On the first look, Artificial Intelligence has made an enormous progress in the last decade. Today's robot are more powerful than in the past and thanks to deeplearning many new domains will be explored in the future. At the same time a major bottleneck is visible which slow down the development dramatically.
The bottleneck in robotics becomes visible if the aim is to automate existing problems, increase the productivity and replace human workers with robots. Typical projects in which robot based automation has failed in the past, and will fail in the future are self-driving cars, automated check desk in the supermarket, household robots and factory automation. Since the beginning in the 1970s many projects were started from different companies and what they have in common is that in 100% of the cases the robots failed in increase the productivity. This kind of description might be surprising because on the first look the problem seems to be easy to solve. Suppose, that a human worker is doing a job in a factory. Wouldn't it be great if a robot arm can be programmed so that the machine can do the task at lower costs? Sure it would be great, but trying out so in reality will become more complicated than the engineers expected first.
What all factory robot projects have in common is, that they do not replacing human work with robots, but they are creating a vision of how factory automation can be done in tomorrow. Pictures, videos and textual description of robots in a factory are not describing how to build such machines, but they are drawing a vision who the future will look like. In reality, robots are useless in a factory. They are not able to increase the productivity by a simple one percent.
The good news is, that in a different domain robotics and artificial intelligence is very successful. It's in university projects which are created around existing project. The first notable example was the micromouse challenge which was developed in the 1980s to test out new robotics pathplanning algorithm. Some years later the Robocup challenge was developed with a similar purpose. A modern domain is called “mario ai” challenge in which an artificial intelligence has to control a game character. In such a synthetic challenge today's hardware and software is very successful. It's possible to create robots which can fulfill the requirements. And it make sense to improve the technology further and develop more advanced competitions.
What is important to know is, that acacemic robotic challenges and real world applications have nothing in common. The paradox situation is, that a state of the art humanoid robot is able to fulfill a complex challenge which includes biped walking and pickup an object. But the same robot is not able to fulfill a more simpler task in a factory. The reason is, that the task1 was constructed as a testbed to develop new robots, while the task2 (in the reality) was there before the robot was build.
The reason why robotics challenges and real world application are two different things isn't researched very well. A naive assumption is, that a real world problem can be reduced in the difficulty to a simple pick&place task in a factory and at the same time the synthetic robot challenge can be increased so that the programmer have to program advanced robots. Unfortunately, both domains are not fitting together. They are two contrasting competitions. That means, with the advent of modern programming challanges like micromouse and robocup, robots haven't become more practical but their ability to increase the productivity in reality has become smaller than ever.
An impressive negative example is that latest generation of a humanoid household robot which is able to clean a table by it's own and can open the refrigator. From a technical point of view, the system is more advanced than any robot before. It is using neural networks, is able to balance during the biped walk and can understand natural language. The surprising fact is, that this sort of technology fails if the goal is to use it in a practical task. That means, the robot is not able to increase the productivity but it will get lower.
The paradox situation is, that from a technical point of view, Artificial Intelligence is available and it is working better than before. Today's technology allows to build advanced vision systems, motion planning software and speech recognition devices. At the same time, the mentioned technology is completely useless. It can only be used in synthetic challenges but it fails to automate existing tasks. That means, from a social perspective a robot isn't working as a tool which can be used with an advantage, but a robot creates a new sort of problems, not available before.
The social description of a tool is, that it will improve something. For example, a hammer allows to fulfill a task. Using a hammer is better than not using the tool. A robot is working different. A robot has more in common with education. It is interesting to investigate who the robot works, but if the robot is ignored it's not a problem. Building a house with robots will take the same time not using a robot.
The problem is not how to program an Artificial Intelligence, but the task is how to use Artificial Intelligence as a tool. The hypothesis is, that this kind of purpose isn't available. Either the machine is a tool, or it's artificial Intelligence. It's not possible to combine both. The distinction can be drawn very precisely. It's the difference between the forth comuter generation which was from 1970-1980 and the fifth computer generation which was planned for the time after 1980s. Forth generation computer technology is a tool. It can be used to increase the productivity. In contrast, Artificial Intelligence and robotics is located in the fifth computer generation which is not a tool, but it flips the situation. That means, fifth computer generation is mostly an academic discplines which tries to educate humans, but not the other way around. Most expert systems and robotics are nothing else than computer based training applications. The idea is, that a human sits in front of the monitor and learns something from the software. Using the technology for increasing the productivity in a factory isn't possible.
The surprising fact is, that with improved artificial Intelligence it will become much harder to utilize a robot as a tool. If the robot contains more advanced technology for example powerful vision algorithms, it will become obvious that the robot is useless under all situations. The system is working great in synthetic challenges but it won't help factory workers to increase the productivity.
No comments:
Post a Comment