Argumentation Henry Prakken SIKS Basic Course Learning and Reasoning May 26th, 2009 Why do agents need argumentation? For their internal reasoning Reasoning about beliefs, goals, intentions etc often is defeasible For their interaction with other agents Information exchange, negotiation, collaboration, … Overview Inference (logic) Abstract argumentation Rule-based argumentation Dialogue Part 1: Inference We should lower taxes Lower taxes increase productivity Increased productivity is good We should lower taxes Lower taxes increase productivity Increased productivity is good We should not lower taxes Lower taxes increase inequality Increased inequality is bad We should lower taxes Lower taxes increase productivity We should not lower taxes Increased productivity is good Lower taxes do not increase productivity USA lowered taxes but productivity decreased Lower taxes increase inequality Increased inequality is bad We should lower taxes Lower taxes increase productivity Prof. P says that … We should not lower taxes Increased productivity is good Lower taxes do not increase productivity USA lowered taxes but productivity decreased Lower taxes increase inequality Increased inequality is bad We should lower taxes Lower taxes increase productivity Prof. P says that … People with political ambitions are not objective We should not lower taxes Increased productivity is good Prof. P is not objective Prof. P has political ambitions Lower taxes do not increase productivity USA lowered taxes but productivity decreased Lower taxes increase inequality Increased inequality is bad We should lower taxes Lower taxes increase productivity Prof. P says that … People with political ambitions are not objective We should not lower taxes Increased productivity is good Prof. P is not objective Prof. P has political ambitions Lower taxes do not increase productivity USA lowered taxes but productivity decreased Lower taxes increase inequality Increased inequality is bad We should lower taxes Lower taxes increase productivity Prof. P says that … People with political ambitions are not objective We should not lower taxes Increased productivity is good Prof. P is not objective Prof. P has political ambitions Lower taxes increase inequality Increased inequality is good Lower taxes do not increase productivity USA lowered taxes but productivity decreased Increased inequality is bad Increased inequality stimulates competition Competition is good Sources of conflict Default generalisations Conflicting information sources Alternative explanations Conflicting goals, interests Conflicting normative, moral opinions … Application areas Medical diagnosis and treatment Legal reasoning Interpretation Evidence / crime investigation Intelligence Decision making Policy design … We should lower taxes Lower taxes increase productivity Prof. P says that … People with political ambitions are not objective We should not lower taxes Increased productivity is good Prof. P is not objective Prof. P has political ambitions Lower taxes increase inequality Increased inequality is good Lower taxes do not increase productivity USA lowered taxes but productivity decreased Increased inequality is bad Increased inequality stimulates competition Competition is good A C B D E Status of arguments: abstract semantics (Dung 1995) INPUT: a pair Args,Defeat OUTPUT: An assignment of the status ‘in’ or ‘out’ to all members of Args So: semantics specifies conditions for labeling the ‘argument graph’. Should capture reinstatement: A B C Possible labeling conditions Every argument is either ‘in’ or ‘out’. 1. An argument is ‘in’ if all arguments defeating it are ‘out’. 2. An argument is ‘out’ if it is defeated by an argument that is ‘in’. Works fine with: A B But not with: A B C Two solutions Change conditions so that always a unique status assignment results A B A A B Use multiple status assignments: A C B B and C A B Unique status assignments Grounded semantics (Dung 1995): S0: the empty set Si+1: Si + all arguments defended by Si ... (S defends A if all defeaters of A are defeated by a member of S) A C B D E Is B, D or E defended by S1? Is B or E defended by S2? A problem(?) with grounded semantics We have: A B We want(?): A B C C D D A problem(?) with grounded semantics A A = Frederic Michaud is French since he has a French name B = Frederic Michaud is Dutch since he is a marathon skater C = F.M. likes the EU since he is European (assuming he is not Dutch or French) D = F.M. does not like the EU since he looks like a person who does not like the EU B C D A problem(?) with grounded semantics E A A = Frederic Michaud is French since Alice says so B = Frederic Michaud is Dutch since Bob says so C = F.M. likes the EU since he is European (assuming he is not Dutch or French) D = F.M. does not like the EU since he looks like a person who does not like the EU E = Alice and Bob are unreliable since they contradict each other B C D Multiple