Hal is a hardware engineer who sometimes programs. He is the former editor of DTACK Grounded and can be contacted through the DDJ offices.
Let's look back at Germany a while before Gutenberg, specifically, at a new city on the edge of the Schwarzwald, the Black Forest. Specialization of labor, which had been with us from the invention of agriculture, was becoming formalized and legalized amongst the emerging middle class by the Guild system. Trade between cities was becoming commonplace.
But not everybody managed to join a Guild; there was still need for common labor. For example, the Guild workers were too busy and too specialized to go into the forest for the firewood needed to cook meals and heat houses in winter. In this small city alone, 80 woodsmen (no woodswomen) earned a living cutting firewood in the forest, bringing the wood into the city, and hawking it in the streets to the populace. One of these woodsmen was a man we'll call "Clumsy."
One day Clumsy's ax struck a knot, and the ax was deflected, striking Clumsy's left foot just below the ankle and ending Clumsy's career as a woodsman. By whatever happenstance, Clumsy survived this industrial accident. While healing, if that's the word, Clumsy had plenty of time to think, and it turned out he was good at that.
So Clumsy invested his life savings in four chain saws, the only ones available, and hired four woodsmen to operate them. He had a custom wagon built with wide, iron tyres, drawn by four oxen, to carry the firewood into the city. And with a large, steady supply of firewood guaranteed, he drew up delivery contracts with individual households. Where the traditional woodsmen spent more time in distribution--hawking firewood from the street is a time-consuming activity, as is carrying the firewood into town on donkey's back--Clumsy's system was highly efficient and could handle the substantial output generated by the four chain saws.
Clumsy's company--what we would today call a company--employed four chain-saw men, a drover, and a youthful drover's assistant to help load and unload that four-ox wagon. These 5.5 employees proved able to produce and deliver as much firewood as 40 traditional woodsmen. And since there remained 75 traditional woodsmen (Clumsy and his four chain-saw men deleted from the former assembly of 80), that meant the city had an oversupply of firewood.
All of you who have taken Economics 101 know that what followed was a short, intense shakeout while the excess 35 traditional woodsmen were forced out of the market. Yep, they had to find some other employment to feed themselves and their wives and children. This economic displacement was Politically Correct: The displacees were all white European males and hence of no concern.
Now we had 40 traditional woodsmen producing half the city's firewood, and Clumsy's 5.5 employees producing the other half. The 5.5 employees generated the same income as the 40 woodsmen, nearly 8 times as much per employee. How should this economic windfall be shared?
It was Clumsy's capital that bought the chain saws, and the improved distribution method was Clumsy's idea. If you think his employees should be paid at the same level as the ordinary woodsmen, with the substantial excess going to Clumsy, then you're a capitalist.
If you think his employees should be paid a substantially higher wage than the regular woodsmen to reflect their improved productivity, then you're a liberal. Clumsy still gets rich, if more slowly.
If you think the economic windfall should be shared amongst all the woodsmen, you're a socialist. And if you think the windfall should properly be shared by all the residents of the city, you're a communist!
I worked out this little scenario in 1965, shortly after taking Economics 101. These days we'd have to include the Greens in our political spectrum--they'd want to absolutely stop the woodsmen to protect the spotted owl--and I don't even want to discuss fringe groups like the syndicalists. How you'd like to distribute Clumsy's economic windfall really does reveal your politics, but in 1965, it proved a very bad idea to go around asking people if they were communists.
Three decades later, this scenario is not a communist detector but a metaphor of our current economic system, with automation causing massive dislocation and unemployment. Yesterday's excess woodsman is today's excess middle manager. What is today's chain saw? Why, spreadsheets and database software. A dozen years ago, Walmart had four layers of management, while Sears had seventeen (honest). Which was more successful? This is not an entirely rhetorical question; if we define success as providing employment for a lot of middle managers and sustenance for their families, then Sears was, back then, more successful.
While middle managers are becoming a threatened species, stenographers, traditionally female, are disappearing under a double whammy: They provided typing services for all those male middle managers. Today, the remaining middle managers almost all use word processors.
