This action might not be possible to undo. Are you sure you want to continue?

Welcome to Scribd! Start your free trial and access books, documents and more.Find out more

**Reasoning for Agreement Technologies
**

.

**Wojtek Jamroga and Leon van der Torre
**

University of Luxembourg

Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies

ESSLLI 2013 @ Duesseldorf, Germany 1/104

Lecture Overview I

1

Organizations: Reasoning about collective play (WJ) Modal logics of individual and coalitional ability in games: ATL. Semantics of ATL in concurrent game structures. Axioms. Semantic variants of ATL: information and recall. Comparing general properties of coalitional play. Discussion.

Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies

ESSLLI 2013 @ Duesseldorf, Germany 2/104

Lecture Overview II

2

Norms: Traditional modal logic and alternatives (LT) Traditional modal deontic logic and its paradoxes. Input/output logic and other alternatives for a new standard. Abstract normative systems. Proof theory based on equivalence and redundancy. System O. Discussion.

Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies

ESSLLI 2013 @ Duesseldorf, Germany 3/104

Discussion. Abstract models of games. Succinct speciﬁcations of security. Decision problems. and abilities: strategic description logics. Description logic. Reasoning about concepts. Equivalence wrt game logics. Pauly’s representation theorem and true playability.Lecture Overview III 3 Models of interaction & dynamics of concepts (WJ) Game-theoretic models of interaction. Impact on axioms. agents. Germany 4/104 . Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

rule based systems. Discussion. Doxastic and game-theoretic structure of trust.Lecture Overview IV 4 Argumentation: Abstract and instantiated (LT) From non-monotonic logic to abstract argumentation and its instantiations: classical logic. and compact representations. Equivalence of frameworks. Leon’s part???? Discussion. Germany 5/104 . Decision procedures based on dialogue games. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. 5 Trust: From beliefs to decisions and back (WJ+LT) Castelfranchi’s view on trust. and preferences.

reasoning (I) Logic and Social Choice Theory (I) Modularity in Ontologies (A) Uniﬁcation in description logics and modal logics (A) Logical Models of Group Decision Making (WS) Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 6/104 .Pointers to Related Courses and Workshops 10 Puzzles and Paradoxes about Knowledge & Belief (F) Logics for multi-agent systems and strat.

Jeff Horty. and Leon van der Torre (eds). Speciﬁcation and Veriﬁcation of Multi-Agent Systems. 2010. 4 5 Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. 2009. Mehdi Dastani. Handbook on deontic logic and normative systems (ﬁrst volume.). and John-Jules Meyer (eds. 2009. Ron van der Meyden.Basic Reading 1 2 3 Handbook on Agreement Technologies (in preparation).). Germany 7/104 . Koen Hindriks. Springer. to appear September 2012). Argumentation in Artiﬁcial Intelligence. Trust Theory: A Socio-Cognitive and Computational Model. Springer. Cristiano Castelfranchi and Rino Falcone. Dov Gabbay. Iyad Rahwan and Guillermo Simari (ed.

Reasoning about Collective Play Reasoning about Collective Play Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 8/104 .. 1. .

1. Reasoning about Collective Play 1. . Agent Temporal Logic 1..1 Agent Temporal Logic Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 9/104 .

Germany 10/104 . Agent Temporal Logic ATL: What Agents Can Achieve ATL: Alternating-time Temporal Logic (Alur et al. 1997-2002) Temporal logic meets game theory Main idea: cooperation modalities Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.1. Reasoning about Collective Play 1.

⟨⟨A⟩⟩Φ: coalition A has a collective strategy to enforce Φ Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Reasoning about Collective Play 1. 1997-2002) Temporal logic meets game theory Main idea: cooperation modalities . Germany 10/104 . Agent Temporal Logic ATL: What Agents Can Achieve ATL: Alternating-time Temporal Logic (Alur et al.1. .

Agent Temporal Logic Syntax . j γ | 3γ | 2γ | γ U γ . Reasoning about Collective Play 1. Germany 11/104 .γ ::= φ | ¬γ | γ ∧ γ | Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.1. . φ ::= p | ¬φ | φ ∧ φ | ⟨⟨A⟩⟩γ .

.γ ::= φ | ¬γ | γ ∧ γ | In fact. Agent Temporal Logic Syntax . Reasoning about Collective Play 1.1. Germany 11/104 . “eventually” and “always” can be derived from “until”: 3 γ ≡ true U γ 2 γ ≡ ¬3 ¬γ Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. φ ::= p | ¬φ | φ ∧ φ | ⟨⟨A⟩⟩γ . j γ | 3γ | 2γ | γ U γ .

. Reasoning about Collective Play 1. Germany 11/104 . “eventually” and “always” can be derived from “until”: 3 γ ≡ true U γ 2 γ ≡ ¬3 ¬γ “Vanilla” ATL: every temporal operator preceded by exactly one cooperation modality ATL*: no syntactic restrictions Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. j γ | 3γ | 2γ | γ U γ . φ ::= p | ¬φ | φ ∧ φ | ⟨⟨A⟩⟩γ .γ ::= φ | ¬γ | γ ∧ γ | In fact.1. Agent Temporal Logic Syntax .

Germany 12/104 .1. Agent Temporal Logic ⟨⟨jamesbond⟩⟩3 (ski ∧ ¬getBurned): “James Bond can go skiing without getting burned” Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Reasoning about Collective Play 1.

Agent Temporal Logic ⟨⟨jamesbond⟩⟩3 (ski ∧ ¬getBurned): “James Bond can go skiing without getting burned” Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.1. Germany 12/104 . Reasoning about Collective Play 1.

Reasoning about Collective Play 1. Agent Temporal Logic ⟨⟨jamesbond⟩⟩3 (ski ∧ ¬getBurned): “James Bond can go skiing without getting burned” ⟨⟨jamesbond. Germany 12/104 .1. bondsgirl⟩⟩fun U shot: “James Bond and his girlfriend are able to have fun until someone shoots at them” Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

Reasoning about Collective Play 2. 1.. Semantics of ATL 1.2 Semantics of ATL Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 13/104 . .

transitions. Semantics of ATL ATL Models: Concurrent Game Structures Agents. actions.1. Germany 14/104 . atomic propositions Atomic propositions + interpretation Actions are abstract Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Reasoning about Collective Play 2.

1. d. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. A concurrent game structure is a tuple M = ⟨Agt. o⟩.1 (Concurrent Game Structure) . Act. where: . Reasoning about Collective Play 2. St. Deﬁnition 1. Semantics of ATL . V . Germany 15/104 .

St. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. d.1 (Concurrent Game Structure) . Semantics of ATL . A concurrent game structure is a tuple M = ⟨Agt. V . o⟩. Reasoning about Collective Play 2. Deﬁnition 1. where: Agt: a ﬁnite set of all agents . Act.1. Germany 15/104 .

1 (Concurrent Game Structure) . A concurrent game structure is a tuple M = ⟨Agt. where: Agt: a ﬁnite set of all agents St: a set of states . Reasoning about Collective Play 2.1. St. o⟩. Act. d. Germany 15/104 . Deﬁnition 1. V . Semantics of ATL . Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

Semantics of ATL . Act.1.1 (Concurrent Game Structure) . where: Agt: a ﬁnite set of all agents St: a set of states V : a valuation of propositions . Germany 15/104 . d. Deﬁnition 1. A concurrent game structure is a tuple M = ⟨Agt. o⟩. V . St. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Reasoning about Collective Play 2.

St. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Semantics of ATL . Germany 15/104 . A concurrent game structure is a tuple M = ⟨Agt. d. Reasoning about Collective Play 2. V . o⟩.1. where: Agt: a ﬁnite set of all agents St: a set of states V : a valuation of propositions Act: a ﬁnite set of (atomic) actions . Act. Deﬁnition 1.1 (Concurrent Game Structure) .

of actions Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. . Germany 15/104 .1 (Concurrent Game Structure) . . Deﬁnition 1. Semantics of ATL . o⟩. A concurrent game structure is a tuple M = ⟨Agt. αk ) to states and tuples . Act. St. d. Reasoning about Collective Play 2. .1. V . . where: Agt: a ﬁnite set of all agents St: a set of states V : a valuation of propositions Act: a ﬁnite set of (atomic) actions d : Agt × St → 2Act deﬁnes actions available to an agent in a state o: a deterministic transition function that assigns outcome states q ′ = o(q. α1 .

Reasoning about Collective Play 2.1. Germany 16/104 . Semantics of ATL Example: Robots and Carriage 1 2 pos0 1 2 pos2 1 pos1 2 Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

wait q1 wait.w ait wa it.1. Reasoning about Collective Play 2.wait push.push pos1 ESSLLI 2013 @ Duesseldorf.push 1 2 pos0 q0 pos0 pu sh.p pos2 1 pos1 2 wait. Semantics of ATL Example: Robots and Carriage wait.push push. Germany 16/104 .p us h it wa h us it. wa sh .wait push.push 2 pu Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies 1 pos2 q2 wait.wait push.

Deﬁnition 1. . A strategy is a conditional plan. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.1. Reasoning about Collective Play 2. Semantics of ATL . Germany 17/104 .2 (Strategy) .

Semantics of ATL .2 (Strategy) . We represent strategies by functions sa : St → Act. Deﬁnition 1. Reasoning about Collective Play 2. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. . A strategy is a conditional plan. Germany 17/104 .1.

A strategy is a conditional plan. memoryless agents Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.1. . Deﬁnition 1.2 (Strategy) . Semantics of ATL . . Germany 17/104 . We represent strategies by functions sa : St → Act. Reasoning about Collective Play 2.

