Download presentation
Presentation is loading. Please wait.
1
Improving Collaboration with Shared Preferences
Loïs Vanhée1,2, Frank Dignum1, Jacques Ferber2 1Utrecht Universiteit 2Université Montpellier 2 Paris, 6th of May 2014
2
Fire fighting Complex problem Life-threatening coordination
Situation seems safe. Others check. Go extinguishing! Checking for victims Pulling water hose Hope others check for threats Complex problem Life-threatening coordination Current methods: help but insufficient What if they shared similar values? Safety > victims > fire Goal: using values for designing MASs
3
Agenda Values Shared Values Design Challenges Conclusion
4
Ordering agent decisions outcomes
Values: Preferences Single Agent Ordering agent decisions outcomes Need comparable outcomes Too complex for concreteness Which foot to start walking? Huge preference order! Combine with a traditional decision process Include anchor points in the code for value-related decisions Easily: abstract core decisions Selection of plans & concrete goals Opportunity: concrete decisions Safety+++ Efficient+++ Plans Check hazards Safety+++ Efficient++ Extinguish Safety++ Efficient+++ Safety++ Efficient++ Safety+++ Efficient+ Safety+ Efficient+++ Safety++ Efficient+ Tom van der Weide 2011
5
Example Safety+++ Efficient+++ clearRoom(): anchor_point_plan_selection { plan: clearRoom()<-true|{openDoor();waterIfNecessary()} expected_outcome: efficiency(high), safety(B(estimateSafetyBlindOpen())) plan: clearRoom()<-true {carefulCheck();openDoor(); carefulCheck();waterIfNecessary()} expected_outcome:efficiency(okay),safety(high) } Safety+++ Efficient++ Safety++ Efficient+++ Safety++ Efficient++ Safety+++ Efficient+ Safety+ Efficient+++ Safety++ Efficient+ Can also be done with goals: Abstract values help selecting pragmatic goals
6
Let’s go multi-agent! Driving the system with shared values
Outcome-related Traditional preferences Behavior-related Corporate values Plans Safety+++ Efficient+++ Check Safety+++ Efficient++ Extinguish Safety++ Efficient+++ Safety++ Efficient++ Safety+++ Efficient+ Drive coordination towards desirable collective action Safety+ Efficient+++ Safety++ Efficient+ Timeliness: Schedules are reliable Everyone is on time Abstractly drive collaboration towards system goals (synergy) Individuals use schedule to be on time More examples in social science studies about culture!
7
Using shared values Designing agents Create expectations Guidelines
Other individuals Society, environment Guidelines Help coordination Directly: matching decisions Prefer to be on time Indirectly: helps cooperation Easily agree on plans favoring timeliness
8
Design challenges Traditional tradeoff: restriction for coordination
Scope and costs limitations Values: abstract preferences (new scope!) Achieving collaboration Abstractly drive agents towards system goals Achieving coordination Generic values patterns Inspiration: cultural studies Main limitations Soft: Compliance with values hard to assert Abstract: Need tactic layer
9
Discussion Shared values helps making decisions
Complementary tool for concrete decisions (selecting pragmatic goals, plans…) Shared values help driving collective action Collaboration Coordination Can be combined with other methods See norms with Jie’s poster, later this week Shared values offer a new perspective for designing MAS that we will explore next months But after the coffee break
10
Thank you for your attention
11
Illustration with Game Theory
Sharing Preferences Illustration with Game Theory Unshared preferences Shared preferences Her Her Trust Check Trust Check Me Me Closest sponge -2 -1 Closest sponge -2 -1 2 -1 1 -1 Cleanest sponge 2 1 Cleanest sponge 2 1 1 1 2 2 Local predictable optimum Nash Equilibrium Long-term goals: collaboration
12
Design Challenges Difficulty
Individual influences, collective consequences Challenge help making decisions without being restrictive Preferences Adding preferences System designer side Strategic Evaluation Evaluation Evaluation Tactic Plan Plan Plan Agent designer side Designing plans Challenge Solutions must match with preferences
13
To my defense… That one was (almost like) new!
Cleaning the dishes Need a sponge, standard plan: pick the closest one: green But, I was not alone! What are you doing? green sponge for cleaning the walls, not the dishes! We have to rewash! To my defense… That one was (almost like) new! A few seconds later… Need to dry hands, standard plan: pick the closest one: green What are you doing? green is for drying dishes, not your hands!
14
What’s the core of the problem?
Green actions are more efficient, they are better! Green actions are dirty, blue ones are better! Green actions are not so dirty! Yes they are! Different values → disagreements Efficiency versus cleanliness Maybe sharing values helps avoiding agreements One values solves two disagreements (and more)! Also work for preventing collaboration failures Sounds like good for MASs! Question: can values be used in MAS design? Now: introduce a new research perspective
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.