When most of you readers think of automation-related economic replacement, what probably springs to mind are the industrial robots on Ford's production lines. Industrial automatons not only don't get sick, take vacations, or go on strike; they also produce more-consistent and therefore more-reliable products. A hand-built automobile is an unreliable automobile. When I was a kid, cars broke down often.
Industrial automatons? The productivity software you have all collectively written to run on personal computers has been responsible for millions of displaced workers. How shall we distribute the excess income produced by a surviving middle manager who uses WordPerfect, Lotus 1-2-3, and dBase IV?
We can't go home again. Nobody wants to un-invent the horse collar so humans can be employed to pull plows. If you try to take away my copy of WordPerfect 5.1, the one I'm using right now, you'll have a real fight on your hands. I'm pleased that my highly reliable car was assembled mostly by industrial robots.
A very few of you may be working on the periphery of robotics. To me, a robot is an autonomous, mobile, industrial automaton whose purpose is to free up human labor; that is, to create unemployment. The first guy I'd like to unemploy is the point man on an infantry patrol (as I write this, American infantrymen have just been sent to the near proximity of Bosnia). But the first robots--very crude and inefficient--are already experimentally deployed, not as infantry. They're hospital orderly/assistants, because hospitals are where the most intense economic activity is found today.
* * *
Now I'd like to introduce you to Dr. Hans Moravec, who has spent the past 25 years working in the field of robotics, and to his book, Mind Children (Harvard University Press, 1988). This book is chock-full of historical information on robotics and its practitioners. For example:
We came out of World War II with many operational analog computers, which aimed guns and bombs and navigated aircraft. This led to the post-WW2 development of "cybernetics," which flourished until the mid '60s, when it was subsumed by the artificial-intelligence community, a product of yet another WW2 development, the digital computer.
Not all activity was confined to the academic community. In 1954, Devol patented the programmable robotic arm (a lineal descendant of the 1801 Jacquard loom). Devol founded Unimation in 1958 to build these arms, and General Motors began using industrial robots in 1961.
John McCarthy was an MIT prof who had a student named Marvin Minsky. McCarthy moved to Stanford University in 1963 in time to found Stanford's Artificial Intelligence Project, later renamed the Stanford Artificial Intelligence Laboratory (SAIL). McCarthy expected to produce a fully intelligent machine in a decade. Sigh. Meanwhile, back at MIT, Minsky's students were connecting TV cameras and mechanical arms to computers.
Stanford had a close relationship to the Stanford Research Institute (SRI), which was generously funded by NASA to develop a remote-controlled lunar rover. With this advantage, Stanford surpassed MIT's robotics group. Following "Shakey" (Stanford's first experimental robot, completed in 1969), SAIL acquired a simulated lunar rover from SRI and used it as the basis of the "Stanford Cart"--the subject of Moravec's PhD thesis.
In the mid '70s, NASA funded Cal Tech's Jet Propulsion Laboratory (JPL) to develop a Mars Rover. Because of the distance to Mars, remote control is impractical even at the speed of light, so the Mars Rover had to be autonomous. JPL developed a prototype, the Robotics Research Vehicle (RRV). Lacking modern microprocessors, the RRV was connected by an umbilical cord to a large computer. Although the RRV was achieving many of its objectives, the project was terminated in 1978 (the 1984 Mars mission was scrubbed).
Moravec, in his 1988 book, expected to see general-purpose robots for the factory and home by the millennium. My take: Follow the money. Housework isn't explicitly valued by our society, and factory workers are a dime a dozen (more or less). Infantry point men can be drafted cheaply; see the Wall in Washington DC. As previously mentioned, the first experimental robots are alive and well and working in hospitals, where today's most intense economic activity is found.
Hans Moravec was a graduate student at Stanford in the 1970s. In 1970, he used a time-sharing DEC PDP-10 mainframe. In 1980, he was still using a DEC 10, this time a KL-10. In contrast to the 1970 machine, it was not very fast for its day (and DEC soon discontinued the product line).
