Dynamic epistemic logic Tristan Charrier Franois Schwarzentruber - - PowerPoint PPT Presentation

dynamic epistemic logic
SMART_READER_LITE
LIVE PREVIEW

Dynamic epistemic logic Tristan Charrier Franois Schwarzentruber - - PowerPoint PPT Presentation

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Dynamic epistemic logic Tristan Charrier Franois Schwarzentruber cole Normale Suprieure Rennes May 13, 2019 1 / 80


slide-1
SLIDE 1

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning

Dynamic epistemic logic

Tristan Charrier François Schwarzentruber

École Normale Supérieure Rennes

May 13, 2019

1 / 80

slide-2
SLIDE 2

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning In the verification/model checking community In philosophy / AI Syntactic specifications

Outline

1

Discussion about modeling actions In the verification/model checking community In philosophy / AI Syntactic specifications

2

Formal definition of event models

3

Model checking

4

Theorem proving

5

Epistemic planning

2 / 80

slide-3
SLIDE 3

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning In the verification/model checking community In philosophy / AI Syntactic specifications

Outline

1

Discussion about modeling actions In the verification/model checking community In philosophy / AI Syntactic specifications

2

Formal definition of event models

3

Model checking

4

Theorem proving

5

Epistemic planning

3 / 80

slide-4
SLIDE 4

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning In the verification/model checking community In philosophy / AI Syntactic specifications

In the verification/model checking community

Program p p q r p a b b a a c a c Transition system Action = an edge

4 / 80

slide-5
SLIDE 5

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning In the verification/model checking community In philosophy / AI Syntactic specifications

In the verification/model checking community

Program p p q r p a b b a a c a c Transition system Action = an edge Epistemic = edges

5 / 80

slide-6
SLIDE 6

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning In the verification/model checking community In philosophy / AI Syntactic specifications

Outline

1

Discussion about modeling actions In the verification/model checking community In philosophy / AI Syntactic specifications

2

Formal definition of event models

3

Model checking

4

Theorem proving

5

Epistemic planning

6 / 80

slide-7
SLIDE 7

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning In the verification/model checking community In philosophy / AI Syntactic specifications

In philosophy / AI

The mechanism of actions is important. Public/private announcement Announce ‘She knows you hold 5♦’ Public action play card 5♦ Private action secretly remove card 5♦ Belief revision learn p although believing ¬p

7 / 80

slide-8
SLIDE 8

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning In the verification/model checking community In philosophy / AI Syntactic specifications

Solution: Dynamic epistemic logic

State Action Classical planning has5♦ pre: has5♦ post: has5♦ := false

slide-9
SLIDE 9

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning In the verification/model checking community In philosophy / AI Syntactic specifications

Solution: Dynamic epistemic logic

State Action Classical planning has5♦ pre: has5♦ post: has5♦ := false DEL [Baltag et al. TARK 1998] [van Ditmarsch et al. 2007] = Kripkean models

  • f

classical planning has5♦ not has5♦ pre: has5♦ post: has5♦ := false pre: true post: −

9 / 80

slide-10
SLIDE 10

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning In the verification/model checking community In philosophy / AI Syntactic specifications

Computing the next state: product update

pre: has5♦ post: has5♦ := false pre: true post: − has5♦ not has5♦ not has5♦ has5♦ not has5♦

10 / 80

slide-11
SLIDE 11

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning In the verification/model checking community In philosophy / AI Syntactic specifications

Outline

1

Discussion about modeling actions In the verification/model checking community In philosophy / AI Syntactic specifications

2

Formal definition of event models

3

Model checking

4

Theorem proving

5

Epistemic planning

11 / 80

slide-12
SLIDE 12

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning In the verification/model checking community In philosophy / AI Syntactic specifications

Syntactic specifications

Game description language

agent a sees the game position

[Love et al. 2008] [Thielscher, IJCAI 2017]

Flatland agent a sees agent b

[Balbiani et al., IGPL 2014] [Gasquet, Goranko, _, AAMAS 2014] [Gasquet, Goranko, _, JAAMAS 2016]

Visibility atoms a sees the truth value of p

[Charrier et al. KR 2016]

Paying attention to public announcements BapayAtt(b) → [p!]BaBbp

[Bolander et al. JoLLI 2016]

Asynchronous announcements [p!][reada]Kap

[Knight et al. MS in CS 2019]

Epistemic gossip [callab]Kasecretb

[van Ditmarsch et al., JAL 2017]

