Dayton AI Class 1

In which we try to explain why we consider artificial intelligence to be a subject most worthy of study, and in which we try to decide what exactly it is, this being a good thing to decide before embarking.

(This post is just my notes on the course videos for week one.)

The purpose of the course is to teach the "very basics of artificial intelligence", and also "to excite you about the field." The course structure is videos to introduce new material, followed by quizzes to test ability to answer questions about AI, and then wrapped up by answer videos. There are also assignments and tests, which don't have answers posted like the quizzes.

Course outline for the first week:
  1. Introduction
  2. Course Overview
  3. Intelligent Agents
  4. Applications of AI
  5. Terminology
  6. Poker Question
  7. Robot Car Question
  8. AI and Uncertainty
  9. Machine Translation
  10. Chinese Translation 1
  11. Chinese Translation 2
  12. Chinese Translation 3
  13. Summary

Intelligent Agents

The most important unifying concept in this intro to AI course is the idea of intelligent agents. Intelligent Agents interact with their Environment. They perceive the state of the Environment through Sensors, and can affect the Environment through Actuators. The big question of AI: what is the function that maps sensors to actuators (the control policy). How does the agent make decisions? The continuous loop from sensors to decision to actuation is the perception-action cycle.

Applications of AI

  • Finance: trading agents
  • Robotics: physical sensors and actuators, machine control agents
  • Games: opponent agents, character agents
  • Medicine: diagnostic agent (expert systems)
  • Web: crawler agents (spiders)

Basic Terminology

Four two-level factors describe basic AI problems.

Observability:
  • Fully Observable: can the agent observe the complete state needed to make the optimal choice?
  • Partially Observable: does the agent need memory in order to make the optimal choice?
Determinism:
  • Deterministic: agents actions uniquely determine the outcome (e.g. chess)
  • Stochastic: outcomes have a certain amount of randomness (e.g. dice games)
State-Space:
  • Discrete: finite states and actions
  • Continuous: the space of possible actions is infinite
Opponent:
  • Benign: the environment has no purpose contrary to the agent's
  • Adversarial: the environment (opponent) is attempting to defeat the agent, many games

Uncertainty Management

AI is the discipline to apply to figure out what to do when you don't know what to do. Reasons for uncertainty:

  • Sensor limits
  • Adversaries
  • Stochastic Environments
  • Laziness
  • Ignorance
Rationality: an agent is rational if it does the right thing. Obviously, we can only be sure of doing the right thing in very simple cases.

Further Reading

Alan Turing asks, Can Machines Think?

No comments:

Post a Comment

Archive