Lecture 19 - Intelligent Agents Flashcards
What is an agent?
A system that is situated in an environment
Capable of autonomous action in the environment to achieve goals or design objectives
What does autonomous imply in relation to intelligent agents?
The agent can act without the intervention of humans or other agents
What must agents be able to do?
Perceive/Understand their environment
Act upon their environment
A central heating thermostat is an example of an agent. What is the environment, sensor and actions of the agent?
Environment: Building whose heating it regulates
Sensor: Its temperature sensor
Actions: Turn heating on, Turn heating off
A pressure cooker safety valve is an example of an agent. What is the environment, sensor and actions of the agent?
Environment: Pressure cooker and its surroundings
Sensor: Pressure sensor
Actions: Open valve, Close valve
A simple spinal withdrawal reflex is an example of an agent. What is the environment, sensor and actions of the agent?
Environment: Animal’s body and surroundings
Sensor: Pain sensors in the skin
Actions: Activate flexor muscles
When is an agent considered to be intelligent?
When it is capable of flexible autonomous behaviour
What does flexible mean in the context of autonomous behaviour and intelligent agents?
Pro active
Conditionally reactive
Socially interactive
What does pro-active mean in the context of flexible autonomous behaviour?
Can take the initiative in order to achieve goals
What does conditionally reactive mean in the context of flexible autonomous behaviour?
Can react in a timely fashion to changes in the environment by selecting the appropriate response from a range of alternatives
What does socially interactive mean in the context of flexible autonomous behaviour?
Can interact with other agents (possibly including humans) to achieve collective goals
Why are expert systems generally not considered to be intelligent agents?
They do not perceive their environment directly - needs a human user
Do not act on the environment - give advice to the user instead
What are two types of intelligent agents?
Reactive agents
Agents with internal state
How do reactive agents make decisions?
Direct mapping from current situation to action:
S->A
No memory of earlier environment states
Markov Decision Process is an example of which type of intelligent agent? why?
Reactive agent
Does not usually use memory to retain previous states
What separates an agent with internal state from a purely reactive agent?
Agents with internal state have some form of memory that future actions are based on.
Reactive agent just maps states to actions
Internal state:
agent_see: SxM -> M
agent_act: M -> A
What are the three main motivations of research into reactive agents?
Approaches based on manipulation of symbolic representations were not well accepted
Intelligent behaviour should be viewed as product of interaction between agent and environment
Intelligent behaviour emerges from interaction of a number of simpler behaviours
What is the Subsumption Architecture?
A reactive agent architecture
What are the two main ideas of subsumption architecture?
Task accomplishing behaviours - decision making realised through set of simple responses to current state
Subsumption Hierarchy - behaviours arranged in layers, lower layers inhibit higher layers - so as you go down the actions get more specialised
What are the advantages of reactive architectures in intelligent agents?
Cheap - low computational power required
Robust - loss of single agent does not seriously disrupt operation of entire set
What are the disadvantages of reactive architectures in intelligent agents?
All decisions based on purely local info so must gather good info to make good decisions
All behaviours are short term - no memory
Emergent behaviour difficult to program
What does BDI stand for, and what is it?
Belief-Desire-Intention
A type of agent with internal state
In multi-agent systems, what are the two types of interaction and what does each involve?
cooperation - must construct plans to achieve collective goals, either centrally or distributed
competition - must negotiate to resolve conflicts
What are the two types of communication in multi agent systems?
Direct communication - agents exchange mssages with each other
Indirect communication - agents communicate by acting on the environment