12 / 80

slide-13
SLIDE 13

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning In the verification/model checking community In philosophy / AI Syntactic specifications

From DEL to epistemic temporal logics

Syntactic specification Models of dynamic epistemic logic p p q r p a b b a a c a c Epistemic temporal model

13 / 80

slide-14
SLIDE 14

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning In the verification/model checking community In philosophy / AI Syntactic specifications

From DEL to epistemic temporal logics

Syntactic specification Models of dynamic epistemic logic p p q r p a b b a a c a c Epistemic temporal model

Easy to specify Succinct Ad-hoc languages Hand-crafted semantics

14 / 80

slide-15
SLIDE 15

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning In the verification/model checking community In philosophy / AI Syntactic specifications

From DEL to epistemic temporal logics

Syntactic specification Models of dynamic epistemic logic p p q r p a b b a a c a c Epistemic temporal model

Elegant Kripkean extension of classical planning Succinct Classification in terms of action types Has probabilistic extension Has extensions that encompass belief revision Perfect-recall only Synchronous only

15 / 80

slide-16
SLIDE 16

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning In the verification/model checking community In philosophy / AI Syntactic specifications

From DEL to epistemic temporal logics

Syntactic specification Models of dynamic epistemic logic p p q r p a b b a a c a c Epistemic temporal model

Elegant Allows for async/no perfect-recall semantics Type of actions lost Not Succinct (usually infinite)

16 / 80

slide-17
SLIDE 17

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning In the verification/model checking community In philosophy / AI Syntactic specifications

Timeline

1918 1930 1940 1950 1960 1970 1980 1990 2000 2010

Logic Verification AI

modal logic epistemic logic dynamic logic DEL Model checking Temporal logics LTL, CTL BDD SAT works! Planning Conformant planning MA-STRIPS dec-POMDP GDL Belief revision ETL Strategic reasoning ATL SL

slide-18
SLIDE 18

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning In the verification/model checking community In philosophy / AI Syntactic specifications

Timeline

1918 1930 1940 1950 1960 1970 1980 1990 2000 2010

Logic Verification AI

modal logic epistemic logic dynamic logic DEL Model checking Temporal logics LTL, CTL BDD SAT works! Planning Conformant planning MA-STRIPS dec-POMDP GDL Belief revision ETL Strategic reasoning ATL SL 18 / 80

slide-19
SLIDE 19

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Examples of actions Definition Effect of actions Dynamic language Expressivity

Outline

1

Discussion about modeling actions

2

Formal definition of event models Examples of actions Definition Effect of actions Dynamic language Expressivity

3

Model checking

4

Theorem proving

5

Epistemic planning

19 / 80

slide-20
SLIDE 20

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Examples of actions Definition Effect of actions Dynamic language Expressivity

Outline

1

Discussion about modeling actions

2

Formal definition of event models Examples of actions Definition Effect of actions Dynamic language Expressivity

3

Model checking

4

Theorem proving

5

Epistemic planning

20 / 80

slide-21
SLIDE 21

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Examples of actions Definition Effect of actions Dynamic language Expressivity

Examples of actions

[baltag1998logic] Example (Public announcement of “p") pre: p post: − a, b Example (Private announcement “p" to a) pre: p post: − pre: true post: − b a a, b

21 / 80

slide-22
SLIDE 22

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Examples of actions Definition Effect of actions Dynamic language Expressivity

Examples of actions

Example (Transfer marble from basket to box)

pre : inBasket post : inBasket := false inBox := true pre : true post : −

b a a, b

22 / 80

slide-23
SLIDE 23

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Examples of actions Definition Effect of actions Dynamic language Expressivity

Outline

1

Discussion about modeling actions

2

Formal definition of event models Examples of actions Definition Effect of actions Dynamic language Expressivity

3

Model checking

4

Theorem proving

5

Epistemic planning

23 / 80

slide-24
SLIDE 24

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Examples of actions Definition Effect of actions Dynamic language Expressivity

Actions

pre: p post: − pre: true post: − b a a, b Definition An event model E = (E, (RE

a )a∈AGT, pre, post) is a tuple where:

E = {e, e′, . . . } is a non-empty finite set of possible events, RE

a ⊆ E × E is an accessibility relation on E for agent a,

pre : E → LEL is a precondition function, post : E × AP → LEL is a postcondition function. A pair (E, e) is called an action, where e represents the actual event of (E, e). A pair (E, E0), for E0 ⊆ E, is a non-deterministic action. The set E0 is the set of triggerable events.

