Back to Contents
You all know the feeling. You have spent what seems like an hour standing in a long line at the post office. You see only three clerks serving customers, even though you see enough spaces for twice that many, and one of those clerks decides to take a break, leaving only two of the slowest people on the planet to serve customers, who always have the most complicated transactions this side of Wall Street. You can almost see the steam coming out of peopleís ears.
In fairness I must add that I have worked as a retail clerk in a small market, a job that has many similarities to that of postal clerk. I found the job tedious, boring, and demoralizing. And that happened before the advent of scanners, which make the job less challenging and, therefore, more tedious and boring. In short, in the occupation of retail clerk we have the perfect job for a soul-less machine. The more so because we generally do not rank retail transactions among our fondest memories of human interaction: the crisp efficiency of a machine in such encounters would certainly compensate the lack of human warmth.
In the evolution of robots into fully autonomous, all-purpose, human-like machines, the development of robotic clerks marks a significant step. But it also gives us a relatively easy step: clerks perform routine and repetitive tasks, just the kind of activity we want computers to carry out for us. And for the most part creating robotic clerks will require little actual invention: many of the parts of the robot already exist. We need only put the parts together and then develop the relevant software to run it.
Start with a motorized wheelchair. Clerks typically work in places with smooth floors, so that gives us a good foundation on which to build these simple robots. The chair will carry the computer that acts as the robotís brain, a radio for the robot to communicate "telepathically" with a supervising robot or the Internet, appropriate sensors, and a pair of manipulators that can mimic the action of human hands.
Certainly assembling a computer and a wheelchair so that the computer controls the movements of the chair doesnít represent a daunting challenge. Technicians have already built computer-controlled vehicles.
Beginning in 2004 the DARPA (Defense Advanced Research Projects Agency) Challenges led to the development of robotic cars that could drive themselves, unaided by humans, over courses that involved 147 miles of unpaved desert roads and those challenges ended up in 2007 with the robot cars maneuvering in urban traffic. Similar challenges could result in the development of robotic clerks of the kind I described above. The value of such clerks to the Department of Defense lies not in their ability to engage in combat (they wonít have such an ability), but in their ability to replace humans in the tedious jobs necessary to run the bases that keep the Army, Navy, Marines, and Air Force working smoothly and efficiently, freeing up humans to take on the more important roles in the military endeavor.
Unlike simpler machines, the robot needs to see and understand what itís doing. We can give robots the sense of vision by using an artificial retina based on a charge-coupled device similar to those used by astronomers. Lenses will project a two-dimensional view of the external world onto the CCD. The challenge for our robotmakers, as yet unmet, consists of giving the robot the ability to pick certain arrays of light, shadow, and color and to associate them with the word that names the object that they represent.
Neural nets provide the best possibility of becoming a system that can pick an image out of the pattern of light on the robotís retina and associate it with a word that names it. To give a robot vision we need to accomplish two tasks: we must train the robot to recognize an object and then we must train the robot to pick the object out of a noisy background. Training a robot to find a dirt road in a desert (as in the DARPA challenge) is difficult certainly, but training a robot to find the canned peas in a grocery store should be relatively easy.
Training a neural net is a little like creating a hologram. Once the neural net is built, either as hardware or as a simulation in a computer, we create a kind of interference pattern by feeding both the required input (the image of a can) and the desired output (the word "can") into the net at the same time. The programmer would then test the net by showing it a variety of cans in different orientations. If the net yields the correct output of "can", the programmer instructs the computer to increase the numerical weights in the net, in essence hardening the hologram. If the net yields the wrong output, the programmer instructs the computer to decrease the numerical weights, in essence softening the hologram for further shaping. The programmer then repeats the training regimen. Once the program is ready for use, it can then be copied directly into other robots. In that fact we see one of the advantages of using electronic computers.
Getting a robot to understand spoken language is easy by comparison and the robot needs to be able to that as well. Programs that understand spoken language already exist, however imperfectly, so we can take the robotís ability to use language as a given.
The robot also incorporates a digital computer that gives it the ability to process the knowledge that it needs to do its job. The neural net thus acts as an interface between the computer and the outer world.
All of this work obliges the robot to expend energy, which it will draw from batteries. When the batteries run down, as they will eventually do, requires that the robot recharge them. Recharging the batteries, for the robots that donít merely sit at a workstation (as the postal clerks tend to do), occurs when a robot comes to a recharging station. Instead of simply plugging into the station and waiting for its battery to charge up, the robot would exchange its depleted battery for a fully charged one.
Of course, you wonít see these machines at Nieman-Marcus, but they will take over Wal-Mart, Target, and other mass-market retailers. You will see them at your local supermarket, working the checkstands and, late at night, stocking the shelves.
That latter task is ideal for our first autonomous robots. All the robot has to do is to find the appropriate location on a shelf, remove the old product, clean the shelf, place the new product on it, and then replace the older product in front of it (grocers call this practice rotating the stock).
But now we have created for ourselves a serious social problem. We will close off an entire class of employment for humans. What happens to those humans?
What happened to elevator operators? The advent of the automated elevator put them all out of work. What happened to them after that? When we replace workers with machines, we generally replace workers who perform the simplest jobs; therefore, we unemploy workers who donít have the competence to perform more demanding jobs. A man who can operate an elevator may not have the ability to learn how to drive a bus.
So what shall we do about the displaced workers? We can choose between two basic possibilities:
I; not paying the robots will lead to a drop in prices and, eventually, to a money-less society. But the deflation that will initiate that change will still cause great social damage. And we may ask how the displaced workers will obtain the money that they need to support themselves. As things stand now, after the unemployment insurance runs out the workers would simply be abandoned to their fate.
II; paying the robots and giving the pay to the displaced workers as pensions will continue the money-based economy, but will eventually free all people to do what they want to do, to do what they do best, to do things that have no economic value, but have other kinds of value.
Which of those choices we make depends upon what kind of people we are. Former president Jimmy Carter once noted that we like to think of these United States of America as a Christian nation, but if we do not care for the poor, especially through our government and its agencies, then we are no such thing. Letís hope that when the time comes the vast majority of Americans demand that we implement the second of the choices above in accordance with our reputation as a fundamentally good and decent people.
Braitenberg, Valentino, "Vehicles: Experiments in Synthetic Psychology", 1984, MIT Press, Cambridge, MA, ISBN 0-262-02208-7.
This is a charming and easily readable book, whose contents also appear in truncated form in the Computer Recreations column in the March 1987 issue of Scientific American.
Back to Contents