Agents Power and Norms Michael Luck Fabiola Lpez

  • Slides: 63
Download presentation
Agents, Power and Norms Michael Luck, Fabiola López y López University of Southampton, UK

Agents, Power and Norms Michael Luck, Fabiola López y López University of Southampton, UK Benemérita Universidad Autonoma de Puebla, Mexico

Part I Michael Luck, Fabiola López y López University of Southampton, UK Benemérita Universidad

Part I Michael Luck, Fabiola López y López University of Southampton, UK Benemérita Universidad Autonoma de Puebla, Mexico

Research Motivations § Agents have limited capabilities § The capabilities of others are needed

Research Motivations § Agents have limited capabilities § The capabilities of others are needed to succeed § Agents are autonomous § Benevolence cannot be taken for granted § Power can be used to influence agents § Powers are neither eternal nor absolute

Research Motivations § Agents and Societies § Societies achieve social order through norms. §

Research Motivations § Agents and Societies § Societies achieve social order through norms. § Agents must have a model of societies. § Agents must be able to recognise normative relationships. § Norms are dynamic concepts. § Agents must be aware of the changes due to norms.

Research Motivations § § § Societies and Autonomous Agents. How can autonomous agents be

Research Motivations § § § Societies and Autonomous Agents. How can autonomous agents be integrated into societies regulated by norms? What does an agent need to deal with norms? What does an agent evaluate before dismissing a norm? How are the goals of an agent affected by social regulations?

Overview § Autonomous Agents § Normative Multi-Agent Systems § Institutional Powers § Personal Powers

Overview § Autonomous Agents § Normative Multi-Agent Systems § Institutional Powers § Personal Powers § Conclusions

Aims § General: § To build a framework to represent agents able to exist

Aims § General: § To build a framework to represent agents able to exist in a society in which social order is achieved through norms. § Particular: § To provide a basic representation of norm-based systems. § To analyse the dynamics of norms. § To describe different kinds of normative relationships that agents might use in decision-making processes. § To identify powers in a society. § To identify personal powers of agents.

Overview § Norms and Normative Agents § Normative Multi-Agent Systems § Dynamics of Norms

Overview § Norms and Normative Agents § Normative Multi-Agent Systems § Dynamics of Norms § Norm Relationships § Conclusions

Multi-Agent Systems § Formal model based on Luck and d’Inverno’s SMART § Autonomous agents

Multi-Agent Systems § Formal model based on Luck and d’Inverno’s SMART § Autonomous agents are essentially defined in terms of their capabilities, goals, beliefs and motivations. § Multi-agent systems are collections of agents from which at least one is autonomous. § Multi-agent systems cannot exist without some interaction among their members. agent framework.

Normative Agents § A normative agent is an autonomous agent whose behaviour is shaped

Normative Agents § A normative agent is an autonomous agent whose behaviour is shaped by the norms it must comply with. § A normative agent must be § able to decide, based on its own goals and motivations, whether a norm must be either adopted or complied with § aware of the consequences of dismissing norms.

Normative Multi-Agent Systems § § A normative multi-agent system is a collection of normative

Normative Multi-Agent Systems § § A normative multi-agent system is a collection of normative agents which are controlled by a set of common norms varying from obligations and social commitments, to social codes. Normative multi-agent systems are characterised by § § the membership of some agents, the norms that members are expected to comply with, norms to enforce and encourage other norms, and norms to legislate.

Normative Systems: Membership § § § Autonomous agents join societies as a way to

Normative Systems: Membership § § § Autonomous agents join societies as a way to satisfy goals whose success relies on the actions of other agents. Members recognise themselves as part of the society by adopting some of its norms. Agents can be part of more than one society. Compliance with norms is never taken for granted. Enforcement and encouragement of norms are needed. Addressees of norms must be members of the system.

Normative Multi-Agent Systems § Disorder and conflicts of interest might appear • § §

Normative Multi-Agent Systems § Disorder and conflicts of interest might appear • § § § These faculties are restricted to specific sets of agents through special sets of norms. These norms specify how some agents have to behave when § § § when norms must be changed, and when punishments and rewards must be applied. norms must be changed, or norm becomes either fulfilled or unfulfilled. Fulfilment of norms is achieved when the corresponding normative goals become satisfied.

