COSC 3402 Artificial Intelligence

Introduction to Artificial Intelligence

Machine or computational AI:

1950 - Alan Turing -> Turing Test
Doesn't define intelligence but if a test is passed the winner is intelligent.
Later people lost interest in Turing test.

Intelligence is a degree of success in solving a particular task by an AI agent. i.e. define intelligence relative to a task.

The Objective To develop computer hardware/software to

1. Investigate a variety of aspects of intelligence.
2. Implement AI agents (systems) that exhibit intelligent behaviour.

Problem Solving is done in AI by munipulation of knowledge.

There can be no intelligence without a knowledge base.

AI Modules:

• Executive
  1. KB (Knowledge Base)
  2. Planner
  3. Scheduler
  4. Actors
• Sensors
• Interface with other agents

1. KR (Knowledge Representation)
2. Search
3. Automated Responce
4. Planning (Acting & Behaviour)
5. Networks
6. Generic Algorithms
7. Machine Learning

Knowledge Representation (KR)

There is no AI without K.R.
Suppose we want to design a Chess program:

• Select move
  This can be random but then it wouldn't be AI.
• Use a list of good possible next moves.
  We have limited space to keep all possible good moves.
• Use a function h(current, next) that tells you how good the next configuration is.
  We need lots of human knowledge to design function h.
• Use the huerristic function on the next few levels.
• Most programs use both the list of good moves specially for opening and ending the game and use function h for the next few moves in advance.

Different levels in AI:

• Fact Level "True" relationships in the environment
• Interface Level How to communicate facts to an AI agents?
• Knowledge Representation Level How to represent knowledge in an AI system?
  • What language to use?
    • Shouldn't take much space in KB
    • Possible choices are
      • Natural Language: expressive
      • Formal Languages: unambiguous but not as expressive
        • Formal language of logic
    • Expressive power of languages:
      • inexact knowledge:
        e.g. temperature of A is too high
      • probabilistic knowledge:
        e.g. Device A works with a probability of 0.7
      • temporal (timed) knowledge:
        e.g. if on(boxA, table1, 1 pm), is it still there at 2pm?
      • model knowledge:
        e.g. It is possible that I'll come back again
        Note: model operations: possibility, necessity are not expressible in classical predicate logic.
      • epistemic knowledge:
        --> One most add epistemic knowledge to predicate knowledge.
  • What operation on KB?
    • retrieve
    • logical reasoning
    • planning with KB
    • etc.

Behaviour of an AI Agent

Defined as the way the agent interacts with the environment. (This includes the interaction with other AI agents and/or human.)

Types of Behaviour

• Static Behaviour minimal or no interaction with the environment
  e.g. A bird that only goes right/left on the screen until the end of screen is reached, then changes direction and y-coor.
• Reactive Behaviour environment determines the choice of actions and activities
  e.g. In a simple adventure game the guard shoots if the player is in sight.
• Goal-oriented Behaviour environment as well as the general goal (objective) of the the agent's behaviour determines the choice of actions and activities
  e.g. In a simple adventure game the player must pass the guard and save the princess.

Static and reactive behaviour can be created using a finite automata.

Example: 
We have a 2D world with some obsticles at some cells. 
Function clear(x,y) is true if there is no obsticle at cell (x,y).
Such that The precondition for MoveRight is that clear(x+1,y) is true 
and so on for other three directions. This can create a F.A.
We can give preferences to moves (try right, if not clear, try down, etc.)

Rule-based systems

a rule: If a, b, c,...,x then do A
a^b^c^...^x ==> do(A)
The system will try the rules in order.

Using logic to implement goal-oriented behaviour >>>