Reaching Agreements � How do agents reaching agreements LECTURE 7: when they are self interested? � In an extreme case (zero sum Reaching Agreements encounter) no agreement is possible — but in most scenarios, there is potential for mutually beneficial agreement on matters of common interest An Introduction to MultiAgent Systems http://www.csc.liv.ac.uk/~mjw/pubs/imas � The capabilities of negotiation and argumentation are central to the ability of an agent to reach such agreements 7-1 7-2 Mechanisms, Protocols, and Strategies Mechanism Design � Negotiation is governed by a particular � Desirable properties of mechanisms: mechanism , or protocol � Convergence/guaranteed success � The mechanism defines the “rules of � Maximizing social welfare encounter” between agents � Pareto efficiency � Mechanism design is designing mechanisms � Individual rationality so that they have certain desirable properties � Stability � Given a particular protocol, how can a � Simplicity particular strategy be designed that individual � Distribution agents can use? 7-3 7-4 Auction Parameters Auctions � Goods can have � private value � publi c/ common value � An auction takes place between an agent � correlated value known as the auctioneer and a collection of � Winner determination may be agents known as the bidders � first price � second price � The goal of the auction is for the auctioneer � Bids may be to allocate the good to one of the bidders � open cry � In most settings the auctioneer desires to � sealed bid maximize the price; bidders desire to � Bidding may be minimize price � one shot � ascending � descending 7-5 7-6 1
English Auctions Dutch Auctions � Most commonly known type of auction: � Dutch auctions are examples of open-cry � first price descending auctions: � open cry � auctioneer starts by offering good at artificially � ascending high value � Dominant strategy is for agent to � auctioneer lowers offer price until some agent successively bid a small amount more than makes a bid equal to the current offer price the current highest bid until it reaches their � the good is then allocated to the agent that valuation, then withdraw made the offer � Susceptible to: � winner’s curse � shills 7-7 7-8 First-Price Sealed-Bid Auctions Vickrey Auctions � Vickrey auctions are: � second-price � First-price sealed-bid auctions are one-shot � sealed-bid auctions : � Good is awarded to the agent that made � there is a single round the highest bid; at the price of the second � bidders submit a sealed bid for the good highest bid � good is allocated to agent that made highest bid � Bidding to your true valuation is dominant � winner pays price of highest bid strategy in Vickrey auctions � Best strategy is to bid less than true valuation � Vickrey auctions susceptible to antisocial behavior 7-9 7-10 Lies and Collusion Negotiation � Auctions are only concerned with the allocation of goods: � The various auction protocols are susceptible richer techniques for reaching agreements are required to lying on the part of the auctioneer, and � Negotiation is the process of reaching agreements on matters collusion among bidders, to varying degrees of common interest � All four auctions (English, Dutch, First-Price � Any negotiation setting will have four components: Sealed Bid, Vickrey) can be manipulated by � A negotiation set: possible proposals that agents can make bidder collusion � A protocol � Strategies, one for each agent, which are private � A dishonest auctioneer can exploit the Vickrey � A rule that determines when a deal has been struck and auction by lying about the 2 nd -highest bid what the agreement deal is � Shills can be introduced to inflate bidding � Negotiation usually proceeds in a series of rounds, with every prices in English auctions agent making a proposal at every round 7-11 7-12 2
