What is the cleanest, easiest way to explain someone who is a non-STEM work colleague the concept of Reinforcement Learning? What are the main ideas behind Reinforcement Learning?
3 Answers
Humans are set loose in the world and go about their days doing stuff.
Whenever they do specific things, their brain sends them good signals (endorphins, joy, etc.) or bad signals (pain, sadness, etc.). They learn through these signals which things they should be doing and which things they shouldn't be doing.
Sometimes the signal is immediate and you know exactly what you're being "rewarded" or "punished" for (e.g. touch a hot stove and it hurts). Sometimes it takes a bit longer and there could be many possible reasons for the brain signal (even a combination of reasons), but you can hopefully figure out what caused it after it happens a few times (e.g. getting a stomach ache a few hours after eating a specific food).
That's basically what Reinforcement Learning is.
-
3$\begingroup$ Good "in plain language" answer based on everyday human experience! $\endgroup$– DukeZhouJun 16, 2020 at 19:38
The famous book Reinforcement learning: an introduction by Sutton and Barto provides an intuitive description of reinforcement learning (that everyone is possibly able to understand).
Reinforcement learning is learning what to do — how to map situations to actions — so as to maximize a numerical reward signal. The learner is not told which actions to take, but instead must discover which actions yield the most reward by trying them.
In the most interesting and challenging cases, actions may affect not only the immediate reward but also the next situation and, through that, all subsequent rewards. These two characteristics — trial-and-error search and delayed reward — are the two most important distinguishing features of reinforcement learning.
In chapter 3, the book also introduces the agent-environment interface, which summarises the cyclic interaction between the agent (aka policy) and the environment (which represents the task/problem that you need to solve).
Every RL algorithm implements a cyclic interaction between an agent and an environment (as illustrated above), where, on each time step $t$, the agent takes an action $A_t$, the environment emits a reward $R_{t+1}$, and the agent and the environment move from state $S_t$ to the state $S_{t+1}$. This interaction continues until some termination criterion is met (for example, the agent dies). While this interaction occurs, the agent is supposed to reinforce the actions that lead to better outcomes (i.e. higher reward).
Reinforcement Learning can be explained by a few equations. However I assume that this is not what you are looking at since the explanation should be for someone having a non-STEM background. Not to say non-STEM folks are not able to understand math equations, but intuition comes easier with words and examples in my opinion.
Reinforcement Learning is about learning an optimal behavior by repeatedly executing actions, observing the feedback from the environment and adapting future actions based on that feedback.
Let's break down the last sentence by the concrete example of learning how to play chess:
Imagine you sit in front of a chess board, not knowing how to play. The optimal behavior you'd like to learn is what moves to execute in order to win the game. So you start learning the game by playing a few moves (actions) with some figures and observing what is happening on the board (environment) and identifying which moves bring you closer victory or give you a better position on the board (feedback). Therefore in future games you will prefer moves which gave you a positive outcome in previous games.
Admittedly this is a very slow process of learning if you don't have a teacher which helps you in the beginning and you would have to play a lot of games until your first victory. But this essentially how computers (and sometimes humans in some sense) learn to do certain things by Reinforcement Learning. Behaviors which lead to positive experiences are collected, memorized and thus reinforced.