CarolSemPesquisa - LES - PUC-Rio

Sweetening Regulated Open Multi-Agent Systems with
a Formal Support for Agents to Reason About Laws
Semantic
Carolina Howard Felicíssimo
Key points of my paper to the AAMAS, 2006
… trying my qualification !
Outline
• Introduction
• An Independent-Domain Ontology for Regulations in Open MAS
(A Generic Normative Ontology)
• Case Study
– Ontology Extension
– Ontology Instantiation
– Implementation
• Related Works
• Conclusions
Carolina Howard Felicíssimo © LES/PUC-Rio
Introduction
•
Provide a Semantic Support for Agents to Reason About Laws in Regulated
Open MAS
What to do?
– Define an approach to specify open MAS’ laws
Define guidelines for the implementation of regulations in open MAS
– Choose the formalism(s) to be used
The more expressive the language, the harder the reasoning !!!
How to balance expressiveness and reasoning?
– Identify the trade-offs about the choice
Benefits:
– Automation of part of the laws’ definitions (Ex.: by inference of inherited laws)
– Check consistency among laws
– Agents can be treat as black-boxes when reasoning about laws
Carolina Howard Felicíssimo © LES/PUC-Rio
An Independent-Domain Ontology for Regulations in Open MAS
Top-Down Modeling - Different (Interaction) Laws’ Levels:
•
•
•
•
Environment’s Laws:
–
Those that are applied to all entities from the environment
–
Are independent of existing organizations, roles being played and interactions
Organization’s Laws:
–
Those that are applied to all entities from the organization
–
Are independent of roles being played and interactions
Role’s Laws:
–
Those that are applied to all entities playing the role
–
Are independent of interactions
Interaction’s Laws:
–
Those that are applied to all entities involved in the interaction
Exceptions: union of disjoint sets: the “perfect world” plus the exceptions
–
Ex.: Elephants: union of the gray elephants + albino elephants
Carolina Howard Felicíssimo © LES/PUC-Rio
An Independent-Domain Ontology for Regulations in Open MAS
Carolina Howard Felicíssimo © LES/PUC-Rio
An Independent-Domain Ontology for Regulations in Open MAS
South America
HP
Dell
HP Brazil
Dell Brazil
Supplier
Supplier
Manufacturer
Distributor
Manufacturer
Customer
Retailer
Dell Uruguai
HP Argentina
Supplier
Supplier
Manufacturer
Manufacturer
Distributor
Retailer
Carolina Howard Felicíssimo © LES/PUC-Rio
Case Study
• TAC Environment’s Laws:
– Trading designed to benefit some other agent at the expense of the
trader's own utility
– Denial-of-service attacks. Agents may not employ API operations for
the purpose of occupying or loading the game servers
– Organizations cannot be composed by more than six manufactures
– The main organization TAC has a fixed life time of 220 days
– To get in, agents have to connect to a game server
Carolina Howard Felicíssimo © LES/PUC-Rio
Case Study
• Organizations’ Laws:
– Human intervention is not allowed
– Is allowed to carry a negative balance in agents’ bank account
– Independent of all roles from the organization, sales can just be made
between the roles:
– Suppliers and manufactures or
– Manufactures and customers
TAC Supply Chain: (model: direct sales to customer)
Supplier -> Manufacturer -> Customer
Carolina Howard Felicíssimo © LES/PUC-Rio
Case Study
• Roles’ Laws:
– Supplier
• Only complete orders are shipped
Exception: in the last day of the game, partial orders will be shipped
• Orders are not shipped before their due dates
• Every order has a down payment of 10%
• Suppliers operate in make-to-order basis
• If multiple days of production are required to satisfy an order, inventory is
carried over. Inventory carrying costs are assumed to be zero
Carolina Howard Felicíssimo © LES/PUC-Rio
Case Study
• Roles’ Laws:
– Manufacture
• Every shipped orders must be paid
• Every manufacture has an assembly cell that cannot process more than 2000
cycles/day
• Every manufacture can only produce if all the required components are
available inventory
• Every manufacture can only ship products if they are available in the
inventory
• Every manufacture with inventory of finished goods and components will be
charged a daily storage cost of 25% - 50% of the nominal price of
components
Carolina Howard Felicíssimo © LES/PUC-Rio
Case Study
• Roles’ Laws:
– Customer
• Every RFQs has to have:
– Product Type
– Quantity Due Date
– Reserve Price
– Penalty amount
– Maximum price per unit that the customer is willing to pay
• Every shipped orders must be paid
• The valid bid with the lowest price has to be chosen
• A randomly choice has to be made when valid bids tied
Carolina Howard Felicíssimo © LES/PUC-Rio
Case Study
• Interactions’ Laws:
– Laws between Manufactures and Suppliers:
• Suppliers have to answer all manufactures’ RFQs
• Manufactures can only send 5 RFQs/day to each supplier for each of the
products offered
• Suppliers have to ignore subsequent orders if more than one type (partial
offer or earliest complete offer) was done in a day
– Laws between Manufactures:
• A manufacture cannot sell to another manufacture
Carolina Howard Felicíssimo © LES/PUC-Rio
Case Study
• Interactions’ Laws:
– Laws between Manufactures and Customers:
• Bid of manufactures must address the entire quantity specified in the RFQ
from customers
• Bid of manufactures must be delivered on the due date specified in the RFQ
• Bid prices of manufactures must be below or equal to the reserve price
specified by the customer in the RFQ
• If a RFQ was sent by a customer to a manufacture, all others manufactures
have to receive RFQs too
– Laws between Suppliers and Customers cannot be addressed because
interactions are not allowed (organizations’ laws)
Carolina Howard Felicíssimo © LES/PUC-Rio
Carolina Howard Felicíssimo © LES/PUC-Rio
Legend:
Environmets’ Laws
Organizations’ Laws
Roles’ Laws
Interactions’ Laws
Carolina Howard Felicíssimo © LES/PUC-Rio
Legend:
Environmets’ Laws
Organizations’ Laws
Roles’ Laws
Interactions’ Laws
Carolina Howard Felicíssimo © LES/PUC-Rio
Case Study - Implementation
• Tac Ontology (tac.owl)
• Tac Rules (tac.rules)
tac.owl
tac.owl
Java Programming
+
tac.rules
Java Programming + INFERENCE
Results
Results
– Automation of part of the laws’ definitions
– Check consistency among laws
Carolina Howard Felicíssimo © LES/PUC-Rio
Related Work
• Enforcement is done based on Interactions’ Laws
– XMLaw
– LGI
– Electronic Institutions
• How to reason with incomplete information and conflits?
– Description Defesiable Logic
• OMINI
...
Carolina Howard Felicíssimo © LES/PUC-Rio
Conclusion
• An approach to regulate open MAS based on ontologies was
defined
• A formal support for agents to base their behavior according to
norms and to reason about action selection was provided
• The independent-domain ontology can be extended and it needs to
be instantiated for specific domains
• Environments’ laws, Organizations’ laws, Roles’ laws and
Interactions’ laws can be specified
• As a future work, the formalism(s) to be used will be chosen
and the trade-offs about the choice will be identified
Carolina Howard Felicíssimo © LES/PUC-Rio
Questions ?