1. Reasoning about Collective Play 2. We represent strategies by functions sa : St → Act. memoryless agents Alternative: perfect recall strategies sa : St+ → Act Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 17/104 . . Deﬁnition 1. A strategy is a conditional plan. Semantics of ATL . .2 (Strategy) .

We represent strategies by functions sa : St → Act. Function out(q. . Semantics of ATL .onward. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. sA ) returns the set of all paths that may result from agents A executing strategy sA from state q . A strategy is a conditional plan. Germany 17/104 .1. Deﬁnition 1.2 (Strategy) . memoryless agents Alternative: perfect recall strategies sa : St+ → Act . . Reasoning about Collective Play 2.

q |= ¬φ iff M. q ̸|= φ.1. q |= p iff p is in V (q ). Semantics of ATL .3 (Semantics of ATL*) . q |= φ1 ∧ φ2 iff M. M. ESSLLI 2013 @ Duesseldorf. M. Reasoning about Collective Play 2. Deﬁnition 1. Germany 18/104 . q |= φ2 . M. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies . q |= φ1 and M.

λ |= Φ. we have M.3 (Semantics of ATL*) . Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies . q |= p iff p is in V (q ). sA ). q |= φ2 . q |= ⟨⟨A⟩⟩Φ iff there is a collective strategy sA such that. q |= φ1 ∧ φ2 iff M. q |= φ1 and M. M. M. ESSLLI 2013 @ Duesseldorf.1. Semantics of ATL . Reasoning about Collective Play 2. Germany 18/104 . q |= ¬φ iff M. q ̸|= φ. M. M. Deﬁnition 1. for every path λ ∈ out(q.

λ |= j γ M. for every path λ ∈ out(q. q |= p iff p is in V (q ). sA ). M. ESSLLI 2013 @ Duesseldorf. λ |= γ1 U γ2 Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies . λ[i. q ̸|= φ. M. iff M. M.. q |= φ1 ∧ φ2 iff M.∞] |= γ1 forall 0 ≤ j ≤ i. λ[j. λ |= Φ. Semantics of ATL . q |= ⟨⟨A⟩⟩Φ iff there is a collective strategy sA such that. M. iff M. λ[1.. Deﬁnition 1..∞] |= γ for all i ≥ 0. we have M.3 (Semantics of ATL*) .. iff M.1. and M. q |= φ2 . q |= φ1 and M. q |= ¬φ iff M. λ |= 2 γ M. M.∞] |= γ .∞] |= γ2 for some i ≥ 0. Reasoning about Collective Play 2. Germany 18/104 . λ[i.

iff M. q |= p iff p is in V (q ). M. iff M. λ |= ¬γ ESSLLI 2013 @ Duesseldorf. λ[i.. M. we have M.∞] |= γ1 forall 0 ≤ j ≤ i. Semantics of ATL . and M. λ |= j γ M. q ̸|= φ. λ |= φ . λ |= γ1 U γ2 Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies M.3 (Semantics of ATL*) . iff M. for every path λ ∈ out(q. M. Germany 18/104 . λ |= Φ.. Reasoning about Collective Play 2.. q |= ⟨⟨A⟩⟩Φ iff there is a collective strategy sA such that. λ[j. iff M. q |= φ2 . q |= φ1 and M. λ ̸|= γ etc. Deﬁnition 1. M. sA ).. λ[1.∞] |= γ . q |= ¬φ iff M. q |= φ1 ∧ φ2 iff M. λ[0] |= φ.∞] |= γ for all i ≥ 0.1. λ |= 2 γ M. M.∞] |= γ2 for some i ≥ 0. iff M. M. λ[i.

q |= ⟨⟨A⟩⟩φ1 U φ2 iff there is sA such that. for every λ ∈ out(q. λ[i] |= φ for all i ≥ 0. we have M. M. q |= φ1 ∧ φ2 iff M. iff there is sA such that. M. sA ). q |= φ1 and M. q |= ¬φ iff M. iff there is sA such that. Semantics of ATL The semantics of “vanilla” ATL can be given entirely in terms of models and states: . Germany 19/104 . q |= φ2 . q |= p iff p is in V (q ). . sA ). sA ). q ̸|= φ. M. λ[1] |= φ. we have M. λ[i] |= φ2 for some i ≥ 0 and M.1. q |= ⟨⟨A⟩⟩ j φ M. for every λ ∈ out(q. λ[j ] |= φ1 for all 0 ≤ j ≤ i. q |= ⟨⟨A⟩⟩2 φ M. Reasoning about Collective Play 2. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. we have M. for every λ ∈ out(q. M.

w ait wa it. wa sh . Semantics of ATL Example: Robots and Carriage wait. Reasoning about Collective Play 2.wait push.push pu pos0 → ⟨⟨1⟩⟩2 ¬pos1 wait.push q0 pos0 pu sh.wait push.push push.p pos2 q2 wait.push it wa h us it.wait push.wait q1 pos1 Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 20/104 .p us h wait.1.

push push. wa sh .wait q1 pos1 Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.wait push.p us h wait. Reasoning about Collective Play 2.wait push.push q0 pos0 pu sh. Semantics of ATL Example: Robots and Carriage wait.push it wa h us it. Germany 20/104 .push pu pos0 → ⟨⟨1⟩⟩2 ¬pos1 wait.1.p pos2 q2 wait.wait push.w ait wa it.

w ait wa it. Germany 20/104 .push push pu pos0 → ⟨⟨1⟩⟩2 ¬pos1 wait.wait push.push wait it wa h us it.1.p pos2 q2 wait.wait q1 pos1 Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.wait push.wait push.push push. Reasoning about Collective Play 2.p us h wait.push wait q0 pos0 pu sh. wa sh . Semantics of ATL Example: Robots and Carriage wait.

push pos2 q2 wait.w a it us h wait.1. Germany 20/104 .push q0 pos0 it wa sh.wait push. Semantics of ATL Example: Robots and Carriage wait. Reasoning about Collective Play 2.push pu wa it.wait push. h pu us it.p pos0 → ⟨⟨1⟩⟩2 ¬pos1 wait.push push.wait push.p wa sh .wait q1 pos1 Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

Semantics of ATL Example: Robots and Carriage wait.wait push.push push.p wa sh .1.wait push.push q0 pos0 it wa sh.wait q1 pos1 Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Reasoning about Collective Play 2.wait push.w a it us h wait. h pu us it. Germany 20/104 .p pos0 → ⟨⟨1⟩⟩2 ¬pos1 wait.push pos2 q2 wait.push pu wa it.

Germany 20/104 .wait push.push pos2 q2 wait.wait push.w a it us h wait.push push.wait q1 pos1 Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Semantics of ATL Example: Robots and Carriage wait.push pu wa it. Reasoning about Collective Play 2.p pos0 → ⟨⟨1⟩⟩2 ¬pos1 wait.p wa sh .1.push q0 pos0 it wa sh. h pu us it.wait push.

wait q1 pos1 Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Semantics of ATL Example: Robots and Carriage wait.wait push.w a it us h wait.push pos2 q2 wait. Germany 20/104 .p pos0 → ⟨⟨1⟩⟩2 ¬pos1 wait.p wa sh .push push.push q0 pos0 it wa sh.wait push.1.wait push. h pu us it.push pu wa it. Reasoning about Collective Play 2.

. 1. . Importance of ATL 1.3 Importance of ATL Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 21/104 . Reasoning about Collective Play 3.

Importance of ATL Connection to Temporal Logic ATL(*) extends the branching-time logic CTL(*): A ≡ ⟨⟨∅⟩⟩ (“for all paths”) E ≡ ⟨⟨Agt⟩⟩ (“there is a path”) Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.1. Germany 22/104 . Reasoning about Collective Play 3.

Reasoning about Collective Play 3.1. Germany 23/104 . Importance of ATL Connection to Temporal Logic Temporal operators allow a number of useful concepts to be formally speciﬁed: Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

Germany 23/104 . Reasoning about Collective Play 3.1. Importance of ATL Connection to Temporal Logic Temporal operators allow a number of useful concepts to be formally speciﬁed: Safety properties: ⟨⟨os⟩⟩2 ¬crash Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

Germany 23/104 . Reasoning about Collective Play 3. bob⟩⟩3 paperAccepted Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Importance of ATL Connection to Temporal Logic Temporal operators allow a number of useful concepts to be formally speciﬁed: Safety properties: ⟨⟨os⟩⟩2 ¬crash Liveness properties: ⟨⟨alice.1.

Germany 23/104 . Importance of ATL Connection to Temporal Logic Temporal operators allow a number of useful concepts to be formally speciﬁed: Safety properties: ⟨⟨os⟩⟩2 ¬crash Liveness properties: ⟨⟨alice. dlr⟩⟩2 (carRequested → 3 carDelivered) (ATL*!) Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Reasoning about Collective Play 3. bob⟩⟩3 paperAccepted Fairness properties: ⟨⟨prod.1.

Importance of ATL Connection to Multi-Agent Systems Validity General properties of systems Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.1. Reasoning about Collective Play 3. Germany 24/104 .

Germany 24/104 . Importance of ATL Connection to Multi-Agent Systems Validity General properties of systems System synthesis Satisﬁability Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Reasoning about Collective Play 3.1.

Germany 24/104 .1. Importance of ATL Connection to Multi-Agent Systems Validity General properties of systems System synthesis Veriﬁcation Satisﬁability Model checking Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Reasoning about Collective Play 3.

Importance of ATL Connection to Games Concurrent game structure = generalized extensive game Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.1. Reasoning about Collective Play 3. Germany 25/104 .

