On the first look, Robotics are promising to replace human work with machines. If self-driving cars, delivery robots, and pick&place robots are available all these jobs can be transfered to robots. This would allow to increase the productivity and reduce the costs. According to the vision, it's only a question of the right software which makes a robot drive by it's own and pick&place objects in a warehouse. Current engineers are on a good path towards fully autonomous robots and within 10 years it will become reality.
The problem with this outlook is, that it's not available in reality, but it's a story told with a purpose. The idea is, that in the near future the new discovered power of Artificial Intelligence will make the work easier for humans. This kind of techno-optimism has to be questioned if it make sense.
The amount of arguments against the wonderful robotics technology is huge. The first argument is, that the computer power which is available to the engineeres has increased dramatically in the last 40 years without any effect to the productivity. The service industry today is the same like 40 years ago. The reason why becomes visible if we are focus on concrete robotics.
There are some projects available which are promissing fast results: the first one is the self-driving car and the second one are pick&place robots. Both tasks are heavily requested by modern economy. Most countries in the world have a large amount of freight which is transported with trucks, and nearly all companies have pick&place tasks. Famous examples are supermarkets and the Amazon warehouse. On the first look the deal is very easy. The engineers are developing self-driving cars which will replace normal cars, and they will develop pick&place robots which can replace human workers.
The perhaps most impressing example of failed robot projects is the self-driving car. Many million of US-dollar were spent to this vision. The funny point is, that todays engineers have mastered the technical challenges and at the same time they haven't. On the one hand, a self-driving car is available and it drives great, but at the same time the car won't be ready in the next 50 years. Explaining this paradox is a bit trickey, and we must go into the details.
Suppose, the self-driving car has to be developed from scratch, because the existing software was accidentally deleted. This gives us the opportunity to understand the inner working from ground up and we can critize the project much better. Creating such a car is not very complicated. For the beginning an RC car will do a great job. It gets an onboard computer, a camera, a lidar and a control software. If all the components are working great, the car is able to drive around in a parcurs and will stop at the red light. Additionally, a parking routine was build in. From a technical perspective the car is ready. So the problem of an autonomous car was solved, isn't it?
And here comes the problem. It's not located in the car itself, but the problem is called “human machine interaction”. This problem is described in the literature poorly. If it's ignored, the engieered have the tendency to believe, that their robots will working fine in reality. Colloquial spoken, human machine interaction is about how a certain robot will perform in the reality. Does the car improve the productivity, can it replace human workers?
The answer is very pessimistic. Even if the car will drive alone great, it won't be able to support human drivers. The problem is, that a synthetic robotics challenge in which the newly designed car drives around is very different from a real driving challenge. What humans are expecting from a machine has to do with productivity. They want to press the on button, and then the machine should provide an added value. Robotics cars doesn't provide such feature. What robots are doing is to create a new kind of challenge not known before.
Perhaps we should go a step backward to describe the overall situation. Classical machines from the industrial revolution are built as autonomous machines. They are activiated and no further interaction is needed. A typical example is a steam engine. If enough fuel is availabe, the steam engine will run many years without interruption. What humans are expecting from robots is the same interaction mode. They believe, a robot is some kind of advanced steam engine which can do more interesting work.
In reality, Artificial Intelligence works as a computer based training. An AI software without a human in the loop doesn't make much sense. An autonomous car is such a computer based training. The idea is, that the self-driving car teaches a human driver how to steer the wheel. Somebody may argue that this feature isn't needed, because the human driver knows already how to drive a car. Instead the idea is, that the car is driving alone without a human behind the steering wheel. Unfortunately, this kind of behavior isn't available. To understand why not we have to go back to our imagined RC car which was created from scratch.
IN theory, it's possible to let the car drive alone in the parcurs. If the task is not important and if no humans are in the parcurs, the autonomous car can do so many hours without interruption. The problem is, that real transportation tasks have to do with important delivery tasks and they have to do with humans who are on the road. That means, the task for the car is to deliver an important freight from A to B and on the path there are other cars which are driven by humans. And this requirements is out of reach for the autonomous car. It wasn't programmed for doing so. If the autonomous car should do the task anyway, a human driver is needed for security reason. Exact in this interaction mode, all the existing self-driving cars are operating. The problem is, that the human driver can't be removed, he is needed by the software.
A naive technician may argue that this is only a small problem which can be solved within 1-2 years. The sad news is, that the so called detail problem was researched 40 years ago in the 1980s the first time, and it wasn't solved since that time. In the 1980s the first larger experiments were done with self-driving cars, hospital robots and pick&place robots. Technically all these devices have worked great. The only problem was the man machine interaction. Perhaps we have to again a small step backward.
What is possible with today's technology is to build robots which succeed in synthetic challenges. The engineers can build robots who are playing soccer, parking a car, steering a car and pick&place objects. The problems are starting, if these wonderful machines should be replacing real workers. This is not possible with today's technology.
No comments:
Post a Comment