24 / 80

slide-25
SLIDE 25

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Examples of actions Definition Effect of actions Dynamic language Expressivity

Deterministic and non-deterministic actions

Deterministic action = single-pointed event model (E, e) pre: p post: p := q pre: true post: − b a a, b Non-deterministic action = multi-pointed event model pre: true post: p := true pre: true post: p := false pre: true post: − b a b a a, b

25 / 80

slide-26
SLIDE 26

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Examples of actions Definition Effect of actions Dynamic language Expressivity

Public actions

Definition An action is said to be public if the accessibility relations in underlying event model are self-loops. pre: true post: p := true pre: true post: p := false a, b a, b

26 / 80

slide-27
SLIDE 27

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Examples of actions Definition Effect of actions Dynamic language Expressivity

Non-ontic actions

Definition An action is said to be non-ontic if the postconditions are trivial: for all e ∈ E, for all propositions p ∈ AP, post(e, p) = p. pre: p post: − pre: true post: − b a a, b

27 / 80

slide-28
SLIDE 28

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Examples of actions Definition Effect of actions Dynamic language Expressivity

Outline

1

Discussion about modeling actions

2

Formal definition of event models Examples of actions Definition Effect of actions Dynamic language Expressivity

3

Model checking

4

Theorem proving

5

Epistemic planning

28 / 80

slide-29
SLIDE 29

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Examples of actions Definition Effect of actions Dynamic language Expressivity

Effect of a public announcement of ϕ: only keep ϕ-worlds

pre: ϕ post: − a, b

In Hintikka’s World: Try on several examples!

ϕ ¬ϕ announcement of ϕ

29 / 80

slide-30
SLIDE 30

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Examples of actions Definition Effect of actions Dynamic language Expressivity

Example of an update product

a b a b a b a b

a a b b a, b a, b a, b a, b

pre: ma post: − pre: true post: − b a a, b

=

a b

a

a b

a

a b

b b a, b

a b

a a, b

a b

b b b a, b

a b

a b a, b

30 / 80

slide-31
SLIDE 31

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Examples of actions Definition Effect of actions Dynamic language Expressivity

Update product: formal definition

Let M = (W , {Ra}a∈AGT, V ) be an epistemic model and E = (E, (RE

a )a∈AGT, pre, post) be an event model.

Definition The update product of M and E is the epistemic model M ⊗ E = (W ⊗, {R⊗

a }a∈AGT, V ⊗) where:

W ⊗ = {(w, e) ∈ W × E | M, w | = pre(e)}, R⊗

a (w, e) = {(w ′, e′) ∈ W ⊗ | wRaw ′ and eRE a e′},

V ⊗(w, e) = {p ∈ AP | M, w | = post(e)(p)}

31 / 80

slide-32
SLIDE 32

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Examples of actions Definition Effect of actions Dynamic language Expressivity

Pointed update products

Definition The successor state of an epistemic state (M, w) by action (E, e) is (M, w) ⊗ (E, e) =def (M ⊗ E, (w, e)) if M, w | = pre(e), otherwise it is undefined. Notation We write e instead of (E, e); We write the word ‘we’ instead of the pair (w, e); We write M ⊗ En for M ⊗ E ⊗ . . . E, n times. We write we1 . . . en | = ϕ instead of M ⊗ En, we1 . . . en | = ϕ.

32 / 80

slide-33
SLIDE 33

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Examples of actions Definition Effect of actions Dynamic language Expressivity

Outline

1

Discussion about modeling actions

2

Formal definition of event models Examples of actions Definition Effect of actions Dynamic language Expressivity

3

Model checking

4

Theorem proving

5

Epistemic planning

33 / 80

slide-34
SLIDE 34

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Examples of actions Definition Effect of actions Dynamic language Expressivity

Dynamic language

Definition The language LDELCK extends LELCK with dynamic modalities and is defined by the following BNF: ϕ ::= ⊤ | p | ¬ϕ | (ϕ ∨ ϕ) | Kaϕ | CGϕ | E, E0ϕ where E, E0 ranges over the set of non-deterministic actions. Definition We extend the definition M, w | = ϕ to LDELCK with the following clause: M, w | = E, E0ϕ if there exists e ∈ E0 s.th. M, w | = pre(e) and M ⊗ E, (w, e) | = ϕ.

