Practice makes perfect.
That’s actually the essence of Reinforcement Learning.
The more the AI system experiences a certain environment, the more it learns how to succeed in it. Beyond this, to truly understand what Reinforcement Learning is, it is necessary to first get down to the foundational level of things.
Current articles and posts on RL are no different from anything else in the AI space. Most pieces on the subject appear to be full of jargon that can be hard to follow for those that are not industry insiders. Despite this, understanding Reinforcement Learning is simpler than you might think.
Firstly, it is a sub-field of Machine Learning.
Secondly, once you already know the general sense of it that we mentioned above, there is really only one other central process to explain in order to define RL.
In RL, an AI system is motivated by achieving some sort of reward that is given to it by the environment that it works in. One contemporary example of a system using RL in this way, today, is Google’s DeepMind.
Now, you might have already read about the AI that beat the human Go master.
That was none other than the DeepMind. To do this, it used RL to learn the game of Go to the point where it could understand the paths to victory better than a human. If you are wondering how this could have happened in the first place, let us begin by thinking about every interaction involving an AI in RL, as a game.
The more that it achieves certain qualitatively measurable results from certain actions that it takes, the more that it continues to play the game using those specific actions. These results are equal to the rewards that we have spoken about.
On the other side of things, if it takes some sort of action like a movement in Go that does not work, then it eventually abandons that action in favor of another that achieves its desired weight. By desired weight, here, we are going back to the idea of this numerical reward. Through grasping this simple formula of trial and error, we then can say we have a basic understanding of how the DeepMind learned to play Go better than a human Go master. Furthermore, we now know have the same level of understanding of how Reinforcement Learning works. Expect future pieces to go deeper into the sub-fields of RL to strengthen this even further.
Resources:
Primary Source:
https://medium.freecodecamp.org/an-introduction-to-reinforcement-learning-4339519de419