labellings A B A B C C D D Status assignments (1) Given Args,Defeat: A status assignment is a partition of Args into sets In and Out such that: 1. An argument is in In if all arguments defeating it are in Out. 2. An argument is in Out if it is defeated by an argument that is in In. A B C Status assignments (1) Given Args,Defeat: A status assignment is a partition of Args into sets In and Out such that: 1. An argument is in In if all arguments defeating it are in Out. 2. An argument is in Out if it is defeated by an argument that is in In. A B C Status assignments (1) Given Args,Defeat: A status assignment is a partition of Args into sets In and Out such that: 1. An argument is in In if all arguments defeating it are in Out. 2. An argument is in Out if it is defeated by an argument that is in In. A B C Status assignments (1) Given Args,Defeat: A status assignment is a partition of Args into sets In and Out such that: 1. An argument is in In if all arguments defeating it are in Out. 2. An argument is in Out if it is defeated by an argument that is in In. A B C Status assignments (1) Given Args,Defeat: A status assignment is a partition of Args into sets In and Out such that: 1. An argument is in In if all arguments defeating it are in Out. 2. An argument is in Out if it is defeated by an argument that is in In. A B C Status assignments (2) Given Args,Defeat: A status assignment is a partition of Args into sets In, Out and Undecided such that: 1. An argument is in In if all arguments defeating it are in Out. 2. An argument is in Out if it is defeated by an argument that is in In. A status assignment is stable if Undecided = . A status assignment is preferred if Undecided is -minimal. In is a stable extension In is a preferred extension A status assignment is grounded if Undecided is -maximal. In is the grounded extension Dung’s original definitions Given Args,Defeat, S Args, A Args: S is conflict-free if no member of S defeats a member of S S defends A if all defeaters of A are defeated by a member of S S is admissible if it is conflict-free and defends all its members S is a preferred extension if it is -maximally admissible S is a stable extension if it is conflict-free and defeats all arguments outside it S is the grounded extension if S is the -smallest set such that A S iff S defends A. S defends A if all defeaters of A are defeated by a member of S S is admissible if it is conflict-free and defends all its members A C B D Admissible? E S defends A if all defeaters of A are defeated by a member of S S is admissible if it is conflict-free and defends all its members A C B D Admissible? E S defends A if all defeaters of A are defeated by a member of S S is admissible if it is conflict-free and defends all its members A C B D Admissible? E S defends A if all defeaters of A are defeated by a member of S S is admissible if it is conflict-free and defends all its members A C B D Admissible? E S defends A if all defeaters of A are defeated by a member of S S is admissible if it is conflict-free and defends all its members A C B D Preferred? E S is preferred if it is maximally admissible S defends A if all defeaters of A are defeated by a member of S S is admissible if it is conflict-free and defends all its members A C B D Preferred? E S is preferred if it is maximally admissible S defends A if all defeaters of A are defeated by a member of S S is admissible if it is conflict-free and defends all its members A C B D Preferred? E S is preferred if it is maximally admissible S defends A if all defeaters of A are defeated by a member of S S is admissible if it is conflict-free and defends all its members A C B D Grounded? E S is groundeded if it is the smallest set s.t. A S iff S defends A S defends A if all defeaters of A are defeated by a member of S S is admissible if it is conflict-free and defends all its members A C B D Grounded? E S is groundeded if it is the smallest set s.t. A S iff S defends A Properties The grounded extension is unique Every stable extension is preferred (but not v.v.) There exists at least one preferred extension The grounded extension is a subset of all preferred and stable extensions … The ‘ultimate’ status of arguments (and conclusions) With grounded semantics: With preferred semantics: A is justified if A g.e. A is overruled if A g.e. and A is defeated by g.e. A is defensible otherwise A is justified if A p.e for all p.e. A is defensible if A p.e. for some but not all p.e. A is overruled otherwise (?) In all semantics: is justified if is the conclusion of some justified argument is defensible if is not justified and is the conclusion of some defensible argument The status of arguments: proof theory Argument games between proponent and opponent: Proponent starts with an argument Then each party replies with a suitable counterargument Possibly backtracking A winning criterion E.g. the other player cannot move An argument is (dialectically) provable