From a descriptive perspective it can be shown, that most robotics projects in the reality fail. That means, the new pick&place robot arm isn't able to increase the productivity at the assembly line. Using not a robot but human workers is from an economic standpoint the better choice. What these description doesn't provide is the reason why.
Apart from anecdotes about failed robotics automation projects in the car industry, in hospitals or in restaurants there is need to give reason, why all these projects have failed. A possible explanation has to do with the social role of a robot in a project. There are two possible roles available:
1. robot as a superstar, which is provided in dedicated robotics challenges like micromouse and robocup
2. robot as a tool, which is requested in automation projects in factories and hospitals
The reason why car companies are starting robot projects in the factory is because they are interested in a robot as a tool for improving the workflow. The hope is, that a robot is able to increase the productivity and reduce the costs. Robots are seen as advanced sort of a hammer or a CNC machine which helps the human workers. This social requirement for a robot can't be realized. All attempt in utilizing a robot as a tool have failed.
Only the first role (robot as a superstar) results into a successful project. Building a 2wheeled robot which is able to travel through a micromouse maze, is an engineering problem which can be solved if enough skills are available in the team. Many succesful demonstrations of the task are recorded in the past, and the experiment can be repeated with new hardware and new engineering teams. Sure, it's possible that the robot gets lost in the maze, but this is only a detail problem, which can be fixed with better programming. In general most robotics challenges can be solved within the given time frame. It's important to know, that the social role in all of these competitions is, that the robot isn't seen as a tool, but as the most important subject. So it's a superstar and the engineers have to improve the machine.
The difference between the social roles isn't only an academic one. It has to do if a project becomes economic productive or not. Dedicated robotics challenges in which a robot is the superstar are costing lots of money but they are providing nothing in return. The micromouse who is traversing the maze doesn't fulfill the needs of an external customer, but the robot was created for it's own. In contrast, real automation projects in a company are focussed on customer needs. The car factories likes to sell a car to a customer, and the robot should do a subtask in the production facility. In such a social role the robot fails.
It make sense to describe the situation from an abstract point of view:
1. CNC machines = forth generation computer = social role as a tool = increase the productivity
2. robots = fifth generation computer = social role as a superstar = lowering the productivity
With such a template it's easy to predict the outcome of a certain project. Using a CNC machine in a robotics challenge won't work, because the CNC machine can't be programmed freely. The same mistake is obvious if a robot is used in an automation project in a company with the aim to increase the productivity. Each of the technology has a certain sweet spot in which the device can be used in a meaningful way. It's interesting to know that outdated CNC machines are able to increase the productive, while advanced robotics aren't able to do so. Sure, every factory can test out the hypthesis for themself. It's possible to start new robot project to proof that the thesis is wrong. But according to the known projects from the past, it can be estimated what will happen.
The productivity paradoxon has to do with using a certain technology for the wrong purpose. In most cases, the idea is to utilize an advanced robot for increasing the productivity in a factory. Such projects will fail, this is called productivity paradox. It's possible to avoid the bottleneck in defining the requirements first. Which means, it's possible to increase the productive or to play with robots, but not at the same time.
It make sense to observe successful robot projects in synthetic challenges closely. In competitions like Robocup, the robot is the superstar. The task he should do is given by the rulebook, for example one requirement is, that a team of robots should play a game of soccer. This includes object recognition, pathplanning and teamplay. The most interesting feature is, that most of the participants are successful in the competition. Which means, that the robots are working great, that they are driving by software and that each year the skills become a bit higher. Without any doubt it's possible to program even biped robot in a way that they are successful in the robocup challenge.
The only thing what is a bit surprising is, that these technology can't transfered into other domains. The origin of the robocup challenge was to create a testbed for experiment with new robotics technology with the longterm goal to use the newly acquired knowledge in practical applications, for example in factory automation. The robocup challenge itself runs great. Each year, the teams are become better and lots of new AI related knowledge was written in the papers around the competition. What is missing is the knowledge transfer into real applications.
The prediction is, that this knowledge transfer is not possible. That means, the advanced robot are succeed in the synthetic challenge, but they fail in real applications outside the competition. To understand the reason why it make sense to observe the robot projects in an academic context.
The typical university driven academic project is not motivated by increasing the productivity, but the main purpose is to explore new knowledge. In the standard case, a team of researchers is unsure how to build a biped robot and they are starting the project to develop new biped walking algorithm. If they are trained well, they get after a while the first results and write a paper about the walking robot machine. This paper motivates other researchers to experiment with more advanced robots. From an academic standpoint such projects are producing sense. Because at the end, many new papers were written, and new technology was developed which was not available before. It's important to know, that the needs of university robotic projects are different from automation projects in the reality. That means, the robot in the lab is capable of biped walking and has a built in vision system but the factory automation project can utilize this technology in a meaningful way.
Or let me explain it from a different point of view. Suppose a university team has build and programmed an advanced humanoid robot which was successful in a robocup challenge. Lots of money was invested in the project, and hundred of researchers have supported the project. The problem is, that from the perspective of factory automation all the written software and all the advanced hardware is useless. It won't increase the productivity at the assembly line.
