When robots start learning from each other

  Many years ago, the war left a desolate desert with many mines and bombs buried in it. To eliminate the bombs and restore peace to the desert, two robots entered the area. Robot No. 1 found a suspicious sphere in the scan and carefully handled it with its tentacles to determine whether it was a stone or a bomb, and if it was a bomb, it would be removed. At this time, the robot No. 2 came along, and it scanned and communicated with No. 1. It also focused on this suspicious object, tried to analyze it, and tried to do it like No. 1.
  Social learning
  this is not real, but an imaginary picture, to demonstrate the ability to learn from each other between the robot. Robot No. 1 is a mature deminer, and No. 2 is learning from No. 1 in a practical way.
  Learning from peers is also called “social learning” by anthropologists and zoologists, that is, by observing or interacting with others to obtain new knowledge and new capabilities, it is widespread in the human and animal world. The role of social learning is crucial, without which it can be said that there is no human cultural development. The same is true for animals.
  Humpback whales feed on small fish and shrimp, and their predation activities are generally under the sea. However, according to scientists’ observations, humpback whales have gradually “mastered” new predation methods in the past 27 years. They have floated to the surface and beat the water with tail fins to chase fish and shrimp for predation, which can enable them to capture more types of food. Scientists speculate that a humpback whale might have found this predator first, and then their peers gradually learned and spread throughout the population.
  Chimpanzees are also known for their ability to learn. Chimpanzees usually break down the stems of herbaceous plants and stick them into ant nests as small sticks to “fish” termites, but the chimpanzee in the African Republic of the Congo has improved this method. They rub the ends of plant stems with their teeth to make them more hairy. Like a brush, this “design” helps chimpanzees catch more termites.
  There are many animals similar to this kind of learning ability. They improve themselves through acquired and mutual learning and improve the adaptive ability of the group. Now, scientists envision robots can do the same, improve themselves and improve their abilities through acquired learning.
  Robot teaches robot
  This assumption was turned into reality by researchers at the Massachusetts Institute of Technology. In 2017, they announced a major breakthrough in social learning for robots, enabling them to learn from each other and take shape.
  Researchers have developed a learning system named “Breakthrough Learning”, which means that after learning, it will break through the original restrictions. Its biggest function is to simplify the robot learning process and make it easier for human teachers to teach. Through breakthrough learning software, a human teacher drags a robot arm through a computer in a virtual scene and teaches it to operate. This process will be transformed into the ability input to the physical robot, so that it really has the corresponding ability. Different from teaching robots in the past, human teachers only need to demonstrate once in a breakthrough learning system.
  Researchers first use a breakthrough learning system to input basic knowledge into two completely different robots so that they know how to deal with objects, such as different treatments for round and square objects. Robot No. 1 has two arms but moves on wheels, and No. 2 is a bipedal humanoid robot. It moves on both legs and has extremely high balance capabilities. First teach No. 1 to detonate the bomb, which is a complex task consisting of a series of steps. Then he brought No. 2 robot in front of it. Under the demonstration of No. 1 and No. 1, combined with his own basic knowledge, he finally integrated the method of demolition of the bomb, which basically can complete the task that only humans taught to No. 1. Therefore, No. 2 was learned.
  Robot culture
  Since then, a new era had begun. Humans teach robots, and robots can continue to teach other robots. That is to say, when a robot masters a capability, it can transfer this skill to other robots, including different shapes, uses, types, and universal humanoid robots.
  So, when robots are more and more, they can continue to learn new knowledge capabilities the day after tomorrow, does it mean that robots can create their own cultural system? Is what human beings doing today the beginning of creating a future robot community?
  Regardless, it is important that social learning for robots needs to be initiated from humans. More importantly, it also requires human intervention, to prevent criminals from abusing technology to endanger society, and to prevent anti-human and anti-moral behaviors from being learned by robots.
  What should cause more contemplation is that if robots can learn from each other and continuously improve themselves, shouldn’t you and me in humans learn by themselves, learn from others and the environment, and make continuous progress?