Timothy J. Norman Challenges for Effective Open Virtual Organisations Timothy J. Norman Department of Computing Science University of Aberdeen [email protected]

Timothy J. Norman

Challenges for Effective Open Virtual Organisations

Timothy J. Norman

Department of Computing Science

University of [email protected]

8th Annual International Workshop on Engineering Societies in the Agents World

Timothy J. Norman

• Open Virtual Organisations– ADEPT– CONOISE/CONOISE-G– e-Institutions

• Semi-Structured Processes in Open Societies– Challenge 1: machine-readable organisations– Challenge 2: recognising the inevitability of conflict– Challenge 3: machinery for resolving conflicts

• Towards mixed initiative (Human-Agent) VOs• Conclusions

Outline

Timothy J. Norman ESAW 2007, Athens 2 of 26

Timothy J. Norman

• What are OVOs?– Open systems – components (agents) may

come and go– Organisation – system has some degree of

established structure & operation

• A tension…– Suppose some agent plays a unique role in

an organisation; that agent leaves!– Suppose a new agent appears offering a

new service; restructure the organisation?

Open Virtual Organisations


Timothy J. Norman

Society vs. Process

processclosed open

society

ope

ncl

ose

d

Example 1

ADEPT

Domain: business process management.

N. R. Jennings, P. Faratin, T. J. Norman, P. O'Brien and B. Odgers (2000) Autonomous Agents for Business Process Management. Int. Journal of Applied Artificial Intelligence 14 (2) 145-189.

ADEPT


Timothy J. Norman

ADEPT

IdentifyRequirementsProfile

CaptureCustomerDetails

CaptureCustomerRequirements

VetCustomer

IdentifyService

FinalCosting

ProvideQuote

CostNetwork

Survey andDesignNetwork

AnalyseRequirements

LegalReview

CustomerOK?

PortfolioItem?

Legal?

Goto A

terminate

A

Yes

Yes

Yes

No

No

No

Business process Organisational structure

negotiation andservice management

DesignerAgency

SurveyTeam

QuantitySurveyor

Design teamagency

peers

Services Service-levelagreements

negotiation andservice management

Credit VettingAgency

LegalAgency

Sales teamAgency

peers

peers

peers

peers

subsidiary

subsidiary

Responsible agents

(service name Vet_Customer inputs ( CCL_Customer customer_details cli man ) outputs( CCL_Decision verdict ) guard ("") body ( sequence { check_CCL ( details = service::customer_details service::verdict = limit ) } -> ( check_CCL ) ) )

=> Service descriptions

=> Enactment through- SLA negotiation- Service & task execution- renegotiation


Timothy J. Norman

+ Decentralised workflow management+ Flexibility – task/service scheduling,

automated SLA (re-)negotiation– Process fixed at design time:

– System organisation– Services

– No workflow-level operational constraints or analysis– Throughput– Service delivery times– Quality control

ADEPT: + & -


Timothy J. Norman

Society vs. Process

processclosed open

society

ope

ncl

ose

d

Example 2

CONOISECONOISE-G

Domain: telecom services configuration.

T. J. Norman, A. Preece, S. Chalmers, N. R. Jennings, M. Luck, V. D. Dang, T. D. Nguyen, V. Deora, J. Shao, W. A. Gray and N. J. Fiddian. (2004). Agent-based formation of virtual organisations, Knowledge-Based Systems 17:103 111.