Normative Roles § From the different kinds of norms in a system, normative roles

Normative Roles § From the different kinds of norms in a system, normative roles for agents can be identified. § Legislators (addressees of legislation norms) § Defenders (addressees of either enforcement or reward norms)

Dynamics of Norms Issue Spread Modification Adoption Abolition Activation Compliance Reward Violation Sanction Dismissal

Dynamics of Norms Issue Spread Modification Adoption Abolition Activation Compliance Reward Violation Sanction Dismissal Non-sanction

Legislation norms Relations of authority legislators members

Legislation norms Relations of authority legislators members

Active norms Relations of responsibility addressees beneficiaries Relations of benefit nt me ce or

Active norms Relations of responsibility addressees beneficiaries Relations of benefit nt me ce or f En ns tio la re defenders

Fulfilled Norms addressees beneficiaries tt gh Ri m led lai oc tit En re

Fulfilled Norms addressees beneficiaries tt gh Ri m led lai oc tit En re s rd wa re s rd ive wa g to defenders

Violated Norms addressees Relations of deception ish n pu to ed titl En defenders

Violated Norms addressees Relations of deception ish n pu to ed titl En defenders beneficiaries

Norm Relationships § Norm relationships can be used by agents to: § To determine

Norm Relationships § Norm relationships can be used by agents to: § To determine empowered situations of agents. § To find reasons to adopt and comply with norms. § To find reasons to provide help. § To take advantage of social benefits in order to satisfy their goals.

Z Specification

Z Specification

Z Specification

Z Specification

Conclusions This work gives the means for agents to reason about norms by providing:

Conclusions This work gives the means for agents to reason about norms by providing: § § § A formal structure of norms that includes the different elements that must be taken into account when reasoning about norms. A formal basic representation of norm-based systems. An analysis and formalisations of the basic kinds of norms that norm-based systems have. An analysis of the dynamics of norms. The set of normative relationships that might emerge by adopting, complying and dismissing norms.

Part II Michael Luck, Fabiola López y López University of Southampton, UK Benemérita Universidad

Part II Michael Luck, Fabiola López y López University of Southampton, UK Benemérita Universidad Autonoma de Puebla, Mexico

Autonomous Agents § Formal model based on Luck and d’Inverno’s § Autonomous agents are

Autonomous Agents § Formal model based on Luck and d’Inverno’s § Autonomous agents are essentially defined in terms of their capabilities, goals, beliefs and motivations. § Interaction among agents results from one agent satisfying the goals of another. SMART agent framework.

Normative Multi-Agent Systems § Norms are mechanisms that a society has in order to

Normative Multi-Agent Systems § Norms are mechanisms that a society has in order to influence the behaviour of agents. § Categories of Norms: § § Obligations Prohibitions § Social Commitments Social Codes A normative agent is an autonomous agent whose behaviour is shaped by the norms it must comply with (AAMAS’ 02)

Normative Multi-Agent Systems § Norm Structure § Normative Goals § Addressees § Context §

Normative Multi-Agent Systems § Norm Structure § Normative Goals § Addressees § Context § Exceptions § Beneficiaries § Rewards § Punishments

Normative Multi-Agent Systems § Normative multi-agent system model at AAMAS’ 02) § Members §

Normative Multi-Agent Systems § Normative multi-agent system model at AAMAS’ 02) § Members § System norms § Legislation norms § Enforcement norms § Reward norms (RASTA’ 02

Normative Multi-Agent Systems § Legislation norms allow some agents to create, modify, and abolish

Normative Multi-Agent Systems § Legislation norms allow some agents to create, modify, and abolish the norms of the system. Legislation normative goals context Issue and abolition of norms permitted punishments rewards legislators . . .

