Here Come the Robots! Bob Seitz
In September, 1976, I published my first personal-computer technology forecast. In it, I predicted a "market explosion" of "home computers" over the next five years. Just six months later (as it turned out), in March, 1977, the first practical personal computers, the Radio Shack TRS-80 and the Commodore P. E. T. 2001 were announced. In 1979, I updated this forecast, projecting the 1989 and year-2000 numbers shown in Table 1 below for a $1,000 computer. Here's what I forecast then for a day 21 years in the future.
Table 1. - Technology Forecast for 1989 and for Year-2000
($1,000) Personal Computers
|RAM||16 Kilobytes||1 Megabyte||64 Megabytes|
|BACKUP||Audio Tape Recorder||Diskette||1 GB Hard Drive|
|PROCESSOR||8-Bit||16-Bit, Multiple Processors||32-Bit, Multiple Processors|
|DISPLAY||32-Column Bl. & Wh||512 X 512, 8-Color||1,024 X 1,024, Full Color|
The term "Multiple Processors" was meant to
refer to the predicted inclusion of separate microprocessor chips
controlling the disk drive, graphics processing, sound, the
printer, etc. The point of this table is that I have had at least
a moderately successful track record at predicting the future of
computer technology. I am now forecasting a similar market
explosion of commercial robotic (automatic) devices over the
course of the next ten years.
Artificial intelligence and its handmaiden, robotics, seem to me to have been two of the last century's more over-hyped concepts. On my bookshelf, I have a 40-year-old book that I picked up at an antique store entitled, "Machines and the Man", dwelling upon the dangers of the imminent age of robotics. Since I'm an antique myself, I remember the way newspaper and magazine articles fretted over the impending wave of overwhelming technological unemployment and unbounded leisure that would be unleashed upon us in the 60's and 70's. What we do with all our spare time? I can't resist quoting the M. I. T. math and cybernetics prodigy, Dr. Norbert Wiener, who wrote,
"Let us remember that the automatic machine... is the precise equivalent of slave labor. Any labor which competes with slave labor must accept the economic conditions of slave labor. It is perfectly clear that this will produce an unemployment situation in comparison with which... the depression of the thirties will seem like a pleasant joke."
So what happened on the way to market? It looks as though robotics is arriving with all the speed of a tired turtle with a bad limp. In his 1991 paper, "The Universal Robot", written for the December 1991 issue of Analog Science Fiction, Dr. Hans Moravec, Director of Carnegie-Mellon University's Mobile Robotics Lab, says it better than I could. He observed that in the early 1970's, researchers at Stanford and MIT began mounting TV cameras and manipulators on wheeled robotic carts and turning them loose in real-world environments. To quote Dr. Moravec,
"What a shock! While the pure reasoning programs did their jobs about as well and about as fast as a college freshman, the best robot control programs took hours to find and pick up a few blocks on a table, and often failed completely, a performance much worse than a six-month old child. This disparity, between programs that reason and programs that perceive and act holds to this day. At Carnegie Mellon University there are two desk-sized computers that can play chess at grandmaster level, within the top 100 players in the world, when given their moves on a keyboard. But present-day robotics could produce only a complex and unreliable machine for finding and moving normal chess pieces.
"In hindsight, it seems that, in an absolute sense, reasoning is much easier than perceiving and acting--a position not hard to rationalize in evolutionary terms. The survival of human beings and their ancestors has depended for hundreds of millions of years on seeing and moving in the physical world, and in that competition large parts of their brains have become efficiently organized for the task. But we didn't appreciate this monumental skill because it is shared by every human being and most animalsit is commonplace. On the other hand, rational thinking, as in chess, is a newly acquired skill, perhaps less than one hundred thousand years old. The parts of our brains devoted to it are not well organized, and, in an absolute sense, we're not very good at it. But until recently, we had no competition to show us up."