34 / 80

slide-35
SLIDE 35

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Examples of actions Definition Effect of actions Dynamic language Expressivity

Dual operator

We define [E, E0] to be ¬E, E0¬. The semantics is: M, w | = [E, E0]ϕ if for all e ∈ E0 we have M, w | = pre(e) implies M ⊗ E, (w, e) | = ϕ; M, w | = E, E0ϕ if there exists e ∈ E0 s.th. M, w | = pre(e) and M ⊗ E, (w, e) | = ϕ.

35 / 80

slide-36
SLIDE 36

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Examples of actions Definition Effect of actions Dynamic language Expressivity

Outline

1

Discussion about modeling actions

2

Formal definition of event models Examples of actions Definition Effect of actions Dynamic language Expressivity

3

Model checking

4

Theorem proving

5

Epistemic planning

36 / 80

slide-37
SLIDE 37

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Examples of actions Definition Effect of actions Dynamic language Expressivity

Expressivity

Theorem DEL and EL have the same expressivity. Idea: we remove the dynamic operators [E, E]. Let us explain it just with public announcements: [ϕ!]ψ : if ϕ holds then after having announced ϕ publicly, ψ holds. [ϕ!]p says the same thing than (ϕ → p) [ϕ!](ψ ∧ χ) says the same thing than ([ϕ!]ψ ∧ [ϕ!]χ) [ϕ!]¬ψ says the same thing than (ϕ → ¬[ϕ!]ψ) [ϕ!]Kaψ says the same thing than (ϕ → Ka[ϕ!]ψ) [ϕ!][ψ!]χ says the same thing than [ϕ ∧ [ϕ!]ψ!]χ General proof in [Baltag, Moss and Solecki, 2003a] DEL is more succinct: [Lutz, AAMAS 2006]

37 / 80

slide-38
SLIDE 38

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Model checking problem Complexity

Outline

1

Discussion about modeling actions

2

Formal definition of event models

3

Model checking Model checking problem Complexity

4

Theorem proving

5

Epistemic planning

38 / 80

slide-39
SLIDE 39

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Model checking problem Complexity

Outline

1

Discussion about modeling actions

2

Formal definition of event models

3

Model checking Model checking problem Complexity

4

Theorem proving

5

Epistemic planning

39 / 80

slide-40
SLIDE 40

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Model checking problem Complexity

Model checking problem

Definition (model checking problem) Input:

An epistemic state A formula, e.g. action1; action2Kap;

Output: yes if Kap action1 action2 no otherwise.

40 / 80

slide-41
SLIDE 41

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Model checking problem Complexity

Outline

1

Discussion about modeling actions

2

Formal definition of event models

3

Model checking Model checking problem Complexity

4

Theorem proving

5

Epistemic planning

41 / 80

slide-42
SLIDE 42

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Model checking problem Complexity

Model checking complexity

public actions any P-complete [van Benthem, 2011] Pspace-complete [Aucher, _, TARK 2013] [Pol et al. 2016]

42 / 80

slide-43
SLIDE 43

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Model checking problem Complexity

State explosion problem

Example Minesweeper easy 8 × 8 with 10 bombs: > 1012 possible worlds.

43 / 80

slide-44
SLIDE 44

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Model checking problem Complexity

State explosion problem

Example Minesweeper 10 × 12 with 20 bombs: > 1025 possible worlds.

44 / 80

slide-45
SLIDE 45

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Model checking problem Complexity

Solution to the state explosion problem

[DBLP:conf/lori/BenthemEGS15], [DBLP:journals/logcom/BenthemEGS18] [Charrier _ AAMAS 2017], [Charrier _ AiML 2018] Succinct representations of epistemic states and actions; Easy to specify by means of accessibility programs; Succinct model checking still in Pspace.

45 / 80

slide-46
SLIDE 46

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Model checking problem Complexity

Impact

Theoretical Existence of a (uniform) strategy in bounded imperfect info games is in Pspace. Implementation: Pspace techniques Symbolic Model checking implemented in Hintikka’s World: by Sébastien Gamblin and Alexandre Niveau (univ. Caen) using BDDs (C wrapper of CUDD compiled in wasm).

46 / 80

slide-47
SLIDE 47

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning

Outline

1

Discussion about modeling actions

2

Formal definition of event models

3

Model checking

4

Theorem proving

5

Epistemic planning

47 / 80

slide-48
SLIDE 48

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning

Theorem proving

Motivation: parametrized verification

for all epistemic states in which p holds:

p Kap action1 action2 p → action1; action2Kap is a theorem, i.e. true in all epistemic states. Definition Input: a formula ϕ; Output: yes if ϕ is a theorem, no otherwise.