Normative Multi-Agent Systems § Enforcement norms are norms which specify what kinds of punishments

Normative Multi-Agent Systems § Enforcement norms are norms which specify what kinds of punishments must be applied when norms are unfulfilled, and who is responsible for the punishment. Norm normative goals context punishments rewards addressees . . . context punishments rewards defenders . . . unsatisfied normative goals Enforcement normative goals

Normative Multi-Agent Systems § Reward norms are norms to specify who is responsible for

Normative Multi-Agent Systems § Reward norms are norms to specify who is responsible for rewards due to norm compliance. Norm normative goals context punishments rewards addressees . . . context punishments rewards defenders . . . satisfied normative goals Reward normative goals

Institutional Powers § Legislation norms Legal Power legislators members

Institutional Powers § Legislation norms Legal Power legislators members

Institutional Powers § Reward norms Legal Reward Power addressees defenders

Institutional Powers § Reward norms Legal Reward Power addressees defenders

Institutional Powers § Enforcement norms Legal Coercive Power defenders addressees

Institutional Powers § Enforcement norms Legal Coercive Power defenders addressees

Institutional Powers § System norms Legal Benefit Power beneficiaries addressees

Institutional Powers § System norms Legal Benefit Power beneficiaries addressees

Personal Powers § Agent capabilities to satisfy goals Ag (g 2) Ag satisfy (g

Personal Powers § Agent capabilities to satisfy goals Ag (g 2) Ag satisfy (g 1) Ag (g 2) Facilitation Power s fit ene b Ag satisfy (g 1) hin der s Ag (g 3) Ag satisfy (g 1) Illegal Coercive Power Ag (g 3)

Personal Powers § Agent benevolence towards a group of agents comrades Ag satisfy (g

Personal Powers § Agent benevolence towards a group of agents comrades Ag satisfy (g 1) Facilitation Power Ag (g 2) Ag satisfy (g 1) Comrade Power

Personal Powers § Agent rewarded by past actions Ag satisfy (g 1) Ag (g

Personal Powers § Agent rewarded by past actions Ag satisfy (g 1) Ag (g 2) Facilitation Power Ag (g 2) Ag satisfy (g 1) Reciprocation Power Ag (g 2) Ag satisfy (g 1) Fulfilled Norm Benefits

Personal Powers § Agents exchange goals Ag satisfy (g 1) Ag (g 2) Facilitation

Personal Powers § Agents exchange goals Ag satisfy (g 1) Ag (g 2) Facilitation Power Ag satisfy (g 3) Ag (g 2) Ag (g 4) Exchange Power Ag (g 4) Facilitation Power Ag (g 4) Ag (g 2) Exchange Power

Z Specification

Z Specification

Conclusions § This work gives the means for agents to identify power in their

Conclusions § This work gives the means for agents to identify power in their current situations of powers in which they are. § § Uses a formal model of systems regulated by norms. Analyses powers due to the role agents play in a society. Analyses powers due to an agent’s capabilities. Provides a taxonomy of powers.

Part III Michael Luck, Fabiola López y López University of Southampton, UK Benemérita Universidad

Part III Michael Luck, Fabiola López y López University of Southampton, UK Benemérita Universidad Autonoma de Puebla, Mexico

Research Motivations v Societies and Autonomous Agents. v How can autonomous agents be integrated

Research Motivations v Societies and Autonomous Agents. v How can autonomous agents be integrated into societies regulated by norms? v What does an agent need to deal with norms? v What does an agent evaluate before dismissing a norm? v How are the goals of an agent affected by social regulations?

Overview § Norms and Normative Agents § The Norm Compliance Process § Strategies for

Overview § Norms and Normative Agents § The Norm Compliance Process § Strategies for Norm Compliance § Experiments with Normative Agents § Conclusions and Additional Work

Norms and Normative Agents § Norm adoption is the process through which an agent

Norms and Normative Agents § Norm adoption is the process through which an agent decides to create an internal representation of a norm. § Norm compliance is the process through which an agent’s goals are updated according to the norms it has decided to comply with.

Norms and Normative Agents § A normative agent is an autonomous agent whose behaviour

Norms and Normative Agents § A normative agent is an autonomous agent whose behaviour is shaped by the norms it must comply with. § A normative agent must be § able to decide, based on its own goals and motivations, whether a norm must be either adopted or complied with. § aware of the consequences of dismissing norms.

