Auctions and Multi-agent Systems 발표자 : 96419-047 오 연 희
Purpose Discuss how a virtual institution can be derived from the Llotja(a fish market) Show how this institution can be applied to define other institutions where the participants can be software agents
Auction Auctions are price-fixing mechanisms : bidding convention Auction houses is institutions : also, other equally relevant conventions are used. (about registration of participants and goods, guarantees, payment, commissions, starting prices, etc...)
Three types of conventions Ontologic and Communicational Conventions Social Conventions Individual Behaviur Rules
But, we are concerned with VI!! The Difference of virtual institution from traditional auction house - Virtual location - software agent Consequently, it could be understood as a multi-agent system model, or more specifically, as its computational realization in the Internet.
Feature of Virtual Institution Message Exchange strong dialogical stance Dialogical system
Llotja (a fish market of Blances) : traditional auction house Five scenes(location, staff) 1. Buyers’ register(AR, ba) 2. Goods’ register(RR, sa) 3. Goods’ show and auction(AH, auct) 4. Credits and goods delivery(DR, bm) 5. Sellers’ settlements(BO, sm)
Scenes in the Blanes Llotja
Illocutions used in an upward bidding round in the Fishmarket
Fishmarket bidding rounds
The Difference of Fishmarket from the Blanes auction The Participants can be computer programs executed in a remote location collisions are managed in the Fishmarket - by a random selection of a winner
Institution FM = <DF, PS, BR> Dialogical Framework Performative Structure : articulation of scenes Individual Behavior Rules FM = <DF, PS, BR>
Performative Structure
Commitments established in an auction house Exchange of auctioned goods their features the exchange conditions
Information structure for Commitments in the Fishmarket The auction’s catalogue : storage of information related to the goods auctioned each seller’s account : identification, the income and commissions each buyer’s account : credit and the purchases mode
Individual Behavior Rules(purpose) The need of making these conventions explicit and more intelligible for users or external observers To facilitate the definition and update of the institution’s conventions
Individual Behavior Rules(example)
Convention Variant of the Fishmarket 1. Closed bidding auction - operator(selected and creditstatus) is added and extended Vickrey bidding : upward and closed bidding - selected operator is changed
Convention Variant of the Fishmarket 2. The same individual rules, the same illocutions and the same operators could be used for any of its variants Use of sufficiently general operators - parametrized
Compoments of Dialogical Framework for Fishmarket Agents a set of identifiers Rol = {boss, auct, sa, sm, ba, bm} {s, b} SR = {<boss, x> : x Rol \ boss} Loc = {AH, RR, AR, DR, BO, M} L the “object language” CL (a communication language) ML (metalanguage to define behaviour rules) T a time model
Models and Implementation FM: auction house A: auction G: auctioned goods A: participating agents E0…Ew: the incidents evolve A = <A, G, <E0..Ew>>
Institutions Agent-mediated institution Agent-based instutions - “Cohabited Mixed Reality Information Space” project Agent-based instutions - “medical protocols” SMASH project
Closing Remarks #1 Artificial Intelligence A interdisciplinary problem other fields of AI
Closing Remarks #2 multi-agent systems Constitute a non-trivial problem a domain in which competitive evaluation is possible auctions are a conveniently scalable problem
Closing Remarks #3 Electronic Commerce Electronic exchange through the Internet The opportunity of developing agents