Imagine a robotic machine that learns to give intentional instructions to another robotic machine. It does this without any human helping or programming specific behaviors. Let’s call this robot One.
What does this mean in detail?
Our robot One uses unsupervised learning to generate a task for another robot that helps him achieve his own goals. Let’s assume One is a household management robot and responsible for keeping the house beautiful, clean, safe and stocked with everything his owners need. One has been in his new household for just 4 weeks and the garden is starting to look uncultivated. He realizes that the grass grew really high.
There are also other robots in the same household that follow strict instructions with no questions asked. For example, there is a lawn mower robot. One is able to directly communicate with this lawn mower robot, but he doesn’t know yet what a lawn mower robot does and what it is good for. A few days ago he had a casual conversation with the lawn mower robot about what he can do. The lawn mower robot said it is in the garden and can mow the lawn.
One doesn’t know what mowing the lawn means, but just the fact that the lawn mower robot is in the garden makes him curious whether the lawn mower robot could help him improve the look of the garden. So when the grass gets way too high, he just asks the lawn mower robot to do whatever it is designed for and observes what happens. And indeed, the grass gets shorter when the lawn mower robot mows the lawn.
What is so special about this?
When we think of robots we typically think of robotic machines that have been programmed by humans. These robots follow a set of specific instructions to execute their tasks. In the above example One is able to generate his own instructions to fulfill his tasks. On the one hand this sounds great as no human is required to tell One exactly how to keep the garden beautiful. He is able to learn everything that is available to him to achieve his goals himself. On the other hand this is a little scary, because One could do things his creators never thought of. Does he have or develop a free will? Does he eventually question why somebody gave him his goals and whether he should still try to achieve them?
One learned to ask the lawn mower robot to mow the lawn if the grass is too high. This allows him to keep the garden beautiful. If One learns this new feature or behavior, he can share his success with all other household robots in the world. This unsupervised crowd sourced learning can lead to exceptional learning speed. If robots can learn new things by themselves and share them among themselves, they can potentially learn to do more things and to do things better than humans.
One knew how to communicate with the lawn mower robot, because he was designed by a human to know it. Another possibility is that robots will learn to communicate with each other without being programmed or following instructions of a human. In this case there is no limit to what they can achieve together. The more such robots will be out there, the more and quicker will they learn.
And finally, once robots get so sophisticated that they understand their own design as well as the tools needed for it, they will eventually be able to reproduce and evolve without human intervention. Combine this with the before mentioned skills of unsupervised crowd sourced learning and learning how to communicate with each other. Now robotic evolution can happen at an unimaginable speed. A real robotic take off.
There are chances and risk involved with the robotic take off. The future will show whether we are able to guide this into a safe and useful direction. Therefore, I propose to at least give it a catchy name so more people think about it and its consequences. I propose robot-to-robot interaction – in short R2R.