Moravec published an article entitled, "Today's Computers, Intelligent Machines, and Our Future" in the February 1979 issue of Analog (a science-fiction magazine) that compared human computational bandwidth to that KL-10 and noted that computers were increasing in speed by a factor of 10 every seven years (equivalent to doubling every two years). As an Analog reader since 1952, when it was Astounding Science Fiction, I read Moravec's article when it first appeared. By 1988 computers had sped up by more than an order of magnitude, so he wrote the book Mind Children to update and expand on the article. 1995 marks yet another x10 improvement, and, according to Moravec, yet another update is in progress.
In 1980, Moravec left Stanford to become director of Carnegie Mellon University's Mobile Robot Laboratory, a part of the Robotics Institute. (Was he frustrated by the inferior computational capacity then available at Stanford?) His interest in comparing human and machine computational bandwidth continued at CMU and is a principal subject of his book.
Moravec has tried several approaches to estimating human computational bandwidth and comparing that to machine computational bandwidth. The one he focuses on in Mind Children is the visual system. He hopes that some of his inevitable errors will tend to cancel and points out that even an error of two orders of magnitude isn't terribly important when machine capability is increasing an order of magnitude every seven years.
Moravec does another very smart (and nonacademic) thing: He concentrates on computational cost effectiveness, what is commonly called "bang per buck." His projected trend line shows that personal computers will, if present trends continue, attain human parity in computational bandwidth shortly after the year 2030 and that before 2040, human-equivalent machine capability will cost less than $1000.
This also means that, in 1995, personal computers are (by Moravec's estimate) about 200,000 times (37 years) slower than the human brain. Moravec specifically claims to work in robotics, not in AI. Amongst practical folk, AI has a bad name. I would suggest that factor of 200,000 has much to do with AI's bad name and its practical failures to this point. Every seven years we drop a digit off that deficit.
Trends do not necessarily continue forever. Bought any $2/megabyte DRAM lately? (See Figure 1.) In my article, "CPU Performance: Where Are We Headed?" (DDJ, January 1994), I predicted that the long-time trend of computer speed doubling every two years would fall off due to the limits of parallelism. I predicted that about 1996, the new computational trend would double (only) every three years. This trend could continue only as long as the dimensions ("design rules") used to fabricate microprocessors could continue to shrink at the established rate (halving every seven years). Other folks (Nick Tredennick, for one) believe that ways will be found to end run these limits.
So the basic problem--the Moravec prediction that robots replacing human beings will become not just a crisis but "oblivion"--may not arrive as scheduled four decades hence. (Read the book's "Prologue" to find that predicted oblivion.)
Moravec discusses the bottom-up versus top-down approaches to robotics. In top-down, you start with a supercomputer and bolt a TV camera and mechanical arm onto it, then try to figure out how to mount wheels. This is the approach Minsky's MIT group took in the mid '60s. Moravec favors the bottom-up approach used at Stanford, where you start with something very much like the prototype Mars rover--a mobile device that has machine vision and manipulative ability--and bolt a small, fast computer onto it. The top-down approach is favored by the AI community, while the bottom-up approach is a dirty-fingernailed philosophical descendant of cybernetics.
Robots, primitive ones, are here today. As they improve, robots as menial workers will migrate downward into areas of less-intense economic activity, and upward into areas of more-intellectual (less-menial) activity. When household robots clean and cook, and the more cerebral ones write columns for computer magazines, robots will have subsumed all human activity and Moravec's oblivion will be upon us.
Every robot deployed represents a person seeking other work. An enormous number of robots will be deployed in the next several decades. We really don't want to un-invent the horse collar, or the printing press, or the steam engine, or the computer. Will we someday want to un-invent robots?
Mind Children is still in print and is only $8.95 in softcover. I know that Computer Literacy Bookstore (Sunnyvale, California) will accept a special order for this book because that's where I got mine. Best of all, this book is not written by some Sunday-supplement journalist hack but by a leading researcher in robotics. If you have any interest in robotics and/or computer trends, you really should read this book.
Figure 1 Memory prices versus time.
Copyright © 1995, Dr. Dobb's Journal