Negotiation in Task-Oriented Domains Machines Controlling and Sharing Resources Imagine that you have three children, each of whom needs to be delivered to a different school each morning. Your neighbor has four children, and also needs to take them to school. Delivery of each child can be modeled as an indivisible task. You and your � Electrical grids (load balancing) neighbor can discuss the situation, and come to an agreement that it is better for both of you (for example, by carrying the � Telecommunications networks (routing) other’s child to a shared destination, saving him the trip). There is no concern about being able to achieve your task by yourself. � PDA’s (schedulers) The worst that can happen is that you and your neighbor won’t come to an agreement about setting up a car pool, in which case you are no worse off than if you were alone. You can only benefit � Shared databases (intelligent access) (or do no worse) from your neighbor’s tasks. Assume, though, that one of my children and one of my neighbors’ children both � Traffic control (coordination) go to the same school (that is, the cost of carrying out these two deliveries, or two tasks, is the same as the cost of carrying out one of them). It obviously makes sense for both children to be taken together, and only my neighbor or I will need to make the trip to carry out both tasks. --- Rules of Encounter , Rosenschein and Zlotkin, 1994 7-13 7-14 The Aim of the Research Heterogeneous, Self-motivated Agents � Social engineering for communities of The systems: machines � are not centrally designed � The creation of interaction environments that � do not have a notion of global utility foster certain kinds of social behavior � are dynamic (e.g., new types of agents) � will not act “benevolently” unless it is in their interest to do so The exploitation of game theory tools for high-level protocol design 7-15 7-16 Attributes of Standards Broad Working Assumption � Efficient : Pareto Optimal � Designers (from different companies, � Stable : No incentive to deviate countries, etc.) come together to agree on � Simple : Low computational and standards for how their automated agents communication cost will interact (in a given domain) � Distributed : No central decision-maker � Discuss various possibilities and their � Symmetric : Agents play equivalent roles tradeoffs, and agree on protocols, strategies, and social laws to be Designing protocols for specific classes of implemented in their machines domains that satisfy some or all of these attributes 7-17 7-18 3
Phone Call Competition Example Distributed Artificial Intelligence (DAI) � Customer wishes to place long-distance call � Distributed Problem Solving (DPS) � Carriers simultaneously bid, sending proposed prices � Centrally designed systems, built-in � Phone automatically chooses the carrier cooperation, have global problem to solve (dynamically) � Multi-Agent Systems (MAS) AT&T Sprint MCI � Group of utility-maximizing heterogeneous agents co-existing in same environment, $0.20 $0.20 $0.23 $0.23 possibly competitive $0.18 $0.18 7-19 7-20 Best Bid Wins Attributes of the Mechanism � Phone chooses carrier with lowest bid � Distributed Carriers have an � Carrier gets amount that it bid � Symmetric incentive to invest effort in � Stable strategic � Simple behavior MCI AT&T Sprint � Efficient $0.20 $0.20 AT&T $0.23 $0.23 $0.18 $0.18 MCI Sprint $0.20 $0.20 “Maybe I can $0.23 bid as high as $0.23 $0.18 $0.18 $0.21...” 7-21 7-22 Best Bid Wins, Gets Second Price Attributes of the Vickrey Mechanism (Vickrey Auction) � Distributed Carriers have no � Phone chooses carrier with lowest bid incentive to � Symmetric � Carrier gets amount of second-best price invest effort in � Stable strategic � Simple behavior AT&T Sprint MCI � Efficient $0.20 $0.20 $0.23 $0.23 AT&T $0.18 $0.18 MCI Sprint $0.20 $0.20 “I have no $0.23 $0.23 $0.18 $0.18 reason to overbid...” 7-23 7-24 4
Domain Theory Postmen Domain � Task Oriented Domains Post Office Post Office 1 2 Agents have tasks to achieve � Task redistribution � TOD TOD a � State Oriented Domains Goals specify acceptable final states � � � c b Side effects � Joint plan and schedules � � Worth Oriented Domains � � f Function rating states’ acceptability � � d e Joint plan, schedules, and goal relaxation � 7-25 7-26 Fax Domain Database Domain “All female Common Database employees 1 2 Common Database making over faxes to TOD TOD $50,000 a TOD TOD a send year.” “All female employees b c with more Cost is than three 2 children.” only to establish f 1 connection e d 7-27 7-28 Slotted Blocks World The Multi-Agent Tileworld SOD SOD WOD WOD hole agents tile 3 B 1 1 2 2 3 A 2 2 5 5 2 1 obstacle 2 34 2 1 1 2 2 3 3 7-29 7-30 5
Recommend
More recommend
