To Do List: Ai Austin

Austin Tate's picture

This is an example plan for an agent.  It is offered as one simple example of what can be shown to agents who enter an I-Room. It is artificially constructed to very trivially show how planning aids, and presentation of options in the context of search and constraint management can radically increase the intelligence brought to bear to assist agents in coordinating their task lists and plans.

There is a powerful, yet simple and easily understandable, conceptual model behind the way that plans are presented, and against which agent capabilities are matched. It is the <I-N-C-A> constraint model of synthesized artifacts. Used for plans, it specifies a "set of constraints on the space of behaviour":

I- Issues to be addressed in the form of questions (e.g. using the 7 question types from the gIBIS methodology, Conklin)

N - Nodes to be added (e.g. typically activities to be perrformed in this context)

C - Constraints between node, domain objects and other elements (e.g. temporal, resource or spatial constraints)

A - Annotations (e.g. rationale, notes on alternatives and advice).


An example of a technical way to present a plan in this form is show here, though it is more usually shown graphically, or as a very simplified view, such as a "to do" list.

 

Issues

  • achieve (P=true) at (begin-of Y)?

Activities

  • perform X [Actions: refine using SOP-1, refine using SOP-2, Done, N/A]
  • perform Y [Actions: Done, N/A]
  • perform Z [Actions: Done, N/A]

Constraints

  • temporal (before (end-of X) (begin-of Y))   X---> Y
  • temporal (before (end-of Z) (calendar date: UT-2010-12-31-23:59:59))

Annotations

  • (P=true) will make the performance of Y more robust
  • (P=true) is an effect of SOP-1