AI Representation

Agents & Environment §

• AI as an Agent
  • ==Agent - Anything that perceives its environment through sensors and acts upon it through actuators==
  • Example: Human Agent
    • Sensors: Eyes, ears, etc.
    • Actuators: Hands, legs, vocal tract
  • Percepts - The content an agent's sensors are perceiving
    • ==Percept Sequence - The complete history of everything the agent has perceived==
  • Agent Function - A mapping of any given percept sequence to an action
  • ==Agent Program - A concrete implementation to realize an agent function internally==
• How do we define a rational agent though?
  • A rational agent should select an action that is expected to maximize its performance measure, given the evidence provided by the percept sequence and whatever built-in knowledge the agent has about the environment
  • But you must define a performance measure for this to work

Specifying the Task Environment §

• Task Environment - the “problems” to which rational agents are the “solutions”
• PEAS
  • Performance measures: How desirable is the sequence of environments the agent has experienced?
  • Environment: What are the key elements in the surroundings of the agent that can influence the agent?
  • Actuators: What allows the agent to take actions?
  • Sensors: What allows the agent to perceive its surroundings?

Agent Programs §

There are multiple ways we could implement an agent function

• Table driven approach
  • P = set of possible percepts
  • T = lifetime of an agent
  • ${\sum }_{t=1}^T|P{|}^t$ = Table size
  • Problem is the table size is WAY too big to be actually usable
• Simple Reflex Agents
  • 
  • Cannot model complicated behavior
• Model-based Reflex Agents
  • 
  • Doesn’t think about the goal
• Goal-based agents
  • 
  • But it can’t tell you exactly how happy the agent is…
• Utility-based agents
  • 
  • Designing a rational agent is to design an agent that maximizes its expected utility
• Spectrum of state representations
  • Atomic
    • Everything is independent from each other
    • Example: City names (West Lafayette, Gary, Chicago, Indianapolis)
  • Factored
    • Multiple measured things in state
    • GPS coordinates, current speed, current direction, current gas in the tank…
  • Structured
    • Objects and relations get built in
    • Objects: Car A, Car B, my car, traffic signs…
    • Relations: Car A is in front of Car B…
  •