Reasoning about Collective Play 3.1. Germany 25/104 . Importance of ATL Connection to Games Concurrent game structure = generalized extensive game ⟨⟨A⟩⟩γ : ⟨⟨A⟩⟩ splits the agents into proponents and opponents γ deﬁnes the winning condition Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

Germany 25/104 . inﬁnite 2-player. zero-sum game Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.1. Reasoning about Collective Play 3. binary. Importance of ATL Connection to Games Concurrent game structure = generalized extensive game ⟨⟨A⟩⟩γ : ⟨⟨A⟩⟩ splits the agents into proponents and opponents γ deﬁnes the winning condition .

1. inﬁnite 2-player. Germany 25/104 . Importance of ATL Connection to Games Concurrent game structure = generalized extensive game ⟨⟨A⟩⟩γ : ⟨⟨A⟩⟩ splits the agents into proponents and opponents γ deﬁnes the winning condition . binary. Reasoning about Collective Play 3. zero-sum game Flexible and compact speciﬁcation of winning conditions Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

Importance of ATL Connection to Games Concurrent game structure = generalized extensive game ⟨⟨A⟩⟩γ : ⟨⟨A⟩⟩ splits the agents into proponents and opponents γ deﬁnes the winning condition . Germany 25/104 . q |= ⟨⟨A⟩⟩γ Model checking ATL corresponds to game solving in game theory! Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Reasoning about Collective Play 3. inﬁnite 2-player.1. binary. zero-sum game Flexible and compact speciﬁcation of winning conditions Solving a game ≈ checking if M.

1. Reasoning about Collective Play 3. Germany 26/104 . Importance of ATL Connection to Games What about the other decision problems? Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

Germany 26/104 . Reasoning about Collective Play 3. Importance of ATL Connection to Games What about the other decision problems? Validity General properties of games Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.1.

Importance of ATL Connection to Games What about the other decision problems? Validity General properties of games Game synthesis Satisﬁability Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 26/104 . Reasoning about Collective Play 3.1.

1. Validities and Axioms 1. Germany 27/104 ..4 Validities and Axioms Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. . Reasoning about Collective Play 4.

1. Theorem 1. Validities and Axioms Fixpoint Properties .4 . Germany 28/104 . Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. . Reasoning about Collective Play 4. The following formulae are valid in ATL (but not in ATL*!): ⟨⟨A⟩⟩2 φ ↔ φ ∧ ⟨⟨A⟩⟩ j ⟨⟨A⟩⟩2 φ j ⟨⟨A⟩⟩3 φ ↔ φ ∨ ⟨⟨A⟩⟩ ⟨⟨A⟩⟩3 φ ⟨⟨A⟩⟩φ1 U φ2 ↔ φ2 ∨ φ1 ∧ ⟨⟨A⟩⟩ j ⟨⟨A⟩⟩φ1 U φ2 .

. . Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 28/104 . Validities and Axioms Fixpoint Properties . Strategy for A can be synthesized incrementally (no backtracking is necessary). The following formulae are valid in ATL (but not in ATL*!): ⟨⟨A⟩⟩2 φ ↔ φ ∧ ⟨⟨A⟩⟩ j ⟨⟨A⟩⟩2 φ j ⟨⟨A⟩⟩3 φ ↔ φ ∨ ⟨⟨A⟩⟩ ⟨⟨A⟩⟩3 φ ⟨⟨A⟩⟩φ1 U φ2 ↔ φ2 ∨ φ1 ∧ ⟨⟨A⟩⟩ j ⟨⟨A⟩⟩φ1 U φ2 .1.4 . Reasoning about Collective Play 4. Corollary . Theorem 1. .

Validities and Axioms What about other general properties of agents and coalitions? Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.1. Germany 29/104 . Reasoning about Collective Play 4.

1. Axioms for ATL (Goranko & van Drimmelen) . (⊥) ¬⟨⟨A⟩⟩ j ⊥ Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 30/104 . Reasoning about Collective Play 4. Validities and Axioms .

Axioms for ATL (Goranko & van Drimmelen) . Reasoning about Collective Play 4. (⊥) ¬⟨⟨A⟩⟩ j ⊥ (⊤) ⟨⟨A⟩⟩ j ⊤ Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Validities and Axioms .1. Germany 30/104 .

Germany 30/104 . Validities and Axioms . (⊥) ¬⟨⟨A⟩⟩ j ⊥ (⊤) ⟨⟨A⟩⟩ j ⊤ (Agt) ¬⟨⟨∅⟩⟩ j ¬φ → ⟨⟨Agt⟩⟩ j φ Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.1. Reasoning about Collective Play 4. Axioms for ATL (Goranko & van Drimmelen) .

Axioms for ATL (Goranko & van Drimmelen) . where A1 and A2 are disjoint Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Validities and Axioms .1. Reasoning about Collective Play 4. (⊥) ¬⟨⟨A⟩⟩ j ⊥ (⊤) ⟨⟨A⟩⟩ j ⊤ (Agt) ¬⟨⟨∅⟩⟩ j ¬φ → ⟨⟨Agt⟩⟩ j φ (S ) ⟨⟨A1 ⟩⟩ j φ1 ∧ ⟨⟨A2 ⟩⟩ j φ2 → ⟨⟨A1 ∪ A2 ⟩⟩ j (φ1 ∧ φ2 ). Germany 30/104 .

Validities and Axioms .1. (⊥) ¬⟨⟨A⟩⟩ j ⊥ (⊤) ⟨⟨A⟩⟩ j ⊤ (Agt) ¬⟨⟨∅⟩⟩ j ¬φ → ⟨⟨Agt⟩⟩ j φ (S ) ⟨⟨A1 ⟩⟩ j φ1 ∧ ⟨⟨A2 ⟩⟩ j φ2 → ⟨⟨A1 ∪ A2 ⟩⟩ j (φ1 ∧ φ2 ). Reasoning about Collective Play 4. where A1 and A2 are disjoint (F P2 ) ⟨⟨A⟩⟩2 φ ↔ φ ∧ ⟨⟨A⟩⟩ j ⟨⟨A⟩⟩2 φ j (GF P2 ) ⟨⟨∅⟩⟩2 (θ → (φ ∧ ⟨⟨A⟩⟩ θ)) → ⟨⟨∅⟩⟩2 (θ → ⟨⟨A⟩⟩2 φ) Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Axioms for ATL (Goranko & van Drimmelen) . Germany 30/104 .

Validities and Axioms .1. (⊥) ¬⟨⟨A⟩⟩ j ⊥ (⊤) ⟨⟨A⟩⟩ j ⊤ (Agt) ¬⟨⟨∅⟩⟩ j ¬φ → ⟨⟨Agt⟩⟩ j φ (S ) ⟨⟨A1 ⟩⟩ j φ1 ∧ ⟨⟨A2 ⟩⟩ j φ2 → ⟨⟨A1 ∪ A2 ⟩⟩ j (φ1 ∧ φ2 ). Germany 30/104 . Reasoning about Collective Play 4. Axioms for ATL (Goranko & van Drimmelen) . where A1 and A2 are disjoint (F P2 ) ⟨⟨A⟩⟩2 φ ↔ φ ∧ ⟨⟨A⟩⟩ j ⟨⟨A⟩⟩2 φ j (GF P2 ) ⟨⟨∅⟩⟩2 (θ → (φ ∧ ⟨⟨A⟩⟩ θ)) → ⟨⟨∅⟩⟩2 (θ → ⟨⟨A⟩⟩2 φ) (F P U ) ⟨⟨A⟩⟩(φ1 U φ2 ) ↔ φ2 ∨ (φ1 ∧ ⟨⟨A⟩⟩ j ⟨⟨A⟩⟩(φ1 U φ2 )) j (LF P U ) ⟨⟨∅⟩⟩2 ((φ2 ∨ (φ1 ∧ ⟨⟨A⟩⟩ θ)) → θ) → ⟨⟨∅⟩⟩2 (⟨⟨A⟩⟩(φ1 U φ2 ) → θ) Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

(⊥) ¬⟨⟨A⟩⟩ j ⊥ (⊤) ⟨⟨A⟩⟩ j ⊤ (Agt) ¬⟨⟨∅⟩⟩ j ¬φ → ⟨⟨Agt⟩⟩ j φ (S ) ⟨⟨A1 ⟩⟩ j φ1 ∧ ⟨⟨A2 ⟩⟩ j φ2 → ⟨⟨A1 ∪ A2 ⟩⟩ j (φ1 ∧ φ2 ).1. φ 1 → φ2 (M P ) φ2 φ1 → φ2 (M on) h ⟨⟨A⟩⟩ φ1 → ⟨⟨A⟩⟩ h φ2 φ (N ec) ⟨⟨∅⟩⟩2 φ Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. where A1 and A2 are disjoint (F P2 ) ⟨⟨A⟩⟩2 φ ↔ φ ∧ ⟨⟨A⟩⟩ j ⟨⟨A⟩⟩2 φ j (GF P2 ) ⟨⟨∅⟩⟩2 (θ → (φ ∧ ⟨⟨A⟩⟩ θ)) → ⟨⟨∅⟩⟩2 (θ → ⟨⟨A⟩⟩2 φ) (F P U ) ⟨⟨A⟩⟩(φ1 U φ2 ) ↔ φ2 ∨ (φ1 ∧ ⟨⟨A⟩⟩ j ⟨⟨A⟩⟩(φ1 U φ2 )) j (LF P U ) ⟨⟨∅⟩⟩2 ((φ2 ∨ (φ1 ∧ ⟨⟨A⟩⟩ θ)) → θ) → ⟨⟨∅⟩⟩2 (⟨⟨A⟩⟩(φ1 U φ2 ) → θ) φ1 . Validities and Axioms . Reasoning about Collective Play 4. Germany 30/104 . Axioms for ATL (Goranko & van Drimmelen) .

