Team Formation between Heterogeneous Actors
Arlette van Wissen
Virginia DignumKobi Gal
Bart Kamphorst
Introduction
How to make decisions in a fast-paced and dynamic domain in which humans and agents interact?
The Problem
2
Introduction
How to make decisions in a fast-paced and dynamic domain in which humans and agents interact?
The Problem
3
IntroductionTeam Formation
N - TT
N
4
To become part of successful teams, one has to deal with the followingsocial dilemmas:
team formationworking together?allocation of payoff?
team maintenance (intention reconciliation)staying together?
- Identify the tradeoffs between fairness, trust and participant type in dynamic team interactions.
- Derive principles that can be used to construct agents that are able to participate effectively in these settings.
IntroductionTeam Formation
5
Introduction
How to make decisions in a fast-paced and dynamic domain in which humans and agents interact?
Humans are not completely rational actors. (Kagel & Roth, 1995)
Social factors influence human behavior.
(Loewenstein, 1989, Camerer, 2003)
The Problem
6Fairness
Introduction
How to make decisions in a fast-paced and dynamic domain
in which humans and agents interact?
- changing environment- no predefined teams- non-binding agreements
Trust
The Problem
7
Introduction
How to make decisions in a fast-paced and dynamic domain in which humans and agents interact?
Social factors influence cooperation between humans and agents.
(van Wissen et al., 2009)
The Problem
8Actor Type
People behave differently towards agents and other people.
(Blount 1995, Sanfey 2003 )
Related WorkTeam Formation
Existing work on human-agent team formation:
• Uses models that generally do not consider trust or nature.
• Focuses on formal analysis.
• Uses simplifying assumptions.
9
Experimental DesignPackage Delivery Domain
Fast paced domain with uncertainty and commitment.
Team Formation: initiators and members.
Social dilemmas:1. working together?2. allocation of payoff?3. staying together?
10
Experimental DesignPackage Delivery Domain
11
Experimental DesignPackage Delivery Problem in Colored Trails
12
Experimental Design
Fairness, Trust
Empirical Methodology
ability to choose team members
&transferable
payoffActor Type
repeated interaction, defection
deception13
Experimental DesignPackage Delivery Problem in CT
- 6 players per game (P)- 3 colors (P/2)
14
Experimental DesignPackage Delivery Problem in CT
- 6 players per game (P)- 3 colors (P/2)
15
Experimental DesignPackage Delivery Problem in CT
- 6 players per game (P)
- 3 colors (P/2)- 6 large packages
(P)- 12 small packages(2P)
16
Experimental DesignPackage Delivery Problem in CT
team formation game with:
- payoff small package = 3
payoff large package with 2 players = 60
payoff large package with 3 player = 180
- imperfect information:players do not have knowledge of all actions and behavior of the players in the game and have partial visibility
17
18
Experimental Design
movie
19
The Experiment
2 experiments, 18 subjects
44% male, 56% female
72% students
5 rounds of 5 – 14 min
payment corresponding to performance
20
Results & DiscussionTeam Formation
Result 1
These factors are shown to significantly affect performance:
joining teams (pearson correlation, r = 0.56)
initiating teams (pearson correlation, r = 0.42)
delivering packages individually (pearson correlation, r = 0.31)
21
Results & DiscussionTeam Formation
Result 2Subjects showed a preference to interact with those they successfully interacted with before.
22
trust historynumber of times subjects successfully
cooperated in the pastin any team configuration
likelihood of future interaction
Results & DiscussionTeam Formation
23
Results & DiscussionIntention Reconciliation
24
Result 3Trust is more important to the decision of defection than the height of outside offers.
> No significant increase of payoff for accepted outside offers. > Small but significant correlation between accepted outside offers
and trust value of initiator (0.29).
Results & DiscussionParticipant Type
Result 4Players offer humans significantly more fair splits than they offer agents.
(combined t-test, p < 0.0001)to people: avg. 94% fairto agents: avg. 82% fair
100 % fair 50 % fair
2-player team 30 15
3-player team 60 30
Result 5The nature of participants does not significantly affect:1. the choice of team partners2. the acceptance of offers3. defection behavior
25
Results & DiscussionParticipant Type
26
Conclusions
People are just as loyal and trusting towards agents
as to humans.agent-initiated teamwork and
working alongside autonomous systems(search-and-rescue, personal assistants, decision
support)
People offer agents less, thereby valuing them
differently from humans. agents need to behave and appear
natural and ‘human-like’ (e-commerce, bidding, games, personal assistant)
27
Conclusions
People prefer to work with players they have
successfully worked with before. refer to previous interactions,
have memory(games, companions)
28
Future Work
1. Computational Model
2. Belief Desire Intention (BDI) Agents
3. Pre-established payoff distributions
29
Thank you.
