Environment types in artificial intelligence and examples
There are several kinds of perspectives to determine types.
1. Fully Observable vs. Partially Observable
In a fully observable environment, the agent knows the full state of the environment. In a partially observable environment, there are some missed sensor data, for example, because of the noise of the sensor. There is also a third type of environment from this perspective called unobservable. For the third type, there is no sensor data. Even if the situation is hopeless, the goal might still be achievable.
- Chess: The agent can see the chessboard and the opponent’s movements. (fully observable)
- Tic-Tac-Toe game: We can see all position for the next move, it was both our movements and the opponent movements. (fully observable)
- Driving: While the driver agent can see the partial environment, it cannot see what is behind that corner. (partially observable)
- Airplane: An airplane has longitude and latitude but, it reads this data from the GPS sensor. And a GPS sensor provides data with a certain amount of noise. If the noise powerful enough to change the airplane state, in this case, we can say observation is partial. (partially observable)
2. Single agent vs. Multiagent
It is easy to understand, an agent trying to solve a crossword puzzle by itself is defined as a single agent environment. But in the chess example, if both sides are controlled by the algorithm, we can say this is the multiagent environment. For example, while one side is set for high performance, the other side is set for low performance. This called a cooperative multiagent environment.
3. Deterministic vs. Stochastic
If we can determine the outcome for a certain state, this is a deterministic environment. Most real-world environments are not deterministic, they are stochastic. On the other hand, maybe the beauty of the world is hidden in its stochastic circumstances. That means the same input doesn't give the same output. Taxi-driving is quite a stochastic environment.
4. Episodic vs. Sequential
Many classification tasks are episodic. The problem divided into several episodes and each episode are independent of each other. But in a sequential environment, there is a need for memory to be able to pass the information from one sequence to another.
5. Dynamic vs. Static
If the environment is changing during the agent’s thinking, this is dynamic; otherwise, it is static. Static environments are quite easy to solve a problem, there is no need to worry about time. For example, taxi driving is very dynamic.
6. Continuous vs. Discrete
While the taxi-driving environment is a continuous environment, chess is a discrete environment (except time during the game).
7. Known vs. Unknown
loading…