Theorem 1. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 31/104 . . The axiomatic system is sound and complete for ATL.5 (Goranko & van Drimmelen 2006) . Reasoning about Collective Play 4.1. Validities and Axioms .

Reasoning about Collective Play 5. Imperfect Information 1. .5 Imperfect Information Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 32/104 .. 1.

1. Reasoning about Collective Play

5. Imperfect Information

**Combining Knowledge and Strategies
**

How can we reason about extensive games with imperfect information?

Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies

ESSLLI 2013 @ Duesseldorf, Germany 33/104

1. Reasoning about Collective Play

5. Imperfect Information

**Combining Knowledge and Strategies
**

How can we reason about extensive games with imperfect information? Let’s put ATL and epistemic logic in one box. We extend CGS with indistinguishability relations ∼a , one per agent We add epistemic operators to ATL

Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies

ESSLLI 2013 @ Duesseldorf, Germany 33/104

1. Reasoning about Collective Play

5. Imperfect Information

**Combining Knowledge and Strategies
**

How can we reason about extensive games with imperfect information? Let’s put ATL and epistemic logic in one box. We extend CGS with indistinguishability relations ∼a , one per agent We add epistemic operators to ATL ; Problems!

Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies

ESSLLI 2013 @ Duesseldorf, Germany 33/104

1. Reasoning about Collective Play

5. Imperfect Information

Example: Robbing a Bank

Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies

ESSLLI 2013 @ Duesseldorf, Germany 34/104

1. Reasoning about Collective Play

5. Imperfect Information

Example: Robbing a Bank

Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies

ESSLLI 2013 @ Duesseldorf, Germany 34/104

1. Reasoning about Collective Play

5. Imperfect Information

**Example: Robbing a Bank
**

′ q0

0⟩

,s ⟨−

⟨− ,s

et

⟩ et 1

q0

⟨t ry

⟨try0 , −⟩

1,

. a

− ⟩ r ⟨t ,− y1 ⟩

q1

⟨try0 , −⟩

q2

open

q3

alarm

Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies

ESSLLI 2013 @ Duesseldorf, Germany 34/104

Reasoning about Collective Play 5. Germany 34/104 .1.− y1 ⟩ q1 ⟨try0 . Imperfect Information Example: Robbing a Bank ′ q0 0⟩ . . −⟩ 1. −⟩ q2 open q3 alarm ′ |= ⟨⟨a⟩⟩3 open q0 ′ q0 |= Ka ⟨⟨a⟩⟩3 open Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. a − ⟩ r ⟨t .s ⟨− ⟨− .s et ⟩ et 1 q0 ⟨t ry ⟨try0 .

−⟩ 1.1. Germany 34/104 .s ⟨− ⟨− .− y1 ⟩ q1 ⟨try0 .s et ⟩ et 1 q0 ⟨t ry ⟨try0 . . a − ⟩ r ⟨t . Imperfect Information Example: Robbing a Bank ′ q0 0⟩ . Reasoning about Collective Play 5. −⟩ q2 open q3 alarm ′ |= ⟨⟨a⟩⟩3 open q0 ′ q0 |= Ka ⟨⟨a⟩⟩3 open Does it make sense? Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

Problem: .1. Imperfect Information Combining Knowledge and Strategies . Germany 35/104 . Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Reasoning about Collective Play 5. Strategic and epistemic abilities are not independent! .

Germany 35/104 . Imperfect Information Combining Knowledge and Strategies .1. Strategic and epistemic abilities are not independent! . Problem: . ⟨⟨A⟩⟩Φ = A can enforce Φ Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Reasoning about Collective Play 5.

1. Problem: . Reasoning about Collective Play 5. ⟨⟨A⟩⟩Φ = A can enforce Φ It should at least mean that A are able to execute (and possibly also identify) the right strategy! Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Imperfect Information Combining Knowledge and Strategies . Strategic and epistemic abilities are not independent! . Germany 35/104 .

Germany 35/104 . Executable strategies = uniform strategies Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Imperfect Information Combining Knowledge and Strategies . Problem: .1. ⟨⟨A⟩⟩Φ = A can enforce Φ It should at least mean that A are able to execute (and possibly also identify) the right strategy! . . Reasoning about Collective Play 5. Strategic and epistemic abilities are not independent! .

Reasoning about Collective Play 5. Deﬁnition 1. Germany 36/104 .1. Strategy sa is uniform iff it speciﬁes the same choices for indistinguishable situations: (no recall:) if q ∼a q ′ then sa (q ) = sa (q ′ ) (perfect recall:) if λ ≈a λ′ then ⇒ sa (λ) = sa (λ). Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. . Imperfect Information Uniform Strategies . where λ ≈a λ′ iff λ[i] ∼a λ′ [i] for every i.6 (Uniform strategy) .

where λ ≈a λ′ iff λ[i] ∼a λ′ [i] for every i.6 (Uniform strategy) . Reasoning about Collective Play 5. .1. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.individual strategies. Strategy sa is uniform iff it speciﬁes the same choices for indistinguishable situations: (no recall:) if q ∼a q ′ then sa (q ) = sa (q ′ ) (perfect recall:) if λ ≈a λ′ then ⇒ sa (λ) = sa (λ). Deﬁnition 1. . Imperfect Information Uniform Strategies . Germany 36/104 . A collective strategy is uniform iff it consists only of uniform .

1. Having a successful strategy does not imply knowing that we have it! . Note: . Imperfect Information . Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 37/104 . Reasoning about Collective Play 5.

1. Note: . Germany 37/104 . Reasoning about Collective Play 5. Knowing that a successful strategy exists does not imply . Having a successful strategy does not imply knowing that we have it! . Imperfect Information .knowing the strategy itself! Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. .

Imperfect Information Levels of Strategic Ability Our cases for ⟨⟨A⟩⟩Φ under imperfect information: 1 There is σ (not necessarily executable!) such that. Φ holds Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. for every execution of σ .1. Germany 38/104 . Reasoning about Collective Play 5.

for every execution of σ . Reasoning about Collective Play 5. for every execution of σ . Φ holds Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 38/104 . Imperfect Information Levels of Strategic Ability Our cases for ⟨⟨A⟩⟩Φ under imperfect information: 1 2 There is σ (not necessarily executable!) such that.1. Φ holds There is a uniform σ such that.

1. Reasoning about Collective Play

5. Imperfect Information

**Levels of Strategic Ability
**

Our cases for ⟨⟨A⟩⟩Φ under imperfect information:

1

2

3

There is σ (not necessarily executable!) such that, for every execution of σ , Φ holds There is a uniform σ such that, for every execution of σ , Φ holds A know that there is a uniform σ such that, for every execution of σ , Φ holds

Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies

ESSLLI 2013 @ Duesseldorf, Germany 38/104

1. Reasoning about Collective Play

5. Imperfect Information

**Levels of Strategic Ability
**

Our cases for ⟨⟨A⟩⟩Φ under imperfect information:

1

2

3

4

There is σ (not necessarily executable!) such that, for every execution of σ , Φ holds There is a uniform σ such that, for every execution of σ , Φ holds A know that there is a uniform σ such that, for every execution of σ , Φ holds There is a uniform σ such that A know that, for every execution of σ , Φ holds

Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies

ESSLLI 2013 @ Duesseldorf, Germany 38/104

1. Reasoning about Collective Play

5. Imperfect Information

**Levels of Strategic Ability
**

Our cases for ⟨⟨A⟩⟩Φ under imperfect information:

1

2

3

4

There is σ (not necessarily executable!) such that, for every execution of σ , Φ holds There is a uniform σ such that, for every execution of σ , Φ holds A know that there is a uniform σ such that, for every execution of σ , Φ holds There is a uniform σ such that A know that, for every execution of σ , Φ holds

From now on, we restrict our discussion to uniform memoryless strategies (unless explicitly stated otherwise).

Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies

ESSLLI 2013 @ Duesseldorf, Germany 38/104

we restrict our discussion to uniform memoryless strategies (unless explicitly stated otherwise).1. for every execution of σ . for every execution of σ . Φ holds A know that there is a uniform σ such that. Φ holds There is a uniform σ such that A know that. Φ holds There is a uniform σ such that. for every execution of σ . Germany 38/104 . Φ holds From now on. Imperfect Information Levels of Strategic Ability Our cases for ⟨⟨A⟩⟩Φ under imperfect information: 1 2 3 4 There is σ (not necessarily executable!) such that. Reasoning about Collective Play 5. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. for every execution of σ .

1. Comparing Variants of Ability 1.6 Comparing Variants of Ability Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Reasoning about Collective Play 6.. . Germany 39/104 .

by comparing validity sets we compare logics in the traditional sense Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Reasoning about Collective Play 6. Germany 40/104 . Comparing Variants of Ability Comparing Different Semantics of ATL Traditional perspective: logic = set of validities Thus.1.

1. Germany 40/104 . Comparing Variants of Ability Comparing Different Semantics of ATL Traditional perspective: logic = set of validities Thus. Reasoning about Collective Play 6. by comparing validity sets we compare logics in the traditional sense Validities capture general properties of games under consideration If two variants of ATL generate the same valid sentences then the underlying notions of ability induce the same kind of games Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

1. Reasoning about Collective Play 6. by comparing validity sets we compare logics in the traditional sense Validities capture general properties of games under consideration If two variants of ATL generate the same valid sentences then the underlying notions of ability induce the same kind of games First step towards devising algorithms for satisﬁability checking Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Comparing Variants of Ability Comparing Different Semantics of ATL Traditional perspective: logic = set of validities Thus. Germany 40/104 .