30
?31
Game FlowInitiator
32
Game FlowInitiator
33
Game FlowMember
34
Game FlowMember
35
Related WorkTrust
For an actor a to be said to trust another actor b with respect to a particular goal g, a must have the following beliefs (castelfranchi, 1998, 2001) :
• Competence Belief: b is useful for achieving g and is able to provide the expected result • Disposition Belief: b is not only capable, but also willing to do what is necessary to achieve g • Dependence Belief: the results and rewards of achieving g depend on the involvement of b • Fulfillment Belief: g will come about due to b’s involvement
36
Results & DiscussionNature
Result 7The nature of participants does not significantly affect the choice of team partners or the acceptance of offers.
37
Introduction
Types of interactions:- cooperative and helpful interactions- competitive interactions- cooperative interactions in competitive scenarios
non-cooperative games - self-interested actors who can make non-binding
commitments - basic modeling unit is the individual
cooperative games- players can make binding commitments- communication and negotiation between the players is allowed- groups of players (teams) may enforce cooperative behavior
- basic modeling unit is the group
team Formation
38
Conceptual Design
Assumptions in CF
Interchangeability
Prev work
All potential members bring the same utility to a team.
Our work
Some potential members are preferred over others.
Membership
Prev work
Membership to a team is equally available to everyone.
Our work
Members have to meet certain requirements for membership to be available to them.
Conflict
Prev work
Conflict is eliminated by making agreements binding.
Our work
No conflict-free environment since agreements are not binding.
39
40
41
42
43
44
45
Introduction
How to make decisions in a fast-paced and dynamic domain in which humans and agents interact?
The Problem
46
What strategy did you use for choosing your team members?
``Whether members were reliable. I could give computers a smaller share without feeling guilty. I did build a good history with one other human player.’’
``I chose the computers mostly, since I thought that they would demand less points.’’
Results & DiscussionNature
47
Conceptual Design
Identify the extent to which these factors affect behavior in heterogeneous systems.
Derive principles that designers of such systems could use to construct agents that are able to participate effectively in teams.
How do nature and trust influence people’s decisions
in mixed team formations?
Objective
48
Outline
1. Introduction2. Related Work
3. Conceptual Design4. Experimental Design5. The Experiment
6. Results & Discussion7. Conclusions
8. Future Work 9. Contributions
49
Results & Discussion
1. Survey
- preferences - strategies
2. Log
- proposals- teams- defections
50
Experimental Design
team formation consists of two steps:
1. The initiator invites members who thereafter join the team.
2. The team delivers the package. attempted team
Package Delivery Domain
51
i
?
Experimental Design
team formation consists of two steps:
1. The initiator invites members who thereafter join the team.
2. The team delivers the package. formed team
Package Delivery Domain
52
i
Results & Discussion
Team formation consists of two steps:
1. The initiator invites members who thereafter join the coalition.
2. The coalition delivers the package.
formed teamA team is formed when the initiator has invited the
requiredmembers for a team and the members have agreed to
be part ofthat team.
Package Delivery Domain
53
Results & Discussion
Team formation consists of two steps:
1. The initiator invites members who thereafter join the coalition.
2. The coalition delivers the package.
successful teamA successful team is a formed team that succeeded to
delivera large package to the goal.
Package Delivery Domain
54
Results & Discussion
Team formation consists of two steps:
1. The initiator invites members who thereafter join the team.
2. The team delivers the package.
failed teamAn unsuccessful team is a team that was dissolved (bydefection) before it was able to deliver the package.
Package Delivery Domain
55
Results & DiscussionIntention Reconciliation
Result 5Players with a high defection rate are as successful as those with a low defection rate.
56
Results & Discussionteam Formation
Result 2Players created more 3-player teams than 2-player teams.
(goodness of fit, p < 0.00001)
Result 3No difference in defection rates in 2- and 3-player teams. 57
Results & DiscussionTeam Formation
successful teams frequency
0 1 0 2 0 4 0 5 1 2 11 3 1 5 2 3 2 4 23 4 13 5 4 5 0 1 20 1 50 2 40 4 51 2 3 121 3 52 3 4 23 4 5 1
58
Top Related