iff proponent has a winning strategy in a game for it. The G-game for grounded semantics: A sound and complete game: Each move replies to previous move Proponent does not repeat moves Proponent moves strict defeaters, opponent moves defeaters A player wins iff the other player cannot move Result: A is in the grounded extension iff proponent has a winning strategy in a game about A. A game tree A F B C D E A game tree A F B C D E P: A A game tree A P: A F O: F B C D E A game tree A P: A F O: F B P: E C D E A game tree A P: A F O: F B P: E C D E O: B A game tree A P: A F O: B O: F B P: E C D E P: C A game tree A P: A F O: B O: F B P: E C P: C E O: D D A game tree A P: A F O: B O: F B P: E C P: C E O: D D P: E The structure of arguments: current accounts Assumption-based approaches (Dung-Kowalski-Toni, Besnard & Hunter, …) K = theory A = assumptions, - is conflict relation on A R = inference rules (strict) An argument for p is a set A’ A such that A’ K |-R p Arguments attack each other on their assumptions Rule-based approaches (Pollock, Vreeswijk, DeLP, Prakken & Sartor, Defeasible Logic, …) K = theory R = inference rules (strict and defeasible) K yields an argument for p if K |-R p Arguments attack each other on applications of defeasible inference rules Aspic system: overview Argument structure based on Vreeswijk (1997) ≈ Trees where Nodes are wff of logical language L closed under negation Links are applications of inference rules Strict (1, ..., 1 ); or Defeasible (1, ..., 1 ) Reasoning starts from knowledge base K L Defeat based on Pollock Argument acceptability based on Dung (1995) ASPIC system: structure of arguments An argument A is: if K with A1, ..., An if there is a strict inference rule Conc(A1), ..., Conc(An) Conc(A) = {} Sub(A) = Sub(A1) ... Sub(An) {A} A1, ..., An if there is a defeasible inference rule Conc(A1), ..., Conc(An) Conc(A) = {} Sub(A) = Conc(A) = {} Sub(A) = Sub(A1) ... Sub(An) {A} A is strict if all members of Sub(A) apply strict rules; else A is defeasible P Q1,R1,R2 K Q1 Q1, Q2 P Q2 R1 R1, R2 Q2 R2 Domain-specific vs. inference general inference rules Flies R1: Bird Flies R2: Penguin Bird Penguin K Bird Penguin R1: , Strict rules: all deductively valid inference rules Bird Flies K Bird Penguin Bird K Penguin K Penguin Flies Bird Flies Penguin Bird ASPIC system: attack and defeat ≥ is a preference ordering between arguments such that if A is strict and B is defeasible then A > B A rebuts B if A undercuts B if Conc(A) = ¬Conc(B’ ) for some B’ Sub(B); and B’ applies a defeasible rule; and not B’ > A Naming convention implicit Conc(A) = ¬B’ for some B’ Sub(B); and B’ applies a defeasible rule A defeats B if A rebuts or undercuts B P Q1 Q2 Q2 R1 R2 V1 V2 S2 V3 T1 T2 Argument acceptability Dung-style semantics and proof theory directly apply! Additional properties (cf. Caminada & Amgoud 2007) Let E be any stable, preferred or grounded extension: If B Sub(A) and A E then B E If the strict rules RS are closed under contraposition, then {| = Conc(A) for some A E } is 1. 2. closed under RS; consistent if K is consistent Argument schemes Many arguments (and attacks) follow patterns. Much work in argumentation theory (Perelman, Toulmin, Walton, ...) Argument schemes Critical questions Recent applications in AI (& Law) Argument schemes: general form Premise 1, …, Premise n Therefore (presumably), conclusion But also critical questions Negative answers are counterarguments Expert testimony (Walton 1996) E is expert on D E says that P P is within D Therefore (presumably), P is the case Critical questions: Is E biased? Is P consistent with what other experts say? Is P consistent with known evidence? Witness testimony Witness W says P Therefore (presumably), P Critical questions: Is W sincere? (veracity) Was P evidenced by W’s senses? (objectivity) Did P occur? (observational sensitivity) Perception P is observed Therefore (presumably), P Critical questions: Are the circumstances such that reliable observation of P is impossible? … Memory P is recalled Therefore (presumably), P Critical questions: Was P originally based on beliefs of which one is false? … ‘Unpacking’ the witness testimony scheme Witness testimony Witness W says “I remember I saw P” Therefore (presumably), W remembers he saw P Therefore (presumably), W saw P Witness Therefore (presumably), P testimony Critical questions: Is W sincere? (veracity) Was P evidenced by W’s senses? (objectivity) Did P occur? (observational sensitivity) ‘Unpacking’ the witness testimony scheme Memory Witness W says “I remember I saw P” Therefore (presumably), W remembers he saw P Therefore (presumably), W saw P Therefore (presumably), P Memory Critical questions: Is W sincere? (veracity) Was P evidenced by W’s senses? (objectivity) Did P occur? (observational sensitivity) ‘Unpacking’ the witness testimony scheme Witness W says “I remember I saw P” Therefore (presumably), W remembers he saw P Perception