Reasoning about Collective Play 6. imperfect recall strategies (r) Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 41/104 . Comparing Variants of Ability Semantic Variants of ATL Memory of agents: Perfect recall (R) vs.1.

imperfect information strategies (i) Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. imperfect recall strategies (r) Available information: Perfect information (I) vs.1. Germany 41/104 . Reasoning about Collective Play 6. Comparing Variants of Ability Semantic Variants of ATL Memory of agents: Perfect recall (R) vs.

imperfect recall strategies (r) Available information: Perfect information (I) vs. Reasoning about Collective Play 6. subjectively successful strategies (is ) Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. imperfect information strategies (i) Success of strategies: Objectively (io ) vs. Comparing Variants of Ability Semantic Variants of ATL Memory of agents: Perfect recall (R) vs. Germany 41/104 .1.

Comparing Variants of Ability Example: Poor Duck Problem q4 q5 look shootL q0 sh o ot . Reasoning about Collective Play 6.1. a R q1 shootL look tR oo sh q2 q3 shot Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 42/104 .

Reasoning about Collective Play 6.1. ATLio R ATLis r ATLio r Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 43/104 . Comparing Variants of Ability Validities in Variants of ATL: Subsumption ATLIR = ATLIr ATLis R .

Ir * ATL ATLio R * ATLis r * ATLio r * Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.1. Comparing Variants of Ability Validities in Variants of ATL*: Subsumption ATLIR * ATLis R * . Reasoning about Collective Play 6. Germany 44/104 .

Germany 45/104 . Comparing Variants of Ability Summary in Plain Words In terms of general properties of games we get the following: Perfect information is a (strict) special case of imperfect information Perfect recall games are (strict) special case of memoryless games Information type has more impact than type of recall Properties of objective and subjective abilities of agents are incomparable ATL* allows for expressing more subtle distinctions than plain ATL Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Reasoning about Collective Play 6.1.

.7 Interesting Stuff: Invalidities Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Interesting Stuff: Invalidities 1. . Reasoning about Collective Play 7. 1. Germany 46/104 .

1. Reasoning about Collective Play

7. Interesting Stuff: Invalidities

Some (In)Validities

⟨⟨a⟩⟩3 p ↔ p ∨ ⟨⟨a⟩⟩ j ⟨⟨a⟩⟩3 p Invalid in all variants with imperfect information

Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies

ESSLLI 2013 @ Duesseldorf, Germany 47/104

1. Reasoning about Collective Play

7. Interesting Stuff: Invalidities

Some (In)Validities

⟨⟨a⟩⟩3 p ↔ p ∨ ⟨⟨a⟩⟩ j ⟨⟨a⟩⟩3 p Invalid in all variants with imperfect information ⟨⟨a⟩⟩(3 p1 ∧ 3 p2 ) ↔ ⟨⟨a⟩⟩3 (p1 ∧ ⟨⟨a⟩⟩3 p2 ∨ p2 ∧ ⟨⟨a⟩⟩3 p1 ) Invalid for imperfect recall

Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies

ESSLLI 2013 @ Duesseldorf, Germany 47/104

1. Reasoning about Collective Play

7. Interesting Stuff: Invalidities

Some (In)Validities

⟨⟨a⟩⟩3 p ↔ p ∨ ⟨⟨a⟩⟩ j ⟨⟨a⟩⟩3 p Invalid in all variants with imperfect information ⟨⟨a⟩⟩(3 p1 ∧ 3 p2 ) ↔ ⟨⟨a⟩⟩3 (p1 ∧ ⟨⟨a⟩⟩3 p2 ∨ p2 ∧ ⟨⟨a⟩⟩3 p1 ) Invalid for imperfect recall ¬⟨⟨∅⟩⟩3 ¬p ↔ ⟨⟨Agt⟩⟩2 p Invalid for subjective ability

Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies

ESSLLI 2013 @ Duesseldorf, Germany 47/104

Germany 48/104 . a tR o sh ot R q1 shootL look q2 q3 shot Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.1. Reasoning about Collective Play 7. Interesting Stuff: Invalidities Some (In)Validities ⟨⟨a⟩⟩3 p ↔ p ∨ ⟨⟨a⟩⟩ j ⟨⟨a⟩⟩3 p Invalid in all variants with imperfect information q4 q5 look shootL q0 sh oo .

s et q0 ⟨t ry ⟨try0 . −⟩ 1. Germany 49/104 . a − ⟩ ⟨t .− 1 ry ⟩ q1 ⟨try0 .s ⟨− ⟨− .1. . −⟩ q2 open q3 alarm Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Interesting Stuff: Invalidities Some (In)Validities ⟨⟨a⟩⟩3 p ↔ p ∨ ⟨⟨a⟩⟩ j ⟨⟨a⟩⟩3 p Invalid in all variants with imperfect information ′ q0 0⟩ ⟩ et 1 . Reasoning about Collective Play 7.

q0 de liv e r q1 clean q2 delivered Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 50/104 .1. Reasoning about Collective Play 7. Interesting Stuff: Invalidities Some (In)Validities ⟨⟨a⟩⟩(3 p1 ∧ 3 p2 ) ↔ ⟨⟨a⟩⟩3 (p1 ∧ ⟨⟨a⟩⟩3 p2 ∨ p2 ∧ ⟨⟨a⟩⟩3 p1 ) Invalid for imperfect recall cle an .

a q1 kiss q2 angry q3 suspicious Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 51/104 .1. Reasoning about Collective Play 7. Interesting Stuff: Invalidities Some (In)Validities ¬⟨⟨∅⟩⟩3 ¬p ↔ ⟨⟨Agt⟩⟩2 p Invalid for subjective ability kiss no t-k iss q0 not-kiss .

to appear. Interesting Stuff: Invalidities References 1 2 3 R.-Y. Journal of Autonomous Agents and Multi-Agent Systems. Alternating-time logic with imperfect recall. 4 Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. T. Bulling and W. Comparing Variants of Strategic Ability: How Uncertainty and Memory Inﬂuence General Properties of Games. Penczek. 2002. LNCS vol. Schobbens. Germany 52/104 . Alternating-time Temporal Logic. Jamroga. 49:672–713. 7388. 210-263. W. and O. Electronic Notes in Theoretical Computer Science.1. pp. 2012. A. Kupferman. 85(2). Speciﬁcation and veriﬁcation of multi-agent systems. Jamroga and W. Journal of the ACM. 2004. Henzinger. N. Lectures on Logic and Computation. P. Alur. Reasoning about Collective Play 7.

1. this enough to reason about organizations? Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Discussion . Is . Interesting Stuff: Invalidities . Germany 53/104 . Reasoning about Collective Play 7.

Germany 54/104 . . 2.. Interaction and Ontologies Interaction and Ontologies Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

Germany 55/104 .. Models of Interaction 2. . Interaction and Ontologies 1. 2.1 Models of Interaction Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

actions. Interaction and Ontologies 1. Models of Interaction Concrete and Abstract Models of Interaction Game theory: the semantics of interaction is deﬁned in terms of possible behaviors and their outcomes “Concrete” game models: states. and transitions are represented explicitly Abstract models: “distill” an abstract representation of individual and coalitional powers Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.2. Germany 56/104 .

a nonempty set of strategies Σi for each player i ∈ N . Models of Interaction Concrete Models: Strategic Games . Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. o. W ) that consists of a nonempty ﬁnite set of players N . a nonempty set of ∏ outcomes W . Deﬁnition 2. Germany 57/104 . {Σi |i ∈ N }. and an outcome function o : i∈N Σi → W which associates an outcome with every strategy proﬁle.1 (Strategic game) . .2. Interaction and Ontologies 1. A strategic game G is a tuple (N.

Models of Interaction Example: Battle of Sexes 1\2 B S B S Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 58/104 .2. Interaction and Ontologies 1.

1 0. 2 Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.2. Germany 58/104 . 0 1. Models of Interaction Example: Battle of Sexes 1\2 B S B 2. 0 S 0. Interaction and Ontologies 1.

2 We are mainly interested in the outcomes of strategies in terms of behavior of the whole system Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Interaction and Ontologies 1. 0 S 0. 1 0. Models of Interaction Example: Battle of Sexes 1\2 B S B 2.2. Germany 58/104 . 0 1.

Interaction and Ontologies 1. Models of Interaction Example: Battle of Sexes 1\2 B S B w1 w2 S w2 w3 We are mainly interested in the outcomes of strategies in terms of behavior of the whole system Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.2. Germany 58/104 .

wait push.push pos1 We are mainly interested in the outcomes of strategies in terms of behavior of the whole system Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.wait q1 wait.push push. Models of Interaction Example: Battle of Sexes wait.wait push.2. Interaction and Ontologies 1.push 1\2 B S q0 pos0 pu it wa sh.push pu sh .wait push.p wa B w1 w2 S w2 w3 wait. Germany 58/104 .p us h pos2 q2 wait.w ait wa it. h us it.

Deﬁnition 2. An effectivity function is a function E : 2Agt → 22 W that associates a family of sets of states with each set of players. Germany 59/104 . Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Interaction and Ontologies 1.2. Intuitively. Models of Interaction Abstract Models: Coalitional Effectivity Models How can we “distill” the powers of agents and coalitions in the game? .2 (Effectivity function) . . elements of E (C ) are choices available to coalition C : if X ∈ E (C ) then by choosing X the coalition C can force the outcome of the game to be in X .

