How to Teach Robots | You Should Listen To ThisMay 14th, 2018
The future of robotics and AI is being created today by companies and innovators that are developing new technologies and implementing them across various industries. As robots become more and more complex questions have emerged as to how significant an impact these machines will have. Will robots replace human capabilities? Will new jobs be created to supervise, animate, and maintain these machines? How will robots learn and adapt to numerous roles and responsibilities? These are the questions that the hosts of Bloomberg’s Decrypted podcasts asked in their episode last week.
The primary issue that this podcasts dives into is the notion of robotic learning and how we can teach machines to adeptly accomplish tasks. This podcast suggests that “before they can take away our jobs, robots need to learn how to do them.” The host, Joshua Brustein outlines two methods of robotic learning that have guided the way we teach robots today: Imitation learning and Reinforcement learning. The first of these two methods builds upon a robot’s cognition of a given task by having a human repeatedly conduct the task themselves and allowing the robot to imitate the correct function. This is the process through which many of the robot products that exist today employ in furthering the capabilities of their machines. The host visits Kindred Robots to explore how ‘pilots’ are teaching sorting robots using this technique. At first the pilot carries out most of the task themselves, but after time the robot takes over and only asks for help when it gets stuck.
The second technique, reinforcement learning, employs a more complex process of trial and error through the use of data to improve the rate at which robots learn. The expectation is that using both methods programmers can teach a robot to “learn to learn,” and thereby reduce the time it takes to understand new tasks. This method layers information and draws from previous instructions to increase the abilities of the robot in question. The result has been that even robot manufactures have been surprised as to the rate at which robots are learning new skills.
Beyond the insight into the theory, science, and practice of teaching robots explained in this podcast, you should listen to this to better understand the growing industry of robotic technologies. You should listen to this to ask deeper questions as to the impact that robots will have on jobs here in the US and abroad, and you should listen to this to hear from leaders in this field as to the possible capabilities of machines in acquiring human like intelligence. This podcast looks into the cases of individuals actively automating away their own usefulness and explores their motivations for doing so.