الذكاء الاصطناعي 1: مثال فلتر الرسائل المزعجة

Study Notes

PEAS Example 2: Spam Filter
- Performance Measure: Minimize false positives and false negatives.
- Environment: User's mail account, server's mail account.
- Actuators: Mark as spam, delete.
- Sensors: Incoming messages, user's account profile.

Goal-based agents: Actions depend on the goal. For example, a robot moving from one room to another will have different actions than a robot moving to a different room.
Cost-based agents: Goal is to minimize the cost of erroneous decisions in the long term. An example is a spam filter that tries to categorize incoming emails correctly (wanted or unwanted).
Comparison Example: Agent 1 makes 12 errors out of 1,000 emails while Agent 2 makes 38 errors. While Agent 1 appears better, the errors made by Agent 2 may be considered more serious if they involve important emails. Therefore error severity might need weighting.

Discusses different characteristics to consider when building an AI-based task environment.

Fully Observable vs. Partially Observable: Does the agent have access to the complete state of the environment? In a game of chess, a player knows all the pieces on the board while in a complex environment like a robot navigating a room, only partial observation is possible, making it harder for the agents to navigate.
Deterministic vs. Stochastic (vs. Strategic): Is the next environment state fully determined by the current state and the agent's action? In a deterministic environment, if an agent makes the same action twice in the same state, the environment will end in the same outcome. Stochastic environments differ because the outcome can change.
Episodic vs. Sequential: Is the agent's experience divided into separate, unconnected decisions, or is it a continuous sequence of actions and observations according to a transition model? In an episodic environment, an agent's decision doesn't need to consider future outcomes. Game of chess, for example, is sequential.
Static vs. Dynamic (vs. Semi-Dynamic): Does the environment change while the agent is thinking? A semi-dynamic environment doesn't change with time, but the agent's performance score does.
Single-Agent vs. Multi-Agent: Is the agent operating alone or with others/other agents in the environment?
Known vs. Unknown: Are the rules of the environment, transition model and reward associated with the states known to the agent?

The document provides a table categorizing task environments (e.g., Chess, Dominos, Medical Diagnosis, Self-Driving Car) based on various properties (fully or partially observable, adversarial, deterministic or stochastic, discrete or continuous, single or multi-agent).

Autonomous Taxi Driver: Is the environment of an autonomous taxi driver a competitive or cooperative multiagent environment?
Task Environment Characterization: Characterize the following environments (Poker Game, Medical Diagnosis, Image Analysis, Interactive English Tutor). Describe the environment, whether known/unknown, fully/partially observable, etcetera in your answer.
Known/Unknown vs. Fully/Partially Observable: Is the distinction between known/unknown environments the same as fully/partially observable?
Episodic/Sequential Environments: Explain episodic environments and their simplicity in comparison to sequential environments.
Communication in Multiagent Environments: Describing environments where communication frequently arises and why it's a rational behaviour.