48 / 80

slide-49
SLIDE 49

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning

Theorem proving is highly intractable

EL Pspace-c [Ladner 1977], [Halpern, Moses, 1992] DEL coNExptime-c [Aucher and _, 2015] ELCK Exptime-c [Halpern, Moses, 1992] DELCK 2Exptime-c [Charrier and _, AiML 2018] + actions + common knowledge + common knowledge + actions

Semi-product modal logics have high complexities; [Gabbay et al. Many-Dimensional Modal Logics: Theory and Applications, 2003] Model checking more practical than theorem proving [Halper, Vardi, KR 1991].

49 / 80

slide-50
SLIDE 50

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Outline

1

Discussion about modeling actions

2

Formal definition of event models

3

Model checking

4

Theorem proving

5

Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

50 / 80

slide-51
SLIDE 51

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Epistemic planning

[Andersen, Bolander, 2011] Epistemic planning initial state repertoire of actions goal yes if

(described in Dynamic epistemic logic)

[Baltag et al. 1998] [van Ditmarsch et al. 2007]

initial state final state satisfying the goal plan

51 / 80

slide-52
SLIDE 52

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Decidability and undecidability of epistemic planning

no postconditions: Boolean postconditions: modal depths of preconditions 1 2 3 ?

52 / 80

slide-53
SLIDE 53

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Decidability and undecidability of epistemic planning

no postconditions: Boolean postconditions: modal depths of preconditions 1 2 3 ? KaKbKap

1 2 3

53 / 80

slide-54
SLIDE 54

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Decidability and undecidability of epistemic planning

no postconditions: Boolean postconditions: modal depths of preconditions 1 2 3 ?

54 / 80

slide-55
SLIDE 55

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Decidability and undecidability of epistemic planning

no postconditions: Boolean postconditions: modal depths of preconditions 1 2 3 ? [Bolander et al. IJCAI 2015] [Charrier et al. IJCAI 2016] [Aucher Bolander IJCAI 2013] [Charrier et al. IJCAI 2016] [Yu, Wen and Liu, IJCAI 2013] [Aucher et al., SR2014] [Douéneau-Tabot et al., AiML 2018] [Andersen, Bolander, 2011] [Lê Cong et al., IJCAI 2018]

55 / 80

slide-56
SLIDE 56

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Outline

1

Discussion about modeling actions

2

Formal definition of event models

3

Model checking

4

Theorem proving

5

Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

56 / 80

slide-57
SLIDE 57

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Epistemic planning is undecidable

Theorem Epistemic planning is undecidable for:

[Andersen, Bolander, JANCL 2011]

  • two

agents + Boolean post

+

md(pre) ≤ 1

+

fixed repertoire

  • f one action

+

6 atomic propositions

  • [Lê Cong, Pinchinat, _, IJCAI-ECAI 2018]

Proof: reduction from halting problem of a small universal cellular automaton.

57 / 80

slide-58
SLIDE 58

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Example: the 110 Rule cellular automaton

. . . 1 1 1 1 1 1 . . .

time

Rules

1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

58 / 80

slide-59
SLIDE 59

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Example: the 110 Rule cellular automaton

. . . 1 1 1 1 1 1 . . . . . . 1 1 1 1 1 1 . . .

time

Rules

1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

59 / 80

slide-60
SLIDE 60

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Example: the 110 Rule cellular automaton

. . . 1 1 1 1 1 1 . . . . . . 1 1 1 1 1 1 . . . . . . 1 1 1 1 1 1 . . .

time

Rules

1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

60 / 80

slide-61
SLIDE 61

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Example: the 110 Rule cellular automaton

. . . 1 1 1 1 1 1 . . . . . . 1 1 1 1 1 1 . . . . . . 1 1 1 1 1 1 . . . . . . 1 1 1 1 1 1 1 1 . . .

time

Rules

1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

61 / 80

slide-62
SLIDE 62

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Example: the 110 Rule cellular automaton

. . . 1 1 1 1 1 1 . . . . . . 1 1 1 1 1 1 . . . . . . 1 1 1 1 1 1 . . . . . . 1 1 1 1 1 1 1 1 . . .

time

Rules

1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

62 / 80

slide-63
SLIDE 63

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Encoding an automaton configuration in a state

· · · 1 1 1 · · · In Hintikka’s World: Cellular automaton

63 / 80

slide-64
SLIDE 64

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Encoding an automaton configuration in a state

