350 likes | 509 Views
Managing Social Influences through Argumentation-Based Negotiation. Present by Yi Luo. Fifth International Joint Conference on AUTONOMOUS AGENTS AND MULTIAGENT SYSTEMS (AAMAS 2006) Workshop: Argumentation in Multi-Agent Systems (ArgMAS)
E N D
Managing Social Influences throughArgumentation-Based Negotiation Present by Yi Luo
Fifth International Joint Conference on AUTONOMOUS AGENTS AND MULTIAGENT SYSTEMS (AAMAS 2006) Workshop: Argumentation in Multi-Agent Systems (ArgMAS) Nishan C. Karunatillake1, Nicholas R. Jennings1, Iyad Rahwan2, Sarvapali D. Ramchurn1. Managing Social Influences through Argumentation-Based Negotiation School of Electronics and Computer Science, University of Southampton, Southampton, UK. Institute of Informatics, The British University in Dubai (Fellow) School of Informatics, University of Edinburgh, Edinburgh, UK. Paper in workshop of AAMAS-06
Background • internal influences Vs. social influence • Internal: intrinsic motivations • External: role, relationship • Example: a teacher is trying to sell a book to his student • Incomplete knowledge: don’t know status in society • Conflict between internal and social influence
Background • Argumentation-based negotiation (ABN) • exchange additional meta-information such as justifications, critics, and other forms of persuasive language • gain a wider understanding of the internal and external influences
Background • Objective: • Propose a ABN framework allowing agents detect, manage and resolve conflicts • Giving agents the capability to challenge their counter parts and obtain the reasons for violating social commitment • simulate to compare the result for agents with and without argumentation in the social context
Social Argumentation Model • Social influence schema • Social Arguments • language and protocol • Decision functions
Social Argumentation Model: Social Influence Schema • Social commitment x: debtor y: creditor θ: action Social commitment is a commitment by agent x to another agent y to perform a stipulated action θ • x attains an obligation toward the y to action θ • y attains certain right to demand (compensation) or require the performance of θ • relationship: encapsulation of social commitments between associated roles
Social Argumentation Model: Social Influence Schema • Act (x, student) and RoleOf(student, student-teacher-relationship)= In (x, student, student-teacher-relationship)
Social Argumentation Model: Social Influence Schema • every two agents combined with an action can be associated together as a social commitment • A set of SCs can be associated together as a relationship • Every two roles in the society can have a relationship
Social Argumentation Model:Social Arguments • Socially influencing decision: argue about validity of reasoning • Dispute a1 is in role r1, SC is a social commitment associated with relationship p • Rebut agent is also is another role which associate another action • Rebut conflicts between two existing obligations, rights and actions • Negotiating social influence: trading • promise to undertake future obligation • Promise not to exercise certain right
Social Argumentation Model:Language and Protocol • Domain language + communication language= Utterance • Domain language: premise about social context conflicts that may face while executing actions • Communication language: elocutionary parties OPEN-DIALOGUE, PROPOSE, ACCEPT, REJECT, CHALLENGE, ASSERT AND CLOSE-DIALOGUE
Social Argumentation Model:Language and Protocol • Protocol • Opening • Conflict recognition: initial interaction, bring the conflict in surface • Conflict diagnosis: establish root cause of the conflict • Conflict management: allows agents to argue addressing the cause of this conflict • Agreement: mutually acceptable solution or agreeing to disagree • Closing
Social Argumentation Model: Decision Making Functionality • Challenge the rejection / end negotiation / forward an alternative proposal • Generating a proposal • If it is capable of performing the reward • If the benefit it gains from the request is greater than the cost of reward • Evaluating a proposal • if it is capable of performing the request • The benefit of the reward is greater than the cost incerred in performing the request
Argumentation Context • Scenario: task allocation • Self-interested agents interact to obtain services to achieve a given set of actions • Agent has: • A list of actions that is required to achieve • Capability to perform actions
Argumentation Context: Scenario • Capability: type + level • Actions: time + capability type + minimum capability level + reward
Argumentation Context: modeling Social Influence • Role-relationship structure • Associated degree of influence: decommitment penalty • Assign roles to actual agents
Argumentation Context: modeling Social Influence Agent a0: • Obligation to provide: - c0 to an agent acting r1; obliged to pay 400 if decommitted. - c1 to an agent acting r1; obliged to pay 100 if decommitted. • Rights to demand: - c0 from an agent acting r1; right to demand 200 if decommitted.
Argumentation Context: modeling Social Influence • Test how agents use argumentation to manage and resolve conflicts created due to incomplete knowledge about their social influence • Provide only a subset of the agent-role map: • perfect knowledge (0% missing knowledge) • Completely unaware of social influence (100% missing knowledge)
Argumentation Context: Agent Interaction • An agent requires a certain capability will generate and forward proposals to another agent, asking him to sell its service in exchange for a certain reward (algorithm 1): propose (do (aj, θj), do (ai, m)) • If the receiving agent perceives this proposal to be viable and believes it is capable of performing it, then will accept it. Otherwise it will reject the proposal (Algorithm 2).
Argumentation Context: Agent Interaction • In case of a reject, the original proposing agent will attempt to forward a modified proposal. The interaction will end either when one of the proposals is accepted or when all valid proposals that the proposing agent can forward are rejected (Algorithm 3). • agents argue: (algorithm 4) • detect conflicts by analyzing the decommitment penalties • Try to resolve it by exchanging their respective justifications • If there are inconsistencies, social arguments are used • If they are both valid, then each agent would point-out alternative justifications via asserting missing knowledge • The defeat-status is computed via a validation heuristic, which simulates a defeasible model
Managing Social Influences • Demanding compensation: Right to demand compensation and the right to challenge non-performance of social commitment
Managing Social Influences • Observation 1: The argumentation strategy allows agents to manage their social influences even at high uncertainty levels. • Observation 2: In cases of perfect information and complete uncertainty, both strategies perform equally. • Observation 3: At all knowledge levels, the argumentation strategy exchanges fewer messages than the non-arguing one.
Managing Social Influences • Observation 4: When there are more social influences within the system, the performance benefit of arguing is only significant at high levels of knowledge incompleteness.
Managing Social Influences • Questioning non-performance Argue-In-First-Rejection and Argue-In-Last-Rejection • Observation 5: The effectiveness of the various argumentation strategies are broadly similar • Observation 6: Allowing the agents to challenge earlier in the dialogue, significantly increases the efficiency of managing social influences.
Conclusion • The incomplete knowledge and the diverse conflicting influences may prevent agents from negotiation • in order to function as a coherent society, agents require a mechanism to manage their social influences in a systematic manner. • Argumentation based approach improve the multi-agent system to form an agreement more effectively and efficiently.
Questions? Thank you