In his 1991 paper, Dr. Moravec also predicted that machine vision wouldn't really be practical until portable computers could perform about 1,000,000,000 calculations per second, and that this might be expected to occur in approximately the year 2000. It is now the year 2000 and he's right on the money. The 500-MHz Motorola G4 and the 800 MHz Intel Pentium III chips are currently capable of speeds in this range. If past is prologue, they will appear in laptops before this year is out. He also estimated in 1991 that the implementation of the machine equivalent of full scale human intelligence would require 10,000,000,000,000 calculations per second, or about 10,000 times the 1,000,000,000 calculations per second that we can muster today. He has since revised that estimate upward to 100,000,000,000 calculations per second. However, he observes that a great deal that's useful can be accomplished with far slower computers. He also predicted that mass-produced general purpose universal robots might begin to appear about now.
One of the questions relevant to robots is the definition of what a robot might be. You might conceivably call a dishwasher or an automatic clothes washer a robot. There are various special purpose "robots" in factories, including robotic arms and visual inspection systems. There are even highly automated factories. Two of the robotic devices I've eagerly awaited are the robotic vacuum sweeper/floor scrubber/waxer/polisher, and the robotic lawnmower. Poulan marketed a solar-powered robotic "automower" in 1995 (an electric billy-goat that munches on grass at Epcot).. It costs $2.500 and is limited to 15,000 sq. ft. lawns. To install it, a wire carryig an RF signal is buried around the perimeter of its pasture. It nibbles randomly over the area enclosed by the wire. Still, for mowing a corporate grass patch, it might be worth the cost. One of its problems has been reliability. Another potential problem is that of theft. If there were many automowers around, it probably wouldn't be long before thieves would grab them up and whisk them to a chop shop, where they would be sold for parts. The Poulan automower has a built-in burglar alarm but that might not necessarily stop a determined burglar. The latest of these devices is the "Trilobyte" by Eureka (www.Eureka.com/whatsnew/robotvac.htm). It's a little battery-powered vacuum sweeper that, like the Poulan automower, roams randomly around the room. Eureka hasn't yet marketed this. A price of <$1.000 a unit is rumored.. Trilobytes might be useful in motels or commercial buildings where labor costs are involved. The first robots will probably be special purpose devices like these that will first appear in commercial settings, and a few years later, will emigrate to the home. And within a few years, they probably won't be considered robots but will be regarded in the same vein as automatic washers or dishwashers. Of course, if they start to talk, listen, and follow instructions...
One of the key enabling technologies for these first-generation, special purpose mobile robots is visual navigation. In the December, 1999, issue of Scientific American, Dr. Moravec has updated his 1991 forecast with an article entitled, "The Rise of the Robots". In it, Dr. Moravec observes that commercial mobile robots, utilizing computers at the 10,000,000 calculations per second level, haven't been very popular. Only about 10,000 of these mental midgets are in use worldwide, and the companies that make them are barely hanging on. A similar situation exists with regard to robotic manipulators. Dr. Moravec remarks that the principal class of commercial mobile robots is that of Automatically Guided Vehicles (AGVs). Early versions of these AGVs were wire followers, but later models used bar codes or navigational landmarks such as walls, corners and doorways. They were only marginally successful, however, because they had a lamentable tendency to become trapped in corners, wander away, or fall down flights of stairs, typically with a mean time between failure (MTBF) of the order of one month. Also, they required expensive reprogramming if any changes were required. However, the first automobiles were usable only by home mechanics, and the first airplanes were flying box-kites. The next generation of these robots, based upon 1,000,000,000-calculation-per-second laptop technology, should permit a new round of mobile, special-purpose robotic devices. And by 2003, low-cost versions of the Itanium II processor should support computational speeds of 10,000,000,000-to-20,000,000,000 computations per second, fueling yet another generation of robots. (More about this later.)
Dr. Moravec's Mobile Robotics Laboratory at Carnegie Mellon is working on 1,000,000,000 computation-per-second robots, with a delivery date in the 2002-2003 time frame. And as he mentions, once mass production begins and competition develops, progress will accelerate and prices will fall.
