Multiagent Systems: Reaching Agreements and Negotiation Mechanisms, Study notes of Electrical and Electronics Engineering

Download Multiagent Systems: Reaching Agreements and Negotiation Mechanisms and more Study notes Electrical and Electronics Engineering in PDF only on Docsity! LE C T U R E 7: R E A C H IN G A G R E E M E N T S A n Introduction to M ultiagentS ystem s h t t p : / / w w w . c s c . l i v . a c . u k / ˜ m j w / p u b s / i m a s / Lecture 7 A n Introduction to M ultiagentS ystem s 1 R eaching A greem ents H ow do agents reaching agreem ents w hen they are self interested? In an extrem e case (zero sum encounter) no agreem entis possible — butin m ostscenarios,there is potentialfor m utually beneficialagreem enton m atters ofcom m on interest. T he capabilities ofnegotiation and argum entation are centralto the ability ofan agentto reach such agreem ents. h t t p : / / w w w . c s c . l i v . a c . u k / ˜ m j w / p u b s / i m a s / 1 Lecture 7 A n Introduction to M ultiagentS ystem s 2 A uctions A n auction takes place betw een an agentknow n as the auctioneer and a collection ofagents know n as the bidders. T he goalofthe auction is for the auctioneer to allocate the good to one ofthe bidders. In m ostsettings the auctioneer desires to m axim ise the price; bidders desire to m inim ise price. h t t p : / / w w w . c s c . l i v . a c . u k / ˜ m j w / p u b s / i m a s / 4 Lecture 7 A n Introduction to M ultiagentS ystem s A uction P aram eters G oods can have private value public/com m on value; correlated value W inner determ ination m ay be firstprice; second price. B ids m ay be open cry sealed bid. B idding m ay be: one shot; ascending descending. h t t p : / / w w w . c s c . l i v . a c . u k / ˜ m j w / p u b s / i m a s / 5 Lecture 7 A n Introduction to M ultiagentS ystem s E nglish A uctions M ostcom m only know n type ofauction: – first-price, – open cry, – ascending. D om inantstrategy is for agentto successively bid a sm all am ountm ore than the currenthighestbid untilitreaches their valuation,then w ithdraw . S usceptible to: – w inners curse; – shills. h t t p : / / w w w . c s c . l i v . a c . u k / ˜ m j w / p u b s / i m a s / 6 Lecture 7 A n Introduction to M ultiagentS ystem s V ickrey A uctions V ickrey auctions are: – second-price; – sealed-bid. G ood is aw arded to the agentthatm ade the highestbid;atthe price ofthe second highestbid. B idding to your true valuation is dom inantstrategy in V ickrey auctions. V ickrey auctions susceptible to antisocialbehavior. h t t p : / / w w w . c s c . l i v . a c . u k / ˜ m j w / p u b s / i m a s / 9 Lecture 7 A n Introduction to M ultiagentS ystem s 3 N egotiation A uctions are only concerned w ith the allocation ofgoods: richer techniques for reaching agreem entsare required. N egotiation is the process ofreaching agreem ents on m atters of com m on interest. A ny negotiation setting w illhave four com ponents: – A negotiation set: possible proposals thatagents can m ake. – A protocol. – S trategies,one for each agent,w hich are private. – A rule thatdeterm ines w hen a dealhas been struck and w hat the agreem entdealis. N egotiation usually proceeds in a series ofrounds,w ith every agentm aking a proposalatevery round. h t t p : / / w w w . c s c . l i v . a c . u k / ˜ m j w / p u b s / i m a s / 10 Lecture 7 A n Introduction to M ultiagentS ystem s 3.1 N egotiation in Task-O riented D om ains Im agine thatyou have three children,each ofw hom needs to be delivered to a differentschool each m orning. Your neighbour has four children,and also needs to take them to school. D elivery ofeach child can be m odelled as an indivisible task. You and your neighbour can discuss the situation,and com e to an agreem entthatitis better for both ofyou (for exam ple,by carrying the other’s child to a shared destination,saving him the trip). T here is no concern aboutbeing able to achieve your task by yourself. T he w orstthatcan happen is thatyou and your neighbour w on’t com e to an agreem entaboutsetting up a car pool,in w hich case you are no w orse offthan if you w ere alone. You can only benefit(or do no w orse) from your neighbour’s tasks. A ssum e, though,thatone ofm y children and one ofm