Reasoning about Preferences in AgentsIJCAI 22 nd July, 2011 Reasoning About Preferences in...

18
Reasoning about Preferences in Agents IJCAI 22 nd July, 2011 Reasoning About Preferences in Intelligent Agent Systems Simeon Visser John Thangarajah James Harland Utrecht University RMIT University Amsterdam, the Netherlands Melbourne, Australia [email protected] {johnt,james.harland}@rmit.edu.au
  • date post

    22-Dec-2015
  • Category

    Documents

  • view

    217
  • download

    1

Transcript of Reasoning about Preferences in AgentsIJCAI 22 nd July, 2011 Reasoning About Preferences in...

Reasoning about Preferences in Agents

IJCAI 22nd July, 2011

Reasoning About Preferences in Intelligent Agent Systems

Simeon Visser John Thangarajah James HarlandUtrecht University RMIT University Amsterdam, the Netherlands Melbourne, [email protected] {johnt,james.harland}@rmit.edu.au

Reasoning about Preferences in Agents

IJCAI 22nd July, 2011

Introduction

“I want to go to Barcelona for a week in July. The hotel must be within 5 minutes walk of Las Ramblas, have a pool and gym, and I want a corner room with a great view of the city, unless the price is more than $120 per night. … Total budget is $2200.”“Done. The best I could do for the hotel was 10 minutes from Las Ramblas, with a pool but no gym or corner room, and for $135 per night. To meet the budget constraint, though, you have to fly on Dodgy Airlines.”

Reasoning about Preferences in Agents

IJCAI 22nd July, 2011

Introduction

“Arrgh! Dodgy Airlines! I’d rather stay in a youth hostel than travel with Dodgy Airlines!”

“Oops! If only I had known that ...”

Reasoning about Preferences in Agents

IJCAI 22nd July, 2011

Overview

Aim: incorporate preferences into Belief-Desire-

Intention (BDI) agents Flexible achievement of goals via plan libraries

Adapt to environmental changes

Extra input from user into computation process

User knows what she wants (and what she doesn’t)

Need to know preferences in advance

Not necessary for preferences to be exhaustive

Reasoning about Preferences in Agents

IJCAI 22nd July, 2011

BDI Agents

Beliefs

Goals

Plans

Events

Actions

Sensors, Messages, …

Actions

observe

think

act

Preferences

Preferences

Reasoning about Preferences in Agents

IJCAI 22nd July, 2011

BDI Execution

Holiday

Holiday Plan

Accommodation Transport

“Holiday time!”“Ok. So any of these combinations will do … “

“Maybe not all of them …”

AND

OROR

Reasoning about Preferences in Agents

IJCAI 22nd July, 2011

Motivations

Holiday

Holiday Plan

Accommodation Transport

1 2

“Which ones do you prefer?”

36 2 1 4 52 6 2 1 5

2

Reasoning about Preferences in Agents

IJCAI 22nd July, 2011

Using Preferences

Plan

Goal 1 Goal N

...

... ...

“Any order will do”

“Any choice will do”

Lisa’s order

Lisa’s order

Reasoning about Preferences in Agents

IJCAI 22nd July, 2011

Representing preferences

Preferences concern the effects of goals and plans

I prefer to minimize the money spent on accommodation I prefer to fly rather than travel by train I prefer to spend at most $500 on transport I prefer payment via credit card … If I can’t get a flight, then …

If the accommodation is a 5* hotel, I prefer to travel with Budget Airlines

Reasoning about Preferences in Agents

IJCAI 22nd July, 2011

Representing preferences

Adapt LPP language (Baier and McIlraith, 2007; Bienvenu et al., 2006)

accommodation.minimize(money) (0) transport.type = plane (0) >> transport.type = train (100) transport(money, 500,≤) (0) book_travel.payment = credit (0) … failure(book_flight) accommodation.type = hotel ∧ accommodation.quality = 5∗ : book_flight.airline = Budget (0)

Reasoning about Preferences in Agents

IJCAI 22nd July, 2011

BDI implementation

BDI process executes as normal, except for:

Generate goal-plan tree for given goal from plan library Calculate resource and property summaries for the goal Use preferences to guide execution of goal-plan tree

Prototype implementation:3,000 lines of Jadex (using metagoal and metaplan)

Reasoning about Preferences in Agents

IJCAI 22nd July, 2011

Computation

Holiday

Holiday Plan

Accommodation Transport

Hotel Backpacker Flight Train Bike

Property summary

Resource summary

Reasoning about Preferences in Agents

IJCAI 22nd July, 2011

Computation

Holiday

Holiday Plan

Accommodation Transport

Hotel Backpacker Flight Train Bike

Reasoning about Preferences in Agents

IJCAI 22nd July, 2011

Computation

Holiday

Holiday Plan

Accommodation Transport

Hotel Backpacker Flight Train Bike

Reasoning about Preferences in Agents

IJCAI 22nd July, 2011

Computation

Holiday

Holiday Plan

Accommodation Transport

Hotel Backpacker Flight Train Bike

Reasoning about Preferences in Agents

IJCAI 22nd July, 2011

Computation

Holiday

Holiday Plan

Accommodation Transport

Hotel Backpacker Flight Train Bike

Reasoning about Preferences in Agents

IJCAI 22nd July, 2011

Conclusions and Further Work

Preferences integrated into BDI execution Extra annotations to goal-plan tree Specifies order of evaluation when appropriate

Further work includes

Acquisition of preferences --- learning? Check preferences for conflicts Interactions between preferences Metrics for preference satisfaction Further implementation and empirical evaluation

Reasoning about Preferences in Agents

IJCAI 22nd July, 2011

Questions?