Models of Interaction Example: Effectivity in Battle of Sexes 1\2 B B w1 S w2 S w2 w3 Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Interaction and Ontologies 1. Germany 60/104 .2.

Models of Interaction Example: Effectivity in Battle of Sexes 1\2 B B w1 S w2 E ({1. 2}) = {{w1 }. {w3 } S w2 w3 Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Interaction and Ontologies 1. {w2 }.2. Germany 60/104 .

w2 . {w1 . 2}) = {{w1 }. w3 }. {w3 }. w3 }} Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. w3 }.2. Germany 60/104 . {w2 . Interaction and Ontologies 1. Models of Interaction Example: Effectivity in Battle of Sexes 1\2 B B w1 S w2 S w2 w3 E ({1. w2 }. {w2 }. {w1 . {w1 .

2. w2 }. w3 }. {w1 . 2}) = {{w1 }. {w2 }. {w1 . w3 } Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Interaction and Ontologies 1. w2 . w3 }} E ({1}) = {{w1 . w2 }. {w1 . {w2 . w3 }. Germany 60/104 . {w3 }. {w2 . Models of Interaction Example: Effectivity in Battle of Sexes 1\2 B B w1 S w2 S w2 w3 E ({1.

Interaction and Ontologies 1. w2 }. w3 }} Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. w2 }. w2 . {w3 }. w3 }.2. 2}) = {{w1 }. w3 }} E ({1}) = {{w1 . {w1 . Models of Interaction Example: Effectivity in Battle of Sexes 1\2 B B w1 S w2 S w2 w3 E ({1. w3 }. w2 . w3 }. {w1 . {w2 }. {w1 . Germany 60/104 . {w2 . {w2 . {w1 .

2}) = {{w1 }. w2 }. w3 }. w2 . Germany 60/104 . {w1 . {w3 }. Models of Interaction Example: Effectivity in Battle of Sexes 1\2 B B w1 S w2 S w2 w3 E ({1. Interaction and Ontologies 1. {w2 . {w1 . w3 }} E ({1}) = {{w1 . w3 }} Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. w2 . w2 }. w3 }. {w2 .2. {w2 . {w1 . {w1 . w3 }. w2 }. w3 }. w3 }} E ({2}) = {{w1 . {w1 . {w2 }. w2 .

w3 }. {w1 . {w2 }. {w2 . w2 }. Models of Interaction Example: Effectivity in Battle of Sexes 1\2 B B w1 S w2 S w2 w3 E ({1. w3 }. {w1 . {w1 . w3 }} E ({1}) = {{w1 . w2 }. w2 . w3 }} E ({2}) = {{w1 . w2 }. w3 }} E (∅) = {{w1 . w2 . w3 }} Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. {w2 . w2 .2. w3 }. {w2 . {w1 . Interaction and Ontologies 1. {w3 }. w3 }. 2}) = {{w1 }. {w1 . w2 . Germany 60/104 .

Exercise . Models of Interaction Abstract Models: Incentives . Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.2. Interaction and Ontologies 1. What mathematical structure would you use to “distill” an abstract view of preferences from utilities? . Germany 61/104 .

E . Deﬁnition 2. Models of Interaction Semantics of ATL in Effectivity Models .3 . Interaction and Ontologies 1. |= ⟨⟨A⟩⟩Φ iff Φ ∈ E (A) Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 62/104 .2.

|= ⟨⟨A⟩⟩Φ iff Φ ∈ E (A) Very neat and elegant! Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.2. Models of Interaction Semantics of ATL in Effectivity Models .3 . E . Deﬁnition 2. Germany 62/104 . Interaction and Ontologies 1.

q |= ⟨⟨A⟩⟩Φ iff EM.2. Interaction and Ontologies 1.q |= ⟨⟨A⟩⟩Φ Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. E . Models of Interaction Semantics of ATL in Effectivity Models . Germany 62/104 . Deﬁnition 2. |= ⟨⟨A⟩⟩Φ iff Φ ∈ E (A) Very neat and elegant! Important: both semantics are equivalent in the sense that: M.3 .

Germany 63/104 .2 Correspondence Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. .. Correspondence 2. Interaction and Ontologies 2. 2.

Correspondence Correspondence Between Concrete and Abstract Models Which effectivity patterns correspond to models of “ordinary” games (i.. our concrete models)? Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Interaction and Ontologies 2.2. Germany 64/104 .e.

Superadditivity: if C ∩ D = ∅. Playable Effectivity Functions Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Deﬁnition 2. Germany 65/104 . Safety: E (C ) ̸= ∅.4 (Playability) . . X ∈ E (C ) and Y ∈ E (D). Liveness: ∅ ∈ / E (C ). Agt-maximality: X ̸∈ E (∅) implies X ∈ E (Agt). then X ∩ Y ∈ E (C ∪ D). An effectivity function E is playable iff the following conditions hold: Outcome monotonicity: X ∈ E (C ) and X ⊆ Y implies Y ∈ E (C ). Correspondence .2. Interaction and Ontologies 2.

Germany 66/104 .2. Interaction and Ontologies 2. Correspondence Playable Effectivity Functions Doesn’t that ring a bell? Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

(S ) ⟨⟨A1 ⟩⟩ j φ1 ∧ ⟨⟨A2 ⟩⟩ j φ2 → ⟨⟨A1 ∪ A2 ⟩⟩ j (φ1 ∧ φ2 ). Germany 66/104 . (⊥) ¬⟨⟨A⟩⟩ j ⊥ (⊤) ⟨⟨A⟩⟩ j ⊤ (Agt) ¬⟨⟨∅⟩⟩ j ¬φ → ⟨⟨Agt⟩⟩ j φ .2. Correspondence Playable Effectivity Functions Doesn’t that ring a bell? . Recall: Next-Time Axioms for ATL . where A1 and A2 are disjoint Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Interaction and Ontologies 2.

Theorem 2. A coalitional effectivity function E corresponds to a strategic game if and only if E is playable. Interaction and Ontologies 2. . Germany 67/104 .5 (Pauly 2001) . Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.2. Correspondence Pauly’s Representation Theorem .

A coalitional effectivity function E corresponds to a strategic game if and only if E is playable. Germany 67/104 . Correspondence Pauly’s Representation Theorem . Interaction and Ontologies 2. . Theorem 2. How to read the result? We characterize the limitations of concrete models Implementability: we characterize which abstract patterns of effectivity can be implemented by concrete models We characterize classes of models for which the semantics of strategic logics if fully equivalent Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.5 (Pauly 2001) .2.

. the result is wrong! Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 67/104 . How to read the result? We characterize the limitations of concrete models Implementability: we characterize which abstract patterns of effectivity can be implemented by concrete models We characterize classes of models for which the semantics of strategic logics if fully equivalent Unfortunately.2. Correspondence Pauly’s Representation Theorem .5 (Pauly 2001) . Theorem 2. A coalitional effectivity function E corresponds to a strategic game if and only if E is playable. Interaction and Ontologies 2.

6 . Interaction and Ontologies 2. = {w ∈ W | w = o(sAgt ) for some strategy proﬁle sAgt }. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. E (∅) is the principal ﬁlter generated by Z . Correspondence Counterexample to Representation Theorem We start with the following observation: .2. Germany 68/104 . Theorem 2. For every effectivity function E of a strategic game.

Correspondence Example: Effectivity in Battle of Sexes 1\2 B B w1 S w2 S w2 w3 E (∅) = {{w1 .2. w2 . Germany 69/104 . Interaction and Ontologies 2. w3 }} Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

Interaction and Ontologies 2. Germany 70/104 . Correspondence Counterexample to Representation Theorem Agt W E (Agt) E (∅) = = = = {a } N {X | X is inﬁnite} {X | X is ﬁnite} Check! Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.2.

Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. An effectivity function E is truly playable iff the following conditions hold: Outcome monotonicity: X ∈ E (C ) and X ⊆ Y implies Y ∈ E (C ). Determinacy: if X ∈ E (N ) then {w} ∈ E (N ) for some w ∈ X. . X ∈ E (C ) and Y ∈ E (D). Safety: E (C ) ̸= ∅. Correspondence Correct Correspondence . Interaction and Ontologies 2. then X ∩ Y ∈ E (C ∪ D). Deﬁnition 2.2. Germany 71/104 . Liveness: ∅ ∈ / E (C ). Agt-maximality: X ̸∈ E (∅) implies X ∈ E (Agt). Superadditivity: if C ∩ D = ∅.7 (True playability) .

A coalitional effectivity function E corresponds to a strategic game if and only if E is truly playable. Theorem 2.8 (Goranko. Interaction and Ontologies 2. Germany 72/104 . . Correspondence Correct Correspondence . Jamroga and Turrini 2011) .2. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

Interaction and Ontologies 2. Theorem 2. and E (∅) is a principal ﬁlter. Correspondence Alternative Characterizations . Germany 73/104 .2. generated by the singleton . E is truly playable iff it is playable and: 1 E (Agt) is a crown (Determinacy) 2 E (∅) has a complete nonmonotonic core 3 E (∅) has a non-empty nonmonotonic core 4 The nonmonotonic core of E (∅) is a singleton.9 . Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

Interaction and Ontologies 2. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 74/104 . .10 . In ﬁnite domains playability implies true playability.2. Correspondence How Big Was the Damage? . Theorem 2.

Germany 74/104 . Theorem 2. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.2. Correspondence How Big Was the Damage? . In consequence. Interaction and Ontologies 2. In ﬁnite domains playability implies true playability.10 . Pauly’s characterization is correct for effectivity functions over ﬁnite sets of outcomes. .

