Classifying AI agents
In this section, we will discuss how best to classify agents and go into more detail about how such a complex system learns. The first classification is between agents that move only in a virtual environment and embodied agents.
Digital agents are confined to a virtual environment. Again, we have varying degrees of interaction with the virtual universe. The simplest agents have interaction with a single user. For example, an agent can be programmed in a virtual environment as a Jupyter notebook, and although it can search the internet, it has rather small, and therefore primarily passive, interactions. There are two subsequent levels of extension:
- Action agents perform actions in a simulated or virtual world. Gaming agents interact with other agents or users. These agents usually have a goal (such as winning a game) and must interact with other players to succeed in achieving their goal. A reinforcement learning algorithm is usually used to train the...