To understand the situation better we have to go back in the 1980s. In that area there was lots of interaction available between universities and factory automation projects. The idea was, to utilize the latest knowledge from the academic domain to increase the productivity in the car industry and build advanced service robot which helps to reduce the costs for the customers. The problem was that most or even all of the university-factory projects have failed. The needs of the factory can't be fulfilled by Robotics-experts, and the latest robots developed in the university are useless for factory automation. As a consequence the collaboration has stopped.
The upraising of synthetic robotics challenge is a sign that the university driven robotics community has built it's own challenges. The aim is no longer to automate existing factories, instead new challenges are created which are needed by the robotics community. Basically spoken, robotics development is working for it's own need. There is no plan to transfer the knowledge from the university into practical applications. Both domains are separated.
Today, both parties are working with opposite technology. In practical automation projects, the well known CNC machines are used which were developed in the 1970s. These machines provides the maximum productivity and help the companies to reduce the costs. On the other hand, the robot projects in the universities are working with different goals. CNC machines are not researched in the university domain, instead the prefered technology is deeplearning, biped robots and modern robot control systems. The prediction is, that in the future the gap will increase. That means, university driven robotics projects and automation projects in the factory have nothing in common. And it's done with different ideology in mind. Simply spoken, both parties have unlearned how to communicate with each other. University researchers who are interested in robotics have no reason to start a project in which a CNC machine is utilized, because this technology is 40 years old and it's not interesting enough from an academic standpoint. On the other hand, automation experts in a car factory have no obligation to introduce modern robotics in the workplace, because these devices are costing too much and doesn't provide a value.
This unwillingness to communicate is something which was not there in the mid 1970s. In that time, university research and the need of the industrial automation was the same. The latest CNC machine technology was developed first by researchers in the lab and then the technology was transfered into the practical domain. With the advent of fifth generation computer the situation changes drastically. Basically spoken academic research and industrial needs have developed into opposite direction.
To understand the reason why we have to describe the situation in the reality. What car companies and the service sector is trying to do is to earn money by providing products. At first, the company is producing a car, and then the car is sold to a customer. The money is reinvested into the factory and more cars are produced. Research projects driven by companies have the obligation to make the process more efficient. The aim is to reduce the costs of producing a car, and if en engineers has an idea how to do so, the factory will use it as soon as possible. The disadvantage of this principle is, that a company is profit oriented. They are only interested in technology which helps to reduce the costs. The problem is, that the entire domain of Artificial Intelligence and robotics won't help to reduce the costs, but it's doing the opposite. From the perspective of a car factory it make no sense to research Artificial Intelligence in detail. Because everytime a detail problem was solved, new problems become obvious. As a consequence, Artificial Intelligence isn't researched by profit oriented companies, but the research is delegated to universities or outsourced into research teams which are not profit oriented.
This kind of hypothesis can be proofed by investigating robotics company in the past. Some examples are available in which companies are trying to earn money by developing robot hardware and software. The most advanced example is the Willow Garage company which has programmed the ROS operating system. What all these companies have in common is, that they are struggle from an economic perspective. The reason is, that created hardware and software finds no customer. Basically spoken nobody likes to pay 100000 US$ for a household robot which can do nothing.
It's not the fault of a certain company, but it has to do that Artificial Intelligence in general has problems to find customers. The market principle is, that a customer pays money and then he gets something in return. A robot works a bit different. If a customer pays 100000 US$ for the PR2 Robot from willow garage he gets nothing in return. What he gets instead is the need to invest more time and more money into the robot.
In a previous blogpost, I have compared robotics projects with a flame which has no purpose. It's possible to throw more fuel into the flame but the flame wouldn't provide something back. The problem is, that market oriented products have to provide an added value for the customer. He pays for example 100 US$ and he expects something in return for the money. And exactly this is missing for AI projects. In the 1980s the naive assumption was, that robotics projects have a long duration. That means, that before the robot can improve the productivity in the company, the engineers will need 10 years in which they can explore the new technology. In the meantime it's known, that the duration is not 10 years, but 100 years and longer. That means, the AI community will research a topic for decades and at the end they won't have something to offer which helps the customer. The problem is not, that the research is hidden behind closed doors. The problem is, that even all the papers are published they are useless for automation tasks.
Perhaps some numbers make the situation more obvious. Each year around 1 million papers about artificial Intelligence were created newly in the Google Scholar directory. Most of them can be downloaded in fulltext. The papers itself are great, the authors are experts on the field and each year they are describing more complicated robots which were build in the laboratory. But a closer look into the paper will show, that nothing new was discovered. They academic community has researched a topic in detail, but they haven't found anything what can be converted into a practical product. The result is, that car factories, hospitals and restaurants are working unchanged since 40 years. The technological development has stopped. No technology is available and all the work is done by human workers.
What we can observe is, that the world is on a low technological level which was froozen in the 1970s and at the same time, the AI revolution has started and the development speed has increased over the years. Latest robotics research is more advanced than ever and nearly each week a breakthrough is available and at the same time the companies in the real world are working with outdated CNC machines, barcode scanners and repetitive human work. The hypothesis is, that the gap can't be overcome and it's described in the literature as the productivity paradox.