Learning in Artificial Intelligence must be approached as a fundamental tool for the maximizing of any agent’s “growth” or development. Artificial Intelligence is based on the concept of perfecting an agent’s response in any environment which implies a deep focus on the agent’s progress through experience. Learning by itself should entail a continuous process as environments change and most likely also include new challenges for the agents that may not have been considered initially by the developer or even acknowledge in the initial data. This is why I believe a reinforcement learning approach is fundamental for any agent that operates in a real-time system. Reinforcement learning is a sub-field of artificial intelligence based on the agent’s need of feedback to achieve a maximum or optimal response throughout its learning. The way we achieve this learning is through continuous analysis of the environment; after every action by the agent the environment has a specific state that can be perceived by the agent, which then based on the newly acquired and store data must develop again a response. If the response maximizes the environment state the agent must be rewarded to be able to reinforce this type of behaviors. This process goes on and after every step the data the agent uses to evaluate each action is more and more accurate leading to better responses every time. Reinforcement learning includes many other types of learning too as it acknowledges the need of an agent to learn in a new environment, to behave accordingly to the specific rules of such environment with the goal of achieving an optimal response and to continue to successfully improve its response based on experience, which is fundamental for a maximized learning and therefore a maximized performance. The decision of which reinforcement learning to use, for example passive or active, negative or positive, must be based on specific conditions and allow the developer for a lot of different possibilities and outcomes which at the end leads to even better performance on a case by case basis.

Source: Artificial Intelligence on Medium