Robotics researchers have been pushing the envelope for the last 30 years since the inception of “artificial intelligence”. The basics of artificial intelligence programming is the modeling of human expertise and mimicking human behavior in a variety of circumstances.
One aspect of artificial intelligence gave rise to expert systems. Complex systems like diesel locomotives are very difficult to repair because of the large number of parts operating together. Human experience accumulated after years of working with diesel locomotives needed to be captured in order to prevent each generation from having to apprentice workers over long periods of time in order to learn how to troubleshoot these systems. So programmers in the early days of AI were employed to learn and program the diagnostic procedures developed by skilled workmen over many years.
These programs were very successful. But in no way do they replace human intelligence and insight. This is simply an example of subtlety in programming a specific area of human experience. Speech recognition continues to be a challenge after decades of effort, limited to transcription applications and simple material handling instructions.
Another area that came up was large scale logistical mapping, another Expert System. What is the most economical way to use airplanes to transport people around the US? When you think of a large air carrier and the number of airplanes, flights, destinations and how they might be mapped together to get the best use out of the airplanes, it is a problem that is too large and complex for a single human to work with. Enter the expert system programmer.
But in none of these cases can a computer program exceed the boundaries of it’s programming. Can the autonomous Jeep get from it’s starting point to it’s destination? Yes. With many man-years of programming and a vast array of computing power, proper deployment of sensors and actuators, and a lot of stored energy.
Can the autonomous Jeep perform any other task? No. Regardless of the sophistication, the machine cannot exceed the boundaries of it’s programming.
Can we teach machines to learn? So far, only in the most crude and rudimentary way. But the course of the learning is again bounded by the programming.
And again, I will defer discussion of true intelligence or consciousness.
But what robotics can do to expand it’s usefulness is to mimic simple human tasking where it is cost effective and where the robot can “outproduce” or exceed the precision of a human. Robotic welding, for example, has reached the point where a basic robot welding cell is less than $50,000. So the cost of entry, the learning curve and complexity of implementing a welding robot cell in a small production facility is very reasonable.
Will robots be used in “human service” applications? Sure. “Robot, vacuum my living room” No sweat. We can already do that with a Roomba only it doesn’t have voice recognition yet. We have robots that can mow the grass in the front yard and avoid shrubs and trees. Very cool.
Will we have robot servants like C3PO in Star Wars? Hopefully more intelligent, C3PO was kind of dumb. Simple tasks like serving a drink at a bar? Yes, that’s been done too, although it doesn’t have philosophical conversations with customers.
Will robots be able to provide basic care in hospitals and for the elderly? Anything is possible. It will come down to how far we can push the envelope of programming, safety and return on cost. Certainly we get robots to get a cold beer from the fridge. But if the fridge is empty can it run out to the store and get us a six pack?
Not anytime soon.