Therefore (presumably), W saw P Therefore (presumably), P Critical questions: Perception Is W sincere? (veracity) Was P evidenced by W’s senses? (objectivity) Did P occur? (observational sensitivity) Applying commonsense generalisations Consc of Guilt P If P then usually Q Therefore (presumably), Q Fleas If Fleas then usually Consc of Guilt People who flea from a crime scene usually have consciousness of guilt Critical questions: are there exceptions to the generalisation? exceptional classes of people may have other reasons to flea Illegal immigrants Customers of prostitutes … Arguments from consequences Action A brings about G, G is good (bad) Therefore (presumably), A should (not) be done Critical questions: Does A also have bad (good) consequences? Are there other ways to bring about G? ... Other work on argument-based inference Reasoning about priorities and defeat Abstract support relations between arguments Gradual defeat Other semantics Dialectical proof theories Combining modes of reasoning ... Part 2: Dialogue ‘Argument’ is ambiguous Inferential structure Single agents (Nonmonotonic) logic Fixed information state Form of dialogue Multiple agents Dialogue theory Changing information state Example P: Tell me all you know about recent trading in explosive materials (request) P: why don’t you want to tell me? P: why aren’t you allowed to tell me? P: You may be right in general (concede) but in this case there is an exception since this is a matter of national importance P: since we have heard about a possible terrorist attack P: OK, I agree (offer accepted). O: No I won’t (reject) O: since I am not allowed to tell you O: since sharing such information could endanger an investigation O: Why is this a matter of national importance? O: I concede that there is an exception, so I retract that I am not allowed to tell you. I will tell you on the condition that you don’t exchange the information with other police officers (offer) Example P: Tell me all you know about recent trading in explosive materials (request) P: why don’t you want to tell me? P: why aren’t you allowed to tell me? P: You may be right in general (concede) but in this case there is an exception since this is a matter of national importance P: since we have heard about a possible terrorist attack P: OK, I agree (offer accepted). O: No I won’t (reject) O: since I am not allowed to tell you O: since sharing such information could endanger an investigation O: Why is this a matter of national importance? O: I concede that there is an exception, so I retract that I am not allowed to tell you. I will tell you on the condition that you don’t exchange the information with other police officers (offer) Example P: Tell me all you know about recent trading in explosive materials (request) P: why don’t you want to tell me? P: why aren’t you allowed to tell me? P: You may be right in general (concede) but in this case there is an exception since this is a matter of national importance P: since we have heard about a possible terrorist attack P: OK, I agree (offer accepted). O: No I won’t (reject) O: since I am not allowed to tell you O: since sharing such information could endanger an investigation O: Why is this a matter of national importance? O: I concede that there is an exception, so I retract that I am not allowed to tell you. I will tell you on the condition that you don’t exchange the information with other police officers (offer) Types of dialogues (Walton & Krabbe) Dialogue Type Dialogue Goal Initial situation Persuasion resolution of conflict conflict of opinion Negotiation making a deal conflict of interest Deliberation reaching a decision need for action Information seeking exchange of information personal ignorance Inquiry growth of knowledge general ignorance Dialogue systems (according to Carlson 1983) Dialogue systems define the conditions under which an utterance is appropriate An utterance is appropriate if it promotes the goal of the dialogue in which it is made Appropriateness defined not at speech act level but at dialogue level Dialogue game approach Protocol should promote the goal of the dialogue Formal dialogue systems Topic language With a logic (possibly nonmonotonic) Communication language Locution + content (from topic language) With a protocol: rules for when utterances may be made Should promote the goal of the dialogue Effect rules (e.g. on agent’s commitments) Termination and outcome rules Negotiation Dialogue goal: making a deal Participants’ goals: maximise individual gain Typical communication language: Request p, Offer p, Accept p, Reject p, … Persuasion Participants: proponent (P) and opponent (O) of a dialogue topic T Dialogue goal: resolve the conflict of opinion on T Participants’ goals: P wants O to accept T O wants P to give up T Typical speech acts: Claim p, Concede p, Why p, p since S, Retract p, Deny p … Goal of argument games: Verify logical status of argument or