Theorem 2. . In consequence.10 . Pauly’s characterization is correct for effectivity functions over ﬁnite sets of outcomes. Interaction and Ontologies 2. In ﬁnite domains playability implies true playability. Correspondence How Big Was the Damage? . Germany 74/104 . No disastrous consequences for axiomatizations as long as the logic has the ﬁnite model property (basic ATL has) Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.2. .

No disastrous consequences for axiomatizations as long as the logic has the ﬁnite model property (basic ATL has) And anyway. who cares about inﬁnite domains? Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Correspondence How Big Was the Damage? . .10 .2. In consequence. Interaction and Ontologies 2. Germany 74/104 . Theorem 2. . Pauly’s characterization is correct for effectivity functions over ﬁnite sets of outcomes. In ﬁnite domains playability implies true playability.

Interaction and Ontologies 3. 2. Concepts and Coalitions 2.3 Concepts and Coalitions Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. .. Germany 75/104 .

Germany 76/104 . Concepts and Coalitions Bringing Ontologies and Strategies Together Background: ATL: probably the most important logic for MAS DL: ﬁrst-order reasoning with conceptual and computational restraint Idea of the work: ATL goes ﬁrst order (with restraint) DL goes strategic More expressive speciﬁcation language . Interaction and Ontologies 3..2. for security speciﬁcations Interesting consequences for reasoning about MAS Nice game theory connection for decision problems Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf..g.useful e.

Concepts and Coalitions First-Order Language: Description Logic ALC Description Logics: languages to talk about individuals. and roles (binary relations) Decidable Many optimization techniques and existing tools Techniques for combinations of modal logic + DL Close to ontologies: OWL. Interaction and Ontologies 3. Semantic Web Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.2. Germany 77/104 . concepts (unary relations).

Concepts and Coalitions Description Logic Syntax: ⊤. C ⊑ D.2. If C is a concept and R is a role name then ∀R. and every concept name is a concept. ⊥ are concepts. Formulae: C (i). C ⊓ D.C are concepts.C and ∃R. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Germany 78/104 . Interaction and Ontologies 3. plus Boolean combinations. and C ⊔ D are also concepts. If C and D are concepts then ¬C .

Germany 79/104 . Interaction and Ontologies 3.2. Concepts and Coalitions Description Logic Example formulae: Secure(c1) Inactive ⊑ Secure (¬Secure) ⊑ (∃attackedBy.⊤) Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

⊤) Semantics: concepts and formulae are interpreted over the domain of interpretation ∆ Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.2. Concepts and Coalitions Description Logic Example formulae: Secure(c1) Inactive ⊑ Secure (¬Secure) ⊑ (∃attackedBy. Interaction and Ontologies 3. Germany 79/104 .

Interaction and Ontologies 4. Combining Concepts and Strategies 2. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.. Germany 80/104 .4 Combining Concepts and Strategies . 2.

Interaction and Ontologies 4. Germany 81/104 . Combining Concepts and Strategies Alternating-time Description Logic We allow for modal concepts and modal sentences. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.2.

⟨⟨clown⟩⟩XHappy Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Combining Concepts and Strategies Alternating-time Description Logic We allow for modal concepts and modal sentences. Germany 81/104 . Interaction and Ontologies 4.2.

Combining Concepts and Strategies Alternating-time Description Logic We allow for modal concepts and modal sentences.2. Germany 81/104 . Interaction and Ontologies 4. Audience ⟨⟨clown⟩⟩XHappy Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

Combining Concepts and Strategies Alternating-time Description Logic We allow for modal concepts and modal sentences. Interaction and Ontologies 4. Germany 81/104 .2. Audience ⊑ ⟨⟨clown⟩⟩XHappy Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

Interaction and Ontologies 4. ⟨⟨ circusOwner⟩⟩F (Audience ⊑ ⟨⟨clown⟩⟩XHappy ) Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Combining Concepts and Strategies Alternating-time Description Logic We allow for modal concepts and modal sentences.2. Germany 81/104 .

Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Concepts: C ::= ⊤ | C0 | {|i|} | ¬C | C ⊓ C | ∃R. Interaction and Ontologies 4. Combining Concepts and Strategies ADL: Syntax We allow for modal concepts and modal sentences.2.C | ⟨⟨C ⟩⟩XC | ⟨⟨C ⟩⟩GC | ⟨⟨C ⟩⟩C U C. Germany 82/104 . Formulae: φ ::= C ⊑ C | ¬φ | φ ∧ φ | ⟨⟨ C⟩⟩Xφ | ⟨⟨ C⟩⟩Gφ | ⟨⟨ C⟩⟩φ U φ.

C | ⟨⟨C ⟩⟩XC | ⟨⟨C ⟩⟩GC | ⟨⟨C ⟩⟩C U C. Concepts: C ::= ⊤ | C0 | {|i|} | ¬C | C ⊓ C | ∃R. Models: (possibly different) terminological interpretations embedded in every state Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.2. Germany 82/104 . Formulae: φ ::= C ⊑ C | ¬φ | φ ∧ φ | ⟨⟨ C⟩⟩Xφ | ⟨⟨ C⟩⟩Gφ | ⟨⟨ C⟩⟩φ U φ. Combining Concepts and Strategies ADL: Syntax We allow for modal concepts and modal sentences. Interaction and Ontologies 4.

d. Interaction and Ontologies 4. A concurrent game structures with terminological interpretation is a tuple M = ⟨∆.11 (Models of ADL) . St. Deﬁnition 2. active. o⟩ is a concurrent game frame. Germany 83/104 . · ⟩ is a (variable) terminological interpretation 3 active : St → F in(∆) \ ∅ deﬁnes the ﬁnite nonempty set of active agents at each state in St . Act. o. Combining Concepts and Strategies ADL: Models . d. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. St. Act. I (q ) 2 ⟨∆.2. · I (q) ⟩ where: 1 ⟨∆.

no p p q12 grab.pos2 } On ={(cheese.table)} Hungry =∅ In =∅ up 2 .2.n o u ′ .pos2 } In =∅ active = {pos1 .table)} Hungry ={pos1 } In =∅ .nop up2 . Combining Concepts and Strategies Example: Dining Possums . pos2 } Possums ={pos1 . ∆ = {pos1 . nop.up1 Possums ={pos1 .table)} Hungry ={pos1 . table} active = {pos1 . Germany 84/104 .up1 active = {pos1 } Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. cheese.nop no p.no p nop. q0 p2 .nop q0 dn . pos2 } Possums ={pos1 . pos2 .pos2 } On ={(cheese.nop up2 . Interaction and Ontologies 4.pos2 } On ={(cheese.

Interaction and Ontologies 4. ˆ C such that for (⟨⟨C ⟩⟩D1 U D2 )I (q) = {x ∈ ∆ | there is sC ∈ Σ every λ ∈ out(q. sC ) there is i such that x ∈ (D2 )I (λ[i]) . i − 1 we have x ∈ (D1 )I (λ[j ]) }. we have x ∈ (D)I (λ[i]) }. . sC ) we have x ∈ (D)I (λ[1]) }. . and for every j = 0. . Combining Concepts and Strategies ADL: Semantics of Concepts ˆC (⟨⟨C ⟩⟩XD)I (q) = {x ∈ ∆ | there is a joint strategy sC ∈ Σ such that for every λ ∈ out(q. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. . . ˆ C such that for every (⟨⟨C ⟩⟩GD)I (q) = {x ∈ ∆ | there is sC ∈ Σ λ ∈ out(q. . sC ) and i = 0. Germany 85/104 . .2. 1.

for each M. q |= φ ∧ ψ iff M. q ̸|= φ. we have M. q |= ¬φ iff M. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. we have M. sC ). λ[i] |= φ for every i ≥ 0. M. Germany 86/104 . Combining Concepts and Strategies ADL: Semantics of Formulae M. M. Interaction and Ontologies 4. q |= ⟨⟨ C⟩⟩φ U ψ iff there is sC ∈ Σ λ ∈ out(q.2. q |= φ and M. q |= ⟨⟨ C⟩⟩Gφ iff there is sC ∈ Σ λ ∈ out(q. q |= ⟨⟨C ⟩⟩Xφ iff there is a joint strategy sC ∈ Σ for each path λ ∈ out(q. λ[1] |= φ. λ[i] |= ψ . M. sC ). q |= ψ . for each M. λ[j ] |= φ for every 0 ≤ j < i. there is i ≥ 0 for which M. ˆ C such that. q |= C ⊑ D iff C I (q) ⊆ DI (q) . ˆ C such that. ˆ C such that. sC ). and M.

Germany 87/104 . Type of coalitional strategies had to be redeﬁned . Interaction and Ontologies 4. Can have interesting consequences for coalitional game theory Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.2. Combining Concepts and Strategies Interesting Features Coalitions are simply concepts Finitely many individuals active at a moment . Very natural interpretation of teams . Agents = active individuals Membership in a coalition can vary .

Germany 88/104 . . ) ⟩⟩G (⟨⟨admin⟩⟩F Perm ) = ⊤ ∧ (⟨⟨admin⟩⟩F ¬Perm ) = ⊤ ⟨⟨ ⊥⟩ speciﬁes that the administrator can grant and deny the permission to any agent. and In for the set of agents who are inside the building. ⟩⟩G(¬In ⊓ ⟨⟨⊤⟩⟩X In ⊑ Perm ) ⟨⟨ ⊥⟩ says that agents who enter the building are only ones that have permission to do so. Let Perm stand for the set of permissions to enter the building.12 (Authorization) .2. . Moreover. Formula ( . Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Combining Concepts and Strategies Examples: Security . Interaction and Ontologies 4. Example 2.

