Behavior-Based Artificial Intelligence

1
Behavior-Based Artificial Intelligence

• Pattie Maes
• MIT Media-Laboratory
• Presentation by
• Derak Berreyesa
• UNR, CS Department

2
Abstract

• Uses behavior-based artificial intelligence as a
  new approach to studying intelligence.
• Deals with autonomous systems that must deal with
  changing goals in an unpredictable environment.

3
Two approaches to AI

• Knowledge-Based AI
• Knowledge based systems have prior knowledge
  about the environment and there task.
• Behavior-Based AI
• React to there changing environment and goals
• They behave in a problem domain.

4
Knowledge-Based AI

• Isolated and advanced competences.
• Closed systems that only connect to the user,
  not the environment.
• Deal with one problem at a time.
• Uses static knowledge structures to solve
  problems.
• Dont have to adapt to changing situations.
• When creating an autonomous system a computer
  takes over as the user.

5
Behavior-Based AI

• Has several competences but they are not as
  intelligent as isolated competence would be.
• Open system that has little time to react.
• Decides itself what the problem to be solved next
  is.
• Relies more on producing behavior than knowledge.
• Emphasis on adaptation.

6
Behavior-Based problem solving

• A system that can learn has to rely less on
  planning.
• Environment can be exploited.
• Learn over time.
• Interacts in society to help accomplish tasks.

7
A Mobile Robot

• Knowledge-Based would create a model of the
  office to update as often as possible. It would
  include the location of the robot itself and the
  location of certain objects.
• Behavior-Based would have specific modules for
  wall following, going through doors, etc.

8
An Interface Agent

• Knowledge-Based will learn all it can about what
  it should prompt the user for and/or try to do
  when the user does something.
• Behavior-Based will have a each separate module
  try to learn about its own specialized task.

9
Conclusion

• Behavior-Based AI have proven to be very
  efficient so far.
• More fundamental research needs to be done.
• A better understanding of the underlying
  principles.