proposition relative to given theory Standards for dialogue systems Argument games: soundness and Dialogue systems: completeness wrt some logical semantics Effectiveness wrt dialogue goal Efficiency, relevance, termination, ... Fairness wrt participants’ goals Can everything relevant be said?, ... Some standards for persuasion systems Correspondence With participants’ beliefs If union of beliefs implies p, can/will agreement on p result? If parties agree that p, does the union of their beliefs imply p? ... With ‘dialogue theory’ If union of commitments implies p, can/will agreement on p result? ... A communication language (Dijkstra et al. 2007) Speech act Attack Surrender request() offer (’), reject() - offer() offer(’) ( ≠ ’), reject() accept() reject() offer(’) ( ≠ ’), why-reject () - accept() - - why-reject() claim (’) - claim() why() concede() why() since S (an argument) retract() since S why() ( S) ’ since S’ (a defeater) concede() concede ’ (’ S) concede() - - retract() - - deny() - - A protocol (Dijkstra et al. 2007) Start with a request Repy to a previous move of the other agent Pick your replies from the table Finish persuasion before resuming negotiation Turntaking: In nego: after each move In pers: various rules possible Termination: In nego: if offer is accepted or someone withdraws In pers: if main claim is retracted or conceded Example dialogue formalised P: Request to tell O: Reject to tell P: Why reject to tell? Embedded persuasion ... O: Offer to tell if no further exchange P: Accept after tell no further exchange Persuasion part formalised O: Claim Not allowed to tell P: Why not allowed to tell? O: Not allowed to tell since telling endangers investigation & What endangers an investigation is not allowed P: Concede What endangers an investigation is not allowed P: Exception to R1 since National importance & National importance Exception to R1 O: Why National importance? P: National importance since Terrorist threat & Terrorist threat National importance Persuasion part formalised O: Claim Not allowed to tell P: Why not allowed to tell? O: Not allowed to tell since telling endangers investigation & What endangers an investigation is not allowed P: Concede What endangers an investigation is not allowed P: Concede Exception to R1 P: Exception to R1 since National importance & National importance Exception to R1 O: Why National importance? P: National importance since Terrorist threat & Terrorist threat National importance Persuasion part formalised O: Claim Not allowed to tell P: Why not allowed to tell? O: Not allowed to tell since telling endangers investigation & What endangers an investigation is not allowed P: Concede What endangers an investigation is not allowed O: Concede Exception to R1 O: Retract Not allowed to tell P: Exception to R1 since National importance & National importance Exception to R1 O: Why National importance? P: National importance since Terrorist threat & Terrorist threat National importance Theory building in dialogue In my 2005 approach to (persuasion) dialogue: Agents build a joint theory during the dialogue A dialectical graph Moves are operations on the joint theory claim Not allowed to tell claim Not allowed to tell why claim Not allowed to tell why since Telling endangers investigation R1: What endangers an investigation is not allowed claim Not allowed to tell why since Telling endangers investigation concede R1: What endangers an investigation is not allowed claim why Not allowed to tell since concede Telling endangers investigation R1: What endangers an investigation is not allowed Exception to R1 R2: national importance Not R1 since National importance claim why Not allowed to tell since concede Telling endangers investigation R1: What endangers an investigation is not allowed Exception to R1 since why R2: national importance Not R1 National importance claim why Not allowed to tell since concede Telling endangers investigation R1: What endangers an investigation is not allowed Exception to R1 since why R2: national importance Not R1 National importance since Terrorist threat Terrorist threat national importance claim why Not allowed to tell since concede Telling endangers investigation R1: What endangers an investigation is not allowed concede Exception to R1 since why R2: national importance Not R1 National importance since Terrorist threat Terrorist threat national importance claim why Not allowed to tell retract since Telling endangers investigation concede R1: What endangers an investigation is not allowed concede Exception to R1 since why R2: national importance Not R1 National importance since Terrorist threat Terrorist threat national importance Research issues Investigation of protocol properties Combinations of dialogue types Mathematical proof or experimentation Deliberation! Multi-party dialogues Dialogical agent behaviour (strategies) ... Further information http://people.cs.uu.nl/henry/siks/siks09.html