Dr. Moravec sees the decade from 2003 to 2010 as an era of automatic, special-purpose appliances. If the forecast of 100 Gigops-computers-by-2010 is correct, "universal" robots will appear by or before 2010 that can pick up clutter, retrieve and deliver things, take inventory, guard homes, open doors, and later, dust, wash dishes, and make beds. By 2010, I should think that it might be reasonable to expect near-human levels of speech recognition, if not speech understanding. I expect to see limited context-sensitive phone-answering services by or before that time. Once 5,000 Gigops computers are available, perhaps by or before 2020, Dr. Moravec feels that a new level of robots with monkey-like intelligence will be feasible. Finally, by or before 2030, a generation of robots with 100,000 Gigops processing powers might make its debut that might aspire, at least from the standpoint of storage and processing power, to humanlike thought processes.
I have a minority opinion that it may not require full human-level processing power to approach human thinking, and that we might not have to wait for desktop computers that can deliver the kinds of processing speeds we're seeking. First, it must take a certain amount of neural circuitry to walk and to balance. That might not be required for wheeled robots. (Dr. Moravec suggests wheels mounted on legs, so that robots can slowly climb stairs or surmount obstacles.) And animals must be able to recognize a predator in a split-second. Robots might be permitted to take somewhat longer to perform such recognition functions. Second, there are special purpose processors like those found in Sega Dreamcast and Nintendo 128 video games that are, perhaps, an order of magnitude more powerful than those found in PC's. Matrox was advertising a 100 Gigops image processing system several years ago. Given a mass market for robots, it might become feasible to develop federations of such specialized processors for robot brains. One might accomplish a lot with a federation of dozens of 100 Gigops processors when they become available. Also, in 10 years, many human functions such as optical character recognition, speech recognition, facial recognition, etc. will require what will probably seem like minor computer capabilities.
There is one remaining question that might bear scrutiny. How much faster can computers get? The Semiconductor Industry Association's (SIA's) 15-Year Roadmap, assembled by 300 committee members in 1997, calls for 13 GHz on-chip clock speeds, with 3 GHz transmission speed across the chip by 2012. This should support sustained computing speeds of 100 Gigops. (It's worth noting that in 1990, two Intel engineers predicted, for the year 2000, a chip with four microprocessors running at 250 MHz and delivering several Gigops. In reality, we'll probably see a 1,000 MHz uniprocessor delivering, perhaps, 1.5 Gigops by year's end.)
The SIA also has a 25-year roadmap that (understandably) isn't yet publicly available.
At the same time, reaching these speeds is not a done deal. The road is getting ever steeper. However, even if computer speeds plateau, there is always the possibility of using multiple processors to do the job, although heat dissipation and power consumption are issues with which one must contend. In the meantime, computer speeds of 10 to 20 billion calculations per second appear to be in the cards for the 2002-2003 time frame. Intel's first incarnation of their new Itanium processor, due out this summer, is said to afford fixed and floating point speeds as high as 6,000 MIPS (Million Instructions Per Second). Next year's "McKinley" model is expected to run as fast as 12,000 MIPS, and their "Madison" chip, appearing in the 2002-2003 time frame, will be faster yet. My private guesstimate is that, within ten years, we will eventually attain at least the 100 Gigops level in a single microprocessor. At that rate, it would require 1,000 such processors to meet the 100,000 Gigops level--if that's necessary--or 100 such processors to reach the 10,000 Gigops range--if that's sufficient. In the end, it also comes down to costs. However, if robots can reduce manufacturing costs, ever more complex machinery might be manufactured at every-lower cost.
One final note: we should probably never underestimate the power of toys and entertainment. The Lego Mindstorms Kit (www.legomindstorms.com) has opened the door to mass robotics. further whetting our appetities for robots. Methinks it won't be long.