ADEPT

Increased system flexibility/adaptation


Incr

ease

d ne

ed to

man

age

trus

t (s

ecur

ity &

priv

acy

mor

e co

mpl

ex)

Incr

ease

d he

tero

gene

ity o

f sy

stem

com

pone

nts

Timothy J. Norman

Offers received SP1 – 10 movies pcm, hourly news SP2 – hourly news SP3 – 120 texts AND 30 mins SP4 – 5 movies pcm, 30 mins

Agents may Bid as an individual Bid representing an existing VO Form a VO to bid

Agents prioritise commitments: may break a prior commitment to bid

Package required by customer Movie subscription News service >50 free text messages per month 30 free phone minutes per month

CONOISE

Quality of offers assessed

Best offer combination identified through reverse auction


Timothy J. Norman

• Open system– Agents may come and go– New services (or service packages) may

be offered

• Open process– VOs exist for lifetime of service– Workflows composed of chained VOs

• No designed-in organisational structure– All agents are peers in competition– ADEPT-like workflow structure can emerge

CONOISE


Timothy J. Norman

Society vs. Process

process

closed open

society

ope

ncl

ose

d

Example 3

e-Institutions

Agents may come and go

Governed by institutional rules

Process defined by institution

A. García-Camino, J.-A. Rodríguez-Aguilar, C. Sierra, and W. Vasconcelos. (2006). Norm-Oriented Programming of Electronic Institutions. Int. Joint Conf. Autonomous Agents and Multi-Agent Systems.

ADEPT

Let’s look a bit closer at this space


Timothy J. Norman

Closed or Open Processes?

processesclosed open

society

ope

n

Institution engenders trust (but agents must trust institution)Agents not permitted to act outside the protocolSecurity and privacy of information honouredQuality control

Agents must decide who to trustFailure may affect reputation → distrustWhat can/should be done & how information is used must be agreed

e-Institutions encapsulate common episodal processes – why require agents to coordinate these at run-time?

But which e-Institution offers the coordination services required?

Challenge 1: Institutional rules/policies must be machine-understandable

e-Institutions


Timothy J. Norman

Semi-Structured Processes

processclosed open

society

ope

n

e-Institutions

semi-structured

Machine-understandable institutions

M. J. Kollingbaum & T. J. Norman (2002). Supervised Interaction, AAMAS

• A language for describing contracts

contract ( LoC, role ( ?cust ) role ( ?sup ) role ( ?bank ) obl ( … ), per ( … ) … )

• Reasoning machinery to interpret and negotiate contracts

• Institutional elements to “host” contract enactment


Timothy J. Norman

• Project PI (agent α) has contract with funding body, β, such that α is:

Example: eScience

Oα:pireport_results(α, R)

FX:Ypublish(D)

Fα:piclaim(X)Pα:piclaim(staff_costs)Pα:piclaim(travel)

– Obliged to report experimental results

– Forbidden from publishing source data

– Limited to spending project funds on staff and travel costs

β

α


Timothy J. Norman

• Suppose α needs to sub-contract a task to generate results

• It has two options:

Example: sub-contracting

OX:Ypublish(D)

OX:Ypay(fee)we also know thatpay(X) A:R→A:R claim(X)

a. Agent γ is a publicly funded organisation that performs the task for free but requires data to be published

b. Agent δ is a private organisation that charges a fee

β

α

γ

δ


Timothy J. Norman

• Regardless of what α does, it will violate the contract with β

Example: conflict

Oα:pireport_results(α, R)FX:Ypublish(D)

Fα:piclaim(X)

Pα:piclaim(staff_costs)

Pα:piclaim(travel)

β

α

γδ

1. No sub-contract


Fα:piclaim(X)


Pα:piclaim(travel)


Fα:piclaim(X)


Pα:piclaim(travel)

2. Sub-contract to γOX:Ypublish(D)

2. Sub-contract to δOX:Ypay(fee)pay(X) A:R→A:R claim(X)

Challenge 2: agents must be able to

recognise conflict situations


Timothy J. Norman

• Informs agent of the implications of signing a contract (adopting the policy of an e-Institution)

• Enables focussed deliberation on what to violate, and hence what sanctions may be imposed

• Agents have social autonomy

• Could also guide conflict resolution…

Utilising Detected Conflicts


Timothy J. Norman

• NoA (Kollingbaum & Norman)– Normative Architecture– Encodes possible plans and active norms in an

adapted RETE network– Efficiently identifies whether a new norm is in

conflict and what with (given options for action)