y neigbours’s children both go to the sam e school (thatis,the costofcarrying outthese tw o deliveries,or tw o tasks,is the sam e as the costof carrying outone ofthem ). Itobviously m akes sense for both children to be taken together,and only m y neighbour or Iw illneed to m ake the trip to carry outboth tasks. h t t p : / / w w w . c s c . l i v . a c . u k / ˜ m j w / p u b s / i m a s / 11 Lecture 7 A n Introduction to M ultiagentS ystem s T he N egotiation S et T he setofdeals over w hich agents negotiate are those thatare: – individualrational – pareto efficient. h t t p : / / w w w . c s c . l i v . a c . u k / ˜ m j w / p u b s / i m a s / 14 Lecture 7 A n Introduction to M ultiagentS ystem s T he N egotiation S etIllustrated utility for agent j utility for agent i utility of conflict deal for j utility of conflict deal for i deals on this line Pareto optim al, hence in the negotiation set this circle delim its the possible deals space of all conflict deal A B C D E from B to C are h t t p : / / w w w . c s c . l i v . a c . u k / ˜ m j w / p u b s / i m a s / 15 Lecture 7 A n Introduction to M ultiagentS ystem s T he M onotonic C oncession P rotocol R ules ofthis protocolare as follow s... N egotiation proceeds in rounds. O n round 1,agents sim ultaneously propose a dealfrom the negotiation set. A greem entis reached ifone agentfinds thatthe dealproposed by the other is atleastas good or better than its proposal. Ifno agreem entis reached,then negotiation proceeds to another round ofsim ultaneous proposals. In round u   ,no agentis allow ed to m ake a proposalthatis less preferred by the other agentthan the dealitproposed attim e u. Ifneither agentm akes a concession in som e round u   ,then negotiation term inates,w ith the conflictdeal. h t t p : / / w w w . c s c . l i v . a c . u k / ˜ m j w / p u b s / i m a s / 16 Lecture 7 A n Introduction to M ultiagentS ystem s N ash E quilibrium A gain... T he Z euthen strategy is in N ash equilibrium : under the assum ption thatone agentis using the strategy the other can do no better than use ithim self... T his is ofparticular interestto the designer ofautom ated agents. Itdoes aw ay w ith any need for secrecy on the partof the program m er. A n agent’s strategy can be publicly know n, and no other agentdesigner can exploitthe inform ation by choosing a differentstrategy. In fact,itis desirable thatthe strategy be know n,to avoid inadvertentconflicts. h t t p : / / w w w . c s c . l i v . a c . u k / ˜ m j w / p u b s / i m a s / 19 Lecture 7 A n Introduction to M ultiagentS ystem s D eception in TO D s D eception can benefitagents in tw o w ays: P hantom and D ecoy tasks. P retending thatyou have been allocated tasks you have not. H idden tasks. P retending notto have been allocated tasks thatyou have been. h t t p : / / w w w . c s c . l i v . a c . u k / ˜ m j w / p u b s / i m a s / 20 Lecture 7 A n Introduction to M ultiagentS ystem s 4 A rgum entation A rgum entation is the process ofattem pting to convince others of som ething. G ilbert(1994) identified 4 m odes ofargum ent: 1. Logicalm ode. “Ifyou acceptthat A and that A im plies B ,then you m ust acceptthat B ”. 2. E m otionalm ode. “H ow w ould you feelifithappened to you?” 3. V isceralm ode. “C retin!” 4. K isceralm ode. “T his is againstC hristian teaching!” h t t p : / / w w w . c s c . l i v . a c . u k / ˜ m j w / p u b s / i m a s / 21 Lecture 7 A n Introduction to M ultiagentS ystem s A bstractA rgum entation C oncerned w ith the overallstructure ofthe argum ent(rather than internals ofargum ents). W rite x y – “argum ent x attacks argum ent y”; – “x is a counterexam ple of y;or – “x is an attacker of y”. w here w e are notactually concerned as to w hat x, y are. A n abstractargum entsystem is a collection or argum ents together w ith a relation “ ” saying w hatattacks w hat. A n argum entis outifithas an undefeated attacker,and in ifall its attackers are defeated. h t t p : / / w w w . c s c . l i v . a c . u k / ˜ m j w / p u b s / i m a s / 24 Lecture 7 A n Introduction to M ultiagentS ystem s A n E xam ple A bstractA rgum entS ystem a b eh f n c d g i j p q m k l h t t p : / / w w w . c s c . l i v . a c . u k / ˜ m j w / p u b s / i m a s / 25