Will artificial intelligence (AI) someday make robots smarter than humans? Fueled by spectacular science-fiction films, gloomy novels and sensational press reports, the public is often afraid of intelligent AI robots that replace, disenfranchise or even subjugate humans. This is reason enough for the automatica to let real robotics and AI experts from industry and research speak.
The opinion of AI and robot experts is clear: There will be no artificial robotic intelligence superior to human intelligence in all areas in the foreseeable future. Special solutions for artificial intelligence are already better than humans in their sub-areas, but, “general artificial intelligence” with its own personality, which mimics the human mind and human intelligence in all its facets, is not in sight if not technically impossible. Therefore, robots even with AI will probably never be able to replace people completely with their enormous adaptability, flexibility and solution competence.
Especially since robots have a very special challenge: they must always interact physically with their environment. And with respect to this practical intelligence, which requires physical interaction of the robot with its environment, even young children are clearly superior to intelligent robots. An example: Today, AI software can beat the best human player in chess, but until now no robot can take a board with chess pieces off a shelf and put the pieces in place to start playing.
However, robots can be significantly better than humans in their respective, very specific task area thanks to AI, especially where rapid evaluation of data for a specific application or reliable task execution (keyword “Smart Factory”) is required. Intelligent-seeing and sensing robots can and should relieve people of heavy, tiring and repetitive tasks—or at least help them in doing so.
Will robots be smarter than humans thanks to AI one day?
A robot that is truly superior to human intelligence in all areas would require general artificial intelligence. This is not possible today and will not be possible in the foreseeable future. The term "artificial intelligence" often gives the wrong impression of fully autonomous systems. This is not in line with our demands on robots. Users want the predictability, high precision and reliability of robots. Not the least from safety aspects, the parameters for robot autonomy in commercial robot applications are therefore set very tightly.
Where are (in the short term) the most sensible or promising applications for AI/machine learning in robotics?
Commercial applications of artificial intelligence in robotics already include detection of their surroundings and reaction to it, for example; this considerably expands the range of functions that robots can perform. Optimization of robot and process performance also results in direct cost savings for companies. Current AI research areas in the field of robotics include the expansion of gripping technology (for example to pick non-rigid objects) and the expansion of robot mobility in non-standardized environments (e.g., rough terrain). Additional areas of focus are gesture and voice control of robots and simplifying robot programming.
Will robots be smarter than humans thanks to AI one day?
Thanks to artificial intelligence, robots will be significantly better than humans in their respective, very specific areas of responsibility, but they will not be able to replace humans completely with their enormous adaptability in the foreseeable future.
Where are (in the short term) the most sensible or promising applications for AI/machine learning in robotics?
Machine learning helps wherever large amounts of data need to be sorted, structured and interpreted. For example, when recognizing objects by means of computer vision, when determining the most effective route through a warehouse or where several robots in a fleet learn from the experiences of the other robots to become better. For other areas of artificial intelligence, such as automated planning or inference, I also see many applications in robot control, making decisions in dynamic environments and generally making the right decision about what to do if something goes wrong.
Will robots be smarter than humans thanks to AI one day?
Robots are already superior to humans in very specific tasks. This applies where criteria such as reliable task execution or the rapid evaluation of data by means of artificial intelligence are required for a specific application. But the challenge for robots is that they always have to interact with their environment. The more complex this is, the more challenging this interaction is. Therefore, the following often applies: What is easy for the robot is difficult for humans and vice versa. Future AI applications will extend the applications where robots can outperform humans. However, systems will not actually become more intelligent than humans with their solution competence and flexibility in the foreseeable future.
Where are (in the short term) the most sensible or promising applications for AI/machine learning in robotics?
The main discipline for machine learning is currently machine vision. This enables robots to reliably detect and grasp a variety of different objects without manual programming effort. Machine learning considerably reduces the set-up effort for new objects and is the door opener for completely new applications in handling, for example in the case of bin-picking in a warehouse. Another promising area is the more efficient collaboration between a person and a robot, in which the latter recognizes the intentions or emotions of the user, for example, and can plan its actions accordingly.
Will robots be smarter than humans thanks to AI one day?
There is no doubt that intelligent robots can and should replace or at least help people in heavy, tiring and repetitive tasks. Intelligent, seeing and feeling robots can be of great help in times of shortage of skilled workers, at least for simple tasks. However, we are still far from the robots in science fiction movies. An example: While today there are chess computers that can beat any chess grandmaster, there is no robot that can take a chess game off a shelf, opens the box with the pieces, take the pieces out of the box, set them on the chessboard and then start playing. With respect to this practical intelligence, which requires physical interaction of the robot with its environment, even young children are clearly superior to today’s intelligent robots. And it will remain like that for a long time to come.
Where are (in the short term) the most sensible or promising applications for AI/machine learning in robotics?
The most promising applications for methods of artificial intelligence are in the field of machine vision, which can be used for greater autonomy and flexibility of robots, for example for object recognition, semantic scene analysis and gripping point recognition. The optimum parameterization of robot movements, for example during welding or assembly operations, is also an application possibility that has already become a reality today. Developments are also focusing on processes for predictive maintenance. Intelligent analysis of large amounts of data from processes can also be used for quality assurance as well as process monitoring and control. For example, to detect if a component has been correctly mounted. In addition, a robot can be programmed via demonstration.
Will robots be smarter than humans thanks to AI one day?
I think the term artificial intelligence is excessive here. We prefer to talk about machine learning, because it is about qualifying our robots in such a way that they can automatically adapt to changing production processes in digitally networked environments. We're working on very specific tasks here, and we don't expect robots to be smarter than humans one day.
Where are (in the short term) the most sensible or promising applications for AI/machine learning in robotics?
A very concrete solution is our mobile robot Helmo. It is already so intelligent that it can move autonomously in manufacturing environments, recognize obstacles and drive around them. In the field of conventional robotics, we are working hard on new, intuitive programming tools for our robots as well as on self-optimizing systems. We can record more than 2.000 machine parameters on our robots and consequently have a reliable database, which we can use to optimize individual applications. Using ML and corresponding algorithms, our robots will automatically be able to optimize themselves without human intervention and adapt perfectly to flexible production processes in Industry 4.0 environments.
Artificial intelligence will also be one of the three trend topics at automatica 2020 (December 8-11). Experience the latest in automated production and gain valuable insights into a future full of entrepreneurial opportunities!