RED CUBE, A LARGE RED BLOCK AND A LARGE GREEN CUBE.
Why?	TO GET RID OF IT.
Why did you do that?	TO CLEAN OFF THE RED CUBE.
Why did you clear off that cube?	TO PUT IT ON A LARGE GREEN CUBE.
Why did you do that?	TO STACK UP THE RED CUBE, A LARGE RED BLOCK AND A LARGE GREEN CUBE.
Why did you do that?	BECAUSE YOU ASKED ME TO.

SHRDLU can assimilate novelty in its environment:

Is there a steeple?	SORRY, I DON’T KNOW THE WORD "STEEPLE."
A "steeple" is a stack which contains two green cubes and a pyramid.	I UNDERSTAND.
Are there any steeples now?	NO.
Build one.	OK. {the machine does so}

SHRDLU makes assumptions to help resolve ambiguities in speech:

Call the biggest block "superblock."	OK.
Have you picked up superblock since we began?	YES.
Is there anything to the right of the red pyramid?	YES, FOUR THINGS: THE BOX, THE RED CUBE, THE BLUE PYRAMID, AND THE BLOCK NAMED SUPERBLOCK.
Why did you drop it?	BY "IT," I ASSUME YOU MEAN SUPERBLOCK... I CAN’T EXPLAIN A NONEXISTENT EVENT.

Can Winograd’s program really think? Is it truly intelligent? We aren’t permitted to dismiss SHRDLU solely because it's "just a computer." The Turing Test for machine intelligence is an objective one: If the entity appears to be thinking, Turing maintained, it is pointless to deny that it is thinking.

Although SHRDLU passes the Test only in a limited way, recent developments in the artificial intelligence field have clearly demonstrated that computers can be taught to interpret meanings by the same processes that humans use.²⁸⁵¹ Dr. A.M. Andrew, cyberneticist at the University of Reading, England, predicts:

Turing’s test will be passed…with no restrictions on topics of conversation or manner of reply by the year 2000 A.D. However, even then the computer will seem like a person behaving rather stiffly and refusing to be drawn into small talk. Perhaps by 2050 A.D. a computer will seem to be someone with whom a joke can be shared, and with whom the conversant identifies to the extent that it becomes important not to hurt the other’s feelings.²⁷⁰⁷

It appears not only possible but probable that many different sentient extraterrestrial races will develop advanced artificial intelligences. It is not yet clear exactly how complex these systems must be. Dr. Marvin Minsky of the MIT Artificial Intelligence Laboratory claims that 10⁶ bits would probably be enough to create true intellect -- provided they "were all in the right place."²² Winograd’s program has about this many -- a million bits -- and the spunky Viking Mars Lander computers carried preprogrammed instructions amounting to a few million bits. In the animal world, this would correspond roughly to the intelligence level of amphibians such as frogs. Says Minsky of Winograd’s SHRDLU:

We see here a computer program that has a small but noticeable fraction of the intelligence of humans. The fraction is somewhere between 10^-6 and 10^-1. I cannot conceive that it would take 10¹² bits {mammal and primate brains) to hold a superintelligent being.²²

Some scientists believe that intellect is the direct consequence of the enormous intricacy of interactions among ten billion active neural components. Experiments performed by Dr. S.A. Kauffman at MIT in 1961 lend some support to this notion. Kauffman wanted to know what would happen if a large number of arbitrary computer components -- "electronic gates" -- were connected to each other at random, with inputs and outputs linked higgledy-piggledy throughout the network. With 100 units, one might suppose that approximately 2¹⁰⁰ (or 10³⁰) different states would be possible, thus rendering the system totally unpredictable. But Kauffman discovered that for a 100-element network there are rarely more than ten distinct cycles of about ten transitions each.²⁶⁷⁸ This result has since been confirmed by others,¹⁷⁸⁵ and demonstrates that intelligence may be a necessary adjunct to complexity.

If alien electronic artificial intellect is possible, how physically small might it be? The theoretical lower limit of cell size is about 400 Angstrom, a bit smaller than the tiniest known living organism (the PPLO). A brain with 10¹⁰ neurons of this size would neatly fill a minute cube one-tenth millimeter on a side.

But artificially designed alien microbrains theoretically could be vastly smaller still. Using molecular electronics with components on the order of 10 Angstrom in size, 10¹⁰ microneurons could be packed into a space of a few microns. This is small enough to hide inside a bacterium, a fact which may have several very interesting consequences.²⁸⁷³ Also, if a brain the size of the head of a pin were constructed, it could house as many as ten million times as many neurons as a single human brain. Alternatively, an intelligent space probe the size of a grapefruit might carry a "city" of billions of advanced cybernetic intellects.

The great late Princeton mathematician von Neumann once calculated the power consumption of brains designed with maximum efficiency, using thermodynamic criteria.¹⁷²⁶ There is a certain minimum amount of energy that must be expended to accomplish a single "binary act," or simple decision, within a brain. In Table 16.4, this value is compared both with human biological neurons and with a variety of modern electronic devices. The relative efficiency of artificial devices is just now passing that of the biological ones, here on Earth, but terrestrial technology clearly still has a long way to go.

Intelligent alien robots may have more closely approached, or already achieved, the ultimate thermodynamic limit of cerebral efficiency.

* Cynics and the otherwise unconvinced are referred to Armer,⁹⁵⁷ Clarke,^55,81 Cosma,⁸⁹⁶ George,⁹⁵² McCarthy,⁸⁵ Michie,⁹⁵³ Puccetti,^71,977 Putnam,¹⁸⁰³ Raphael,²⁶⁸⁷ Rose,⁵⁸³ Sagan,^318,2552 Turing,⁹⁵⁵ and Wesley.¹⁷¹⁷

** The system, while not perfect, is capable of carrying on a real-time discourse. Analysis of and response to each sentence requires from 5-20 seconds.¹⁷⁵

Last updated on 6 December 2008