Hasbob stand for the set of cryptographic objects possessed by Alice and Bob. Interaction and Ontologies 4. Then.13 (Public key cryptography) . respectively. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Let Hasalice . Example 2. formula ⟨⟨ bob⟩⟩FHasbob (skalice ) ⟨⟨ alice⟩⟩GHasalice (skalice ) ∧ ¬⟨ speciﬁes that Alice can maintain her secret key. Combining Concepts and Strategies Examples: Security . while Bob cannot get hold of it. . Germany 89/104 .2.

Alternatively.⟨ alice⟩⟩G(Has -skalice = {|alice|}). Then.2. formula ⟨⟨ bob⟩⟩FHasbob (skalice ) ⟨⟨ alice⟩⟩GHasalice (skalice ) ∧ ¬⟨ speciﬁes that Alice can maintain her secret key. Hasbob stand for the set of cryptographic objects possessed by Alice and Bob. ⟨ . while Bob cannot get hold of it. Germany 89/104 . Combining Concepts and Strategies Examples: Security . Interaction and Ontologies 4.13 (Public key cryptography) . let Has -skalice refer to the set of individuals that possess the secret key of Alice: . Let Hasalice . Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Example 2. respectively.

Interaction and Ontologies 5. Formal Results 2.5 Formal Results Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. 2.. Germany 90/104 . .

2.14 . Theorem 2. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. and CDL . Interaction and Ontologies 5. ALCO. Formal Results Expressivity . ADL is strictly more expressive than ATL. Germany 91/104 .

Interaction and Ontologies 5.2.15 . and reasonably small . ADL model checking is P-complete The usual catch: model must be given. Formal Results Decision Problems: Model Checking . Germany 92/104 . Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Theorem 2. ﬁnite.

Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Formal Results Decision Problems: Satisﬁability .17 . . ADL satisﬁability is undecidable . ADL does not have the ﬁnite model property . Interaction and Ontologies 5. Conjecture 2. Theorem 2. Germany 93/104 .2.16 .

Interaction and Ontologies 5. Formal Results Decision Problems: Anything in Between? Typical decision problem for the DL community: satisﬁability Typical decision problem for the MAS logics community: model checking Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.2. Germany 94/104 .

Germany 94/104 . Interaction and Ontologies 5.2. Formal Results Decision Problems: Anything in Between? Typical decision problem for the DL community: satisﬁability Typical decision problem for the MAS logics community: model checking What about combining the two? Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

2. Formal Results Decision Problems: Anything in Between? Typical decision problem for the DL community: satisﬁability Typical decision problem for the MAS logics community: model checking What about combining the two? . Germany 94/104 . Interaction and Ontologies 5. realizability Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

2. Germany 95/104 . ⟩⟩G(⟨⟨admin⟩⟩F Perm = ⊤) Example: F ⊕? |= ⟨⟨ ⊥⟩ . Formal Results Frame Satisﬁability (≈ Mechanism Design) . Interaction and Ontologies 5. answers whether there exists a model M extending F such that M. Deﬁnition 2. a state q in it. given a concurrent action frame F . . Frame satisﬁability is the decision problem which. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. and a formula φ.18 (Frame satisﬁability) . q |= φ.

Germany 95/104 . given a concurrent action frame F .19 . Deﬁnition 2.18 (Frame satisﬁability) . answers whether there exists a model M extending F such that M. Frame satisﬁability for ADL is decidable Proof by reduction to satisﬁability of the “nexttime” fragment of ADL where the sets of concepts occurring inside and ⟩⟩. . Frame satisﬁability is the decision problem which. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. a state q in it. q |= φ. ⟨⟨·⟩⟩ are disjoint outside ⟨⟨ ·⟩ .2. ⟩⟩G(⟨⟨admin⟩⟩F Perm = ⊤) Example: F ⊕? |= ⟨⟨ ⊥⟩ . and a formula φ. Theorem 2. Formal Results Frame Satisﬁability (≈ Mechanism Design) . Interaction and Ontologies 5. .

Open Issues 2.6 Open Issues Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Interaction and Ontologies 6.. . 2. Germany 96/104 .

Open Issues Open Issues: Technical Prove or disprove undecidability of the sat problem Look for decidable fragments (singleton coalitions?) Complexity of frame satisﬁability (known to be between EXPTIME and 2NEXPTIME) Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Interaction and Ontologies 6.2. Germany 97/104 .

Interaction and Ontologies 6. Germany 98/104 .2. Open Issues Open Issues: Conceptual What do we get with the higher expressivity? what kind of speciﬁcations? Worth pursuing: natural interpretation of coalitions Speciﬁcation and veriﬁcation of institutional scenarios (variable coalitions) Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

Germany 98/104 . Interaction and Ontologies 6.2. Open Issues Open Issues: Conceptual What do we get with the higher expressivity? what kind of speciﬁcations? Worth pursuing: natural interpretation of coalitions Speciﬁcation and veriﬁcation of institutional scenarios (variable coalitions) Variable coalitions in cooperative game theory Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf.

2. Open Issues Open Issues: Conceptual What do we get with the higher expressivity? what kind of speciﬁcations? Worth pursuing: natural interpretation of coalitions Speciﬁcation and veriﬁcation of institutional scenarios (variable coalitions) Variable coalitions in cooperative game theory Realizability as the logical correspondent of: (a) mechanism design. Germany 98/104 . (b) coalition formation (depends on which concepts we leave open for ﬁxing) Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Interaction and Ontologies 6.

Interaction and Ontologies 6.. Germany 99/104 .?) Properties of group knowledge in evolving groups Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Open Issues Open Issues: Conceptual Imperfect information: Adding the epistemic angle Uniformity of strategies Problems with variable coalitions (do the members know who is in the coalition.2..

2.. Interaction and Ontologies 6. Open Issues Open Issues: Conceptual Imperfect information: Adding the epistemic angle Uniformity of strategies Problems with variable coalitions (do the members know who is in the coalition. Germany 99/104 .?) Properties of group knowledge in evolving groups Security: knowledge of objects (such as cryptographic keys) ⟨⟨ bob⟩⟩FHasbob (skalice ) ⟨⟨ alice⟩⟩GHasalice (skalice ) ∧ ¬⟨ Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf..

?) Properties of group knowledge in evolving groups Security: knowledge of objects (such as cryptographic keys) ⟨⟨ bob⟩⟩FKnowsbob (skalice ) ⟨⟨ alice⟩⟩GKnowsalice (skalice ) ∧ ¬⟨ Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Interaction and Ontologies 6. Germany 99/104 ..2. Open Issues Open Issues: Conceptual Imperfect information: Adding the epistemic angle Uniformity of strategies Problems with variable coalitions (do the members know who is in the coalition..

Germany 99/104 . Realizability for "hybrid" reasoning about knowledge Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Interaction and Ontologies 6..?) Properties of group knowledge in evolving groups Security: knowledge of objects (such as cryptographic keys) ⟨⟨ bob⟩⟩FKnowsbob (skalice ) ⟨⟨ alice⟩⟩GKnowsalice (skalice ) ∧ ¬⟨ .. Open Issues Open Issues: Conceptual Imperfect information: Adding the epistemic angle Uniformity of strategies Problems with variable coalitions (do the members know who is in the coalition.2.

Cambridge Univ. 2003. and P. Press. V. Nardi. Interaction and Ontologies 6. Goranko. Effectivity functions. Peleg. 2013. Baader. 288-314. Proceedings of ECAI 2012. University of Amsterdam. and P. PhD thesis. eds. 2001. Logic for Social Software. pp. Concepts. McGuinness. 15(1). pp. D. Wojciech Jamroga. game forms. 2012.F. W. 67-80. and Applications. 26(2). Calvanese. Open Issues References 1 2 3 4 5 B. and Coalitions in Alternating Time. Patel-Schneider. Journal of Autonomous Agents and Multi-Agent Systems. Turrini. Social Choice and Welfare. Agents. D. games.2. pp. F. and rights. Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. D. The Description Logic Handbook: Theory. Strategic games and truly playable effectivity functions. Implementation. 438-443. Pauly. Jamroga. Germany 100/10 .L. M.

this enough for semantic alignment? Wojtek Jamroga and Leon van der Torre · Reasoning for Agreement Technologies ESSLLI 2013 @ Duesseldorf. Is . Open Issues . Discussion . Germany 101/10 . Interaction and Ontologies 6.2.

- History Essay
- FISCHER Enthusiasm
- mushroom magazine 10/2010
- Christoph Keller Verbal Nonverbal
- Public Speaking Outline
- Ruinas de Berlin
- Bagchi, Wells. Varieties of Mathematical Prose (1998)(22s)
- Switzerland
- (7) Capital Adequacy by Anujit Mitra.ppt
- ee-assessment-sheet (Repaired)
- Untitled
- Vedic Mythology 1000011594
- Capturing the Audience (Part I)
- europa
- Conversational Implicature
- Praetorius - Magnificat Cantor, SATTB + organ 1st 3 sets(full score with organ)
- Introductions and Conclusions
- CLOCKSS Adds Its 12th Archive Node at Humboldt University
- ProMatic P Garage Door Operator
- Art on my Mind_ThePoeticsofSoul
- Idiom a Ti City
- Formal Semantics
- Fronius TP 2000
- Written Communication
- 001208071-An-01-Ml-funk Uhr Mit Datum Und Tempe de en Fr Nl
- the master key sistem
- Brazil, D (1995) a Grammar of Speech - Introduction
- Construir no Construído
- 02 World History BLMs
- Manual Car Amplifier Helix A4 Competition

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue listening from where you left off, or restart the preview.

scribd