This unit explores several learning methods that are used in the field of robotics. Some time is spent exploring the concept of reinforcement learning which I discussed in the previous unit, however the unit dives deeper and also covers supervisor learning as an exciting area with big potential for robotics.
Table of Contents
Reading – Chapter 21: Things Keep Getting Better – Learning
Learning is a challenging thing for anything to do, particularly because there is so much uncertainty that goes with the process of learning and concepts, actions, and information need to continue to be tested and confirmed. Depending on the amount and type of information available different modes of learning may be more appropriate for different problems. Learning is an exciting field though, and by combining different learning methods it should be possible to achieve even better results for the learner.
Food For Thought 21 -1
Some people find the idea of robot’s learning frightening, because they are uncomfortable with the notion of robot behavior being anything but completely predictable. Does adding learning to the robot’s abilities make it more unpredictable? If so, how can this be helped? If not, how can people’s worries be put to rest?
I think that when people are worried about robots learning they are worried more about other tasks that we tend to associate with learning. I think it would be very helpful to take the time to understand what learning is and what learning is not. Learning is the accumulation of knowledge or experience, and it is the application of the knowledge to improve understanding. This then leads to the question, what is understanding? I would argue that understanding is the use of knowledge to predict the outcome of a situation or action. So by this logic learning is then the ability to refine and verify predicted outcomes of actions, or situations. When the predictions or expectations are shown to be erroneous, the underlying assumptions are then revisited and new expectations are drawn for future encounters with similiar actions, situations, or challenges.
Now the important thing I want to point out here is that learning is not intrinsically a creative ability. As humans, we operate with emotion and intuition. We make leaps, we go beyond what we know based upon what we feel. Someone that can learn is not necessarily someone that can make an intuitive leap. It is our intuition and our emotion that allows us to make leaps and which gives rise to our creativity.
I believe that a robot that can learn will be a valuable tool and will be immensely helpful in processing learning and knowledge, but without emotion and intuition will be unable to devise new concepts and thoughts. The robot will be able to iterate, but will be stunted when it comes to innovation.
I want to point out that even in the animal kingdom, we can train many animals to do many different things. These animals learn that task. They build associations between stimulus and the best course of action for receiving a reward. The animals, while sometimes emotional, do not however show intuitive leaps, that allow them to learn are create a new skill without having been taught. They are reliable in their behavior, and while there is variation and personality involved, they will not spontaneously do something that is outside their nature or to which they have not been trained.
So while a robot is going to be a complex system, and we cannot know its entire future of actions and states, we can say that there are limitations upon that system and within those limitations we can have reasonable expectations for the behavior and outcome of a robotics system. I believe that a learning robot will be as predictable as the programming has made it to be. I do not expect a learning robot to intrinsically be an emotional or intuitive system, as such I do not expect the robot to be an unpredictable system.
While this lesson was focused on robotics, I feel as though it has touched upon approaches in artificial intelligence and upon our understanding of psychology. This lesson, as with many of the lessons in this course feels as though it is a simple introduction that shows the reader some of the various other disciplines that are integral to robotics and invites them to continue to learn as their interests are piqued.
December 11th, 2021
Feature image: Photo by Pavel Danilyuk: https://www.pexels.com/photo/a-robot-playing-chess-8438880/