Classes of AI Agents

October 10, 2019

General AI

Classes of AI Agents

Agents in Artificial Intelligence

Info #1 – Link

By: Sahil_Bansall


Classes of Agents

This link goes over some of the basic setups for a few different classes of AI, which are as follows:

  • Simple Reflex Agents
  • Model-Based Reflex Agents
  • Goal-Based Agents
  • Utility-Based Agents
  • Learning Agent
Simple Reflex Agents

These agents simply react to the current percept, with no regard to percept history. It uses condition-action rules, which simply map a state/condition to an action. So these take in some input with sensors of some type, and then depending on solely that input, it decides an action to take at that time.

Model-based Reflex Agents

These find a rule whose condition matches the current situation. This agent adds an internal state, which is updated with each percept depending on the percept history. This internal state allows the agent to create its own model of the world that cannot be perceived. This agent updates its state with information on how the world evolves independently from the agent, and how the agent actions affect the world.

This type of agent starts to work well with a partially observable environment, since it can make its own “guess” at the full world by making its own models.

Goal-based Agents

These agents make decisions to reach a goal (desired situations). Every action they take gets them closer to the goal. This generally involves using searching and planning.

Utility-based Agents

These agents base actions of preference (utility). How “happy” an agent is is sometimes used to describe its utility. These use utility functions to map a state to a real number which can evaluate what action will make it most happy.

Learning Agent

This type of agent starts with a basic knowledge but then acts and adapts through learning. This learning is done through the use of 4 key conceptual components:

  • Learning Elements: makes improvements by learning from environment
  • Critic: Gives feedback to learning element to inform agent how well it is doing relative to a fixed performance standard
  • Performance Element: selects action
  • Problem Generator: suggests actions that will lead to new and informative experiences