Let’s start with the ML Agents key point, or the Machine Learning key points.
Any Machine Learning engine learns by observing its environment, how react or how is changing, in response to its actions. Trying to guess which actions are the most optimus to obtain rewards, or which he needs to avoid in order not to get penalties.
But each action, even if it is the same, is different, since it is affected by previous changes in the environment.
Our role is try to create an environment well balanced combining the three key points. This balance is the difference between a well trained agent and an Agent unable to learn.