• FOUND! (Vasconcelos, Kollingbaum & Norman)– First-Order Unification for Norm Deliberation– Norms combined with constraints on their

application and domain axioms– Also tells us how the conflict can be resolved

Conflict detection mechanisms


Timothy J. Norman

Example: NEO

Non-combatant Evacuation OperationReports are received of NGO workers requiring assistanceTop priority mission to evacuate NGO workersTask allocated to Team A commanderTeams A and B are different coalition partners

• Team A– Based on carrier off

coast (South)– UAVs with sensors to

provide on-going visual surveillance

– Group of helicopters within range of NGO workers

• Team B resources– Based to the North-East

on land– Group of helicopters

within range of NGO workers

– Mechanised infantry


Timothy J. Norman

Example: NEO Norms

• Most effective plan: deploy UAV; assess situation; deploy helis; coordinate with UAV intelligence

• Team A operates under the following norms:– Commander is obliged to evacuate the NGO

workers (from NEO mission policy)– Team A is forbidden from sharing UAV-obtained

intelligence with other coalition partners (from coalition operations policy)

– Helicopters are forbidden from flying in bad weather (from safety policy)


Timothy J. Norman

Example: NEO Enactment

• AUV is deployed• Information from flight operations:

deteriorating weather conditions• Conflict:

– Continuing with Heli operation from carrier would violate safety policy

– Delegating Heli evacuation to team B would violate coalition operations policy in sharing UAV intelligence

– Failing to continue with NEO will violate mission policy


Timothy J. Norman

Need for Conflict Resolution

• Time and safety-critical scenario• Norm violation not an option• Must resolve conflict

– Automatically, or– By requesting human intervention into the

decision-making

Challenge 3: Agents should have machinery to suggest resolutions to or to automatically resolve conflicts


Timothy J. Norman

Example: Conflict Resolution

• FOUND! used to identify norm conflict• FOUND! offers resolution on the basis of meta-

policy (or conflict resolution strategy) legis superioris

• Proposal presented to Team A commander:– Curtail coalition operations policy regarding UAV

intelligence sharing in this instance– Delegate Heli evacuation of NGO workers to team B– Instruct AUV intelligence group and team B Heli

group to coordinate the rescue mission


Timothy J. Norman

Human-Agent Teamwork

• This example – Further demonstrates the utility of norm conflict

detection and resolution– Illustrates the need to refocus on teamwork– But not agent teamwork (ala. Cohen & Levesque;

Tambe; etc.), humans and agents working as a team

• Agents can use these techniques to aid humans in making complex decisions– Monitoring and restricting information flows– Managing and supporting trusted virtual

environments within which humans operate


Timothy J. Norman

• Open Virtual Organisations need– Models of Norms/Contracts/Policies that are

machine-readable– Mechanisms to efficiently identify norm

conflicts– Mechanisms to resolve conflicts, or to

coherently present solutions to users

Summary


Timothy J. Norman

Other Dimensions

processclosed open

Increased process

complexity


e-InstitutionsIncreased

system scale

OVOs

This is where we

need to go!

Timothy J. Norman

• ADEPT (1995-1998) – Nick Jennings• CONOISE/CONOISE-G (2000-2006)

– Alun Preece (now Cardiff) Nir Oren (now KCL) Nick Jennings Mike Luck

• NoA (2000-2003) – Martin Kollingbaum (now CMU)

• FOUND! – Wamberto Vasconcelos (Aberdeen)

• ITA (International Technology Alliance) (2006-2016) – Wamberto Vasconcelos, Derek Sleeman, Katia Sycara (CMU), Simon Parsons (CUNY)

• ALIVE (FP7) – from 2008

Acknowledgements


Documents

Timothy J. Norman Challenges for Effective Open Virtual Organisations Timothy J. Norman Department of Computing Science University of Aberdeen [email protected]