談判過程中的混合智能.pdf

《談判過程中的混合智能.pdf》由會員分享，可在線閱讀，更多相關《談判過程中的混合智能.pdf（20頁珍藏版）》請在三個皮匠報告上搜索。

1、Hybrid Intelligence in Negotiation ProcessesCATHOLIJN M.JONKEROur Future with Artificial IntelligencePotential&benefits of AIEmpower autonomyExpand experienceNew activitiesStrenghten democraciesMisuse&side-effects of AIReduce autonomyReplace experienceRedundantEndanger democraciesArtificial Intellig

2、ence Hybrid IntelligencePotential&benefits of AIEmpower autonomyExpand experienceNew activitiesStrenghten democraciesMisuse&side-effects of AIReduce autonomyReplace experienceRedundantEndanger democraciesAlign AI with human&social values&create Hybrid IntelligenceMeaningful human control&Hybrid Inte

3、lligenceNegotiation(Support)SystemsWhy negotiation is difficult for humans Leaving money on the table(sub-optimal outcomes for both)Bounded rationality(outcome space is too big)Slowly discover what they want while negotiating(preference elicitation)Satisficing(rationalising your intuitions)Position

4、bargaining vs understanding underlying concerns Self-reinforcement(proud of the outcome)And the usual Laziness(preparation)Emotions(negative effect on rationality)combining the strengths of machine and human HumanNSSUnderstands natural languageGood at calculating(bids)Recognizes emotionsKeeps overvi

5、ewHas emotions Stays calmGeneral knowledgeNegotiation knowledgeThe Pocket Negotiator Project:development of a next generation negotiation support systemSWOT analysis(2007)PreparationHumanSymb.AIPrivate preparation:Domain modellingUser modellingOpponent modellingAlternativesBidding:Strategy determina

6、tionBid evaluationNext bid determinationBidding analysisClosing:ContractRelationshipsReflectionJoint exploration:Domain modellingUser preferencesOpponent modellingRelationshipsPhases of Negotiation:Potential for AI(2007)Computer scienceIndustrial EngineeringBusiness2121Pocket Negotiator increases ou

7、tcome utility(p 0.01)Some people manage quite well on their ownReyhan Aydoan and Catholijn M.Jonker,“Bidding Support by the Pocket Negotiator Improves Negotiation Outcomes”,in:Recent Advances in Agent-based Negotiation:Applications and Competition Challenges,Springer,2023.The average number of Paret

8、o Optimal bids over all groups split out according to support-and starting conditions.Average number of bids made per negotiation:10All groupsNS-SS-NSS-Pareto5.56.2NS-Pareto2.22.6Pareto OptimalitysignificantNot significantalso per groupWhat support was used?GroupsBidsPareto OptimalityPareto ClicksSu

9、ggestionsComputer science students 1078.51Design students9.63.64.91.1Business students11.16.46.73.8All10.35.96.7*2.2*:In the Support-condition:Significant correlation between the number of bids on the Pareto Frontier and the type of support used by the participants:Spearmans Rho=0.61,p 0.01Self-dire

10、ctedThe autonomy diagonal of negotiationNegotiation analysis Negotiation heuristicsGame theory Trading botsSelf-sufficientResearch agenda1.Domain knowledge and preference elicitation2.Long-term perspective3.User trust and adoption4.Learning strategiesSWOT(2017)&Research Agenda2007HumanSymb.AIPrepara

11、tionHumanSymb.AIPrivate preparation:Domain modellingUser modellingOpponent modellingAlternativesBidding:Strategy determinationBid evaluationNext bid determinationBidding analysisClosing:ContractRelationshipsReflectionJoint exploration:Domain modellingUser preferencesOpponent modellingRelationshipsPr

12、eparationHumanSymb.AI2007ExplorationHumanSymb.AI20072007201720172017PreparationHumanSymb.AI2017Phases of Negotiation:Potential for AI(2017)From the 2017 Research agenda1.Domain knowledge and preference elicitation2.Long-term perspective3.User trust and adoption4.Learning Strategies SWOT(2024)&Resear

13、ch AgendaWhat opportunities offer LLMs?Negotiation SupportAutomated Negotiating AgentsPicture fromSWOT(2024)&Research AgendaThe NegotiatorLlaMaFrom the 2017 Research agenda1.Domain knowledge and preference elicitation2.Long-term perspective3.User trust and adoption4.Learning Strategies By ChatGPTIll

14、 help you advocate for yourself and get better outcomes.Become a great negotiator.2007HumanSymb.AIPrivate preparation:Domain modellingUser modellingOpponent modellingAlternativesBidding:Strategy determinationBid evaluationNext bid determinationBidding analysisClosing:ContractRelationshipsReflectionJ

15、oint exploration:Domain modellingUser preferencesOpponent modellingRelationshipsPreparationHumanSymb.AI2007ExplorationHumanSymb.AI20072007201720242017PreparationHumanSymb.AI20172024202420242017Phases of Negotiation:Potential for AI(2024)HumanSymb.AILLMsSWOT 2024 AI potential for negotiation020406080100Situation awarenessHuman factorsNegotiation overiewExplainableNegotiation strategiesTrust and AdoptionHumanLLMRLSymb.AIThe way forwardLLMs instrumental for usabilityReinforcement Learning for soliditySymbolic AI(explainability,causal learning)Hybrid Intelligence is the way forward