Argumentation-based negotiation Rahwan, Ramchurn, Jennings, McBurney, Parsons and Sonenberg, 2004 Presented by Jean-Paul Calbimonte
Negotiation? interactions Mutually acceptable agreement Scarce resources: time money services, anything Automate! Approaches Game-theoretic analysis Heuristic-based Argumentation-based
Classical negotiation Agents exchange proposals Limitations Computational constraints (analytical game-theoretic) Sub-optimal approximate outcomes (heuristic-based) Require extensive empirical evaluation (heuristic) No additional info exchanged Agents preferences need to be fully characterized Lack information to evaluate/compare proposals Fixed agents preferences proposals
Argumentation-based negotiation Exchange additional information Argument: piece of information Reasons for refusal Justification of a proposal Change object of negotiation Promise rewards Emit threats Components External elements Own elements
Non-ABN Elements Proposal Database Locution Interpretation Opponent/ Environment Model & Mental Attitudes Proposal Evaluation Generation Locution Generation Outgoing Locutions Incoming Locutions query / update query Proposal content propose / accept / reject
ABN elements Proposal Database Locution Interpretation Opponent/ Environment Model & Mental Attitudes Proposal Evaluation Generation Locution Generation Outgoing Locutions Incoming Locutions query / update query Proposal content propose / accept / reject Argument Interpretation Argument Generation Argument Selection Argument content
External elements Communication language Interaction between agents Locutions, utterances Domain language Concepts & meta information about the world Negotiation protocol Conventions, rules of dialogue: who is allowed to say what? Information stores Keep track of utterances, behavior Internal, external repositories accept reject propose reject(b,a,Price=$200 and Item=palmI30)
Communication & domain language Rich languages, clear semantics request(j,i, Do(i,α),Do(i,α) Do(j,β)) Express preferences Express plans, intentions Standardized domain languages Heterogeneous environments Semantic & Syntactic interoperability i j Do(i, α ) Do(i, α ) Do(j, β )
Negotiation protocol Interaction, dialogue rules Finite state machines Dialogue games Enforce fairness, rule consistency Termination, avoid infinite interactions Enusre guaranteed success Endless interactions? Conformance checking Utterance acceptable? Admission to negotiation
Information stores Keep track of past utterances Internal Centralized Commitment stores Promise to initiate, execute, maintain Defend claims Commitment rules Retract from commitments Allowed Forced
Elements of ABN agents Argument and proposal evaluation Incoming arguments, update mental state Argument and proposal generation Generate set of candidates Argument selection Choosing a candidate
Argument and proposal evaluation Objective considerations Argument acceptability, evaluation Subjective considerations Consider own preferences and motivations Combine objective + subjective Combine belief arguments and value arguments Provide unified framework for evaluation of goal, belief, plan, etc. Probabilistic evaluation
Argument and proposal generation Generate candidate arguments Increase, maximize utilities Rule-based generation Characterization of possible arguments Take into account other factors: Protocol Authority Expectations Utility Honesty
Argument selection Choose candidate argument ABN strategy Use strength order: common practice, appeal, self interest, promise,threat Use factors: trust-utility I need this but I don’t trust you Opponents information, Probabilistic models, uncertainty Learning techniques Consider behavior & mental model behind
Conlcusions ABN: increasing importance Enables rational dialogue Richer models of negotiation How agents use objective reasoning to reach subjective goals? Challenges Communicative rationality Social influence Trust Mediated negotiation Complexity