· · · 1 1 1 · · · In Hintikka’s World: Cellular automaton

64 / 80

slide-65
SLIDE 65

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Encoding an automaton configuration in a state

· · · 1 1 1 · · · In Hintikka’s World: Cellular automaton

65 / 80

slide-66
SLIDE 66

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Encoding an automaton configuration in a state

· · · 1 1 1 · · · In Hintikka’s World: Cellular automaton

66 / 80

slide-67
SLIDE 67

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Encoding an automaton configuration in a state

· · · 1 1 1 · · · In Hintikka’s World: Cellular automaton

67 / 80

slide-68
SLIDE 68

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Encoding an automaton configuration in a state

· · · 1 1 1 · · · In Hintikka’s World: Cellular automaton

68 / 80

slide-69
SLIDE 69

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Outline

1

Discussion about modeling actions

2

Formal definition of event models

3

Model checking

4

Theorem proving

5

Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

69 / 80

slide-70
SLIDE 70

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

(Infinite) epistemic temporal structures

time

Epistemic planning: first-order query ∃x, goal(x)

70 / 80

slide-71
SLIDE 71

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Decidability when pre/post are Boolean

Theorem ([DBLP:conf/ijcai/YuWL13], [DBLP:journals/corr/AucherMP14]) When pre/post are Boolean, epistemic planning is decidable. Epistemic temporal structures are automatic Epistemic planning is a first-order-query first-order-query on automatic structures is decidable. Theorem ( [Douéneau-Tabot, Pinchinat and _, 2018]) Even decidable for goals in epistemic linear µ-calculus.

71 / 80

slide-72
SLIDE 72

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Automatic structure = defined by automatas

N iseven?

  • ,

, AN Aiseven? A AN enc : N → {1}∗ n → 1n start 1

72 / 80

slide-73
SLIDE 73

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Example of an automatic structure

N, iseven?,

Aiseven?

even start

  • dd

1 1

73 / 80

slide-74
SLIDE 74

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Example of an automatic structure

S = N, iseven?, 2 5 iff “11 11111" 2 5 iff word 1 1 1 1 1 1 1

  • is accepted by A

A

state start sink 1 1

  • ,
  • 1
  • 1
  • 1
  • 74 / 80
slide-75
SLIDE 75

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Outline

1

Discussion about modeling actions

2

Formal definition of event models

3

Model checking

4

Theorem proving

5

Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

75 / 80

slide-76
SLIDE 76

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Strategies

Definition A strategy for player a is a function σ that maps any history we1...en to a deterministic epistemic action in the repertoire of a. Definition A uniform strategy for player a is a strategy σ such that if we1...en ∼a ue′

1...e′ n then

σ(we1...en) = σ(ue′

1..e′ n)

76 / 80

slide-77
SLIDE 77

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Undecidability even for Boolean pre/post

Theorem [Reif, Peterson, 1979] [Coulombe and Lynch, Def. 1, p. 14:7, FUN 2018] [Maubert et al., IJCAI 2019] The existence of uniform strategies for two players against an environment for achieving a goal ϕ is undecidable. Decidability cases public actions [Belardinelli et al., 2017] [Maubert et al., IJCAI 2019] hierarchical information [Maubert et al., 2018] [Maubert et al., IJCAI 2019]

(picture idea from Raphael Berthon)

77 / 80

slide-78
SLIDE 78

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Complexity results on epistemic planning

  • ne centralized

many planner players [Bolander et al. IJCAI 2015] [Maubert et al., 2019] public NP-c PSPACE-c announcements public PSPACE-c EXPTIME-c actions Boolean decidable undecidable pre/post [Reif, Peterson, 1979] all undecidable Uninformed semantics.

78 / 80

slide-79
SLIDE 79

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Perspectives: DEL and formal language theory

Question Is epistemic planning one agent (pre md 1, ✟✟ ✟ post) decidable? First-order query is decidable First-order query is undecidable Automatic structures Turing-complete structures . . . . . . . . . . . . . . . . . . . . . . . . Pushdown automata? Caucal hierarchy?

79 / 80

slide-80
SLIDE 80

Discussion about modeling actions Formal definition of event models Model checking Theorem proving Epistemic planning Undecidability of epistemic planning Decidability when pre/post are Boolean Generalize to multi-player setting

Perspectives

Connection with logics for reasoning about strategies such as Alternating temporal-time logic, Strategy Logic, etc. [Maubert et al., 2019] Describing protocols/policies

80 / 80