Norms and Normative Agents § § Compliance with norms is § enforced through punishments,

Norms and Normative Agents § § Compliance with norms is § enforced through punishments, and § encouraged through rewards. Neither punishments nor rewards are effective without being related to the current goals of an agent. § Punishments must hinder important goals. § Rewards must benefit important goals.

Norm Compliance: norm processing active norms intended norms rejected norms

Norm Compliance: norm processing active norms intended norms rejected norms

Norm Compliance: affected goals normative goals hindered by normative gs rewards benefited from rewards

Norm Compliance: affected goals normative goals hindered by normative gs rewards benefited from rewards punishments hindered by punishments intended norms rejected norms

Norm Compliance: updating goals hindered by normative gs goals current goals normative goals benefited

Norm Compliance: updating goals hindered by normative gs goals current goals normative goals benefited from rewards hindered by punishments

Strategies for Norm Compliance § Social § § All norms are rejected. intended norms

Strategies for Norm Compliance § Social § § All norms are rejected. intended norms Fearful § § All norms are complied with. Rebellious § § norm A norm including punishments is always complied with. Greedy § A norm including rewards is always complied with. norm with punishment norm with reward

Analysis of active norms nonconflicting norms hindered by normative gs conflicting norms = hindered

Analysis of active norms nonconflicting norms hindered by normative gs conflicting norms = hindered by normative gs active norms

Pressured Strategy Non-conflicting norms are complied with if their punishments hinder any existing goal.

Pressured Strategy Non-conflicting norms are complied with if their punishments hinder any existing goal. hindered by normative gs hindered by punishments = non conflicting norms intended norms

Pressured Strategy Conflicting norms are complied with if the goals hindered by punishments are

Pressured Strategy Conflicting norms are complied with if the goals hindered by punishments are more important than the goals hindered by normative goals. hindered by normative gs hindered by punishments conflicting norms > hindered by normative gs intended norms

Opportunistic Strategy Non-conflicting norms are complied with if the offered rewards might benefit a

Opportunistic Strategy Non-conflicting norms are complied with if the offered rewards might benefit a goal. hindered by normative gs benefited from rewards = non conflicting norms intended norms

Opportunistic Strategy Conflicting norms are complied with if associated rewards benefit more important goals

Opportunistic Strategy Conflicting norms are complied with if associated rewards benefit more important goals than those that might be hindered by normative goals. hindered by normative gs benefited from rewards conflicting norms > hindered by normative gs intended norms

Z Specification

Z Specification

Z Specification

Z Specification

Experiments with Normative Agents Agent Strategies for non Strategies for conflicting norms Social Rebellious

Experiments with Normative Agents Agent Strategies for non Strategies for conflicting norms Social Rebellious Selfish Pressured & Opportunistic Social-Selfish

Experiments with Normative Agents § Individual performance is the proportion of personal goals that

Experiments with Normative Agents § Individual performance is the proportion of personal goals that become satisfied under the presence of norms. § Social contribution represents the proportion of norms complied with by an agent who has its own goals. § Experiments were run § by varying the number of conflicts between the goals of an agent and the normative goals of the corresponding norms (from 0% to 100%), and § by taking different sizes for the sets of current goals and active norms.

Experiments with Normative Agents § § Internal and external conditions were similar for all

Experiments with Normative Agents § § Internal and external conditions were similar for all agents. § Agents have similar goals § Similar norms become active at the same time. § The importance of each goal is also the same for all agents. Complete social control was assumed. § All punishments were applied. § All offered rewards were given.

Experiments with Normative Agents

Experiments with Normative Agents

Conclusions § A formal structure of norms that includes the different elements that must

Conclusions § A formal structure of norms that includes the different elements that must be taken into account when reasoning about norms. § A formal model to incorporate the process of normcompliance into a BDI-like agent architecture. § A set of strategies that agents might follow to decide when norms must be complied with. § Different ways to combine strategies to define complex normative behaviours. § An analysis of normative agent behaviour when total social control is exerted.