Deception: An epistemic planned event? Shikha Singh, Deepak Khemani - - PowerPoint PPT Presentation

▶
deception an epistemic planned event
SMART_READER_LITE
LIVE PREVIEW

Deception: An epistemic planned event? Shikha Singh, Deepak Khemani - - PowerPoint PPT Presentation

Mar 15, 2023 237 likes •1.35k views

Deception: An epistemic planned event? Shikha Singh, Deepak Khemani Department of Computer Science & Engineering IIT Madras Logic and Cognition Pre-Conference Workshop ICLA-2019 Shikha Singh, Deepak Khemani Deception: An epistemic planned

slide-1
SLIDE 1

Deception: An epistemic planned event?

Shikha Singh, Deepak Khemani

Department of Computer Science & Engineering IIT Madras Logic and Cognition Pre-Conference Workshop ICLA-2019

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 1 / 55

slide-2
SLIDE 2

Outline

1

Motivation

2

Lying and Cognition

3

Lies and Deception Defining-Modelling Lies (Existing work) Engineering Lies

4

Epistemic Planning Classical Planning Our approach State-of-the-art: Epistemic Planning

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 2 / 55

slide-3
SLIDE 3

Motivation: The Gruffalo

Examples are better than precept...

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 3 / 55

slide-4
SLIDE 4

Motivation: The Gruffalo

Examples are better than precept...

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 4 / 55

slide-5
SLIDE 5

Motivation: The Gruffalo

Examples are better than precept...

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 5 / 55

slide-6
SLIDE 6

Motivation: The Gruffalo

Examples are better than precept...

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 6 / 55

slide-7
SLIDE 7

Motivation: The Gruffalo

Examples are better than precept...

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 7 / 55

slide-8
SLIDE 8

Motivation: The Gruffalo

Examples are better than precept...

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 8 / 55

slide-9
SLIDE 9

Motivation: The Gruffalo

Examples are better than precept...

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 9 / 55

slide-10
SLIDE 10

Motivation: The Gruffalo

Examples are better than precept...

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 10 / 55

slide-11
SLIDE 11

Motivation: The Gruffalo

Examples are better than precept...

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 11 / 55

slide-12
SLIDE 12

Motivation: The Gruffalo

Examples are better than precept...

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 12 / 55

slide-13
SLIDE 13

Motivation: The Gruffalo

Examples are better than precept...

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 13 / 55

slide-14
SLIDE 14

Motivation: The Gruffalo

Examples are better than precept...

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 14 / 55

slide-15
SLIDE 15

Motivation: The Gruffalo

Examples are better than precept...

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 15 / 55

slide-16
SLIDE 16

Motivation: The Gruffalo

Examples are better than precept...

The Story Ends here...!!

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 16 / 55

slide-17
SLIDE 17

Lying and Understanding Theory of Mind (TOM)

Social and Cognitive Correlates of Childrens Lying Behavior [TL08]1

“Two hypothesis regarding the relationship between theory of mind understanding and lie-telling”

1Victoria Talwar and Kang Lee, Social and cognitive correlates of childrens lying

behavior, Child development 79 (2008), no. 4, 866881.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 17 / 55

slide-18
SLIDE 18

Lying and Understanding Theory of Mind (TOM)

Social and Cognitive Correlates of Childrens Lying Behavior [TL08]1

“Two hypothesis regarding the relationship between theory of mind understanding and lie-telling” First hypothesis (TOM1):

1Victoria Talwar and Kang Lee, Social and cognitive correlates of childrens lying

behavior, Child development 79 (2008), no. 4, 866881.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 17 / 55

slide-19
SLIDE 19

Lying and Understanding Theory of Mind (TOM)

Social and Cognitive Correlates of Childrens Lying Behavior [TL08]1

“Two hypothesis regarding the relationship between theory of mind understanding and lie-telling” First hypothesis (TOM1):

Relationship between children’s lie-telling and their first-order belief understanding: telling a lie requires deliberately creating a false belief in the mind of another.

1Victoria Talwar and Kang Lee, Social and cognitive correlates of childrens lying

behavior, Child development 79 (2008), no. 4, 866881.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 17 / 55

slide-20
SLIDE 20

Lying and Understanding Theory of Mind (TOM)

Social and Cognitive Correlates of Childrens Lying Behavior [TL08]1

“Two hypothesis regarding the relationship between theory of mind understanding and lie-telling” First hypothesis (TOM1):

Relationship between children’s lie-telling and their first-order belief understanding: telling a lie requires deliberately creating a false belief in the mind of another.

Second hypothesis (TOM2):

1Victoria Talwar and Kang Lee, Social and cognitive correlates of childrens lying

behavior, Child development 79 (2008), no. 4, 866881.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 17 / 55

slide-21
SLIDE 21

Lying and Understanding Theory of Mind (TOM)

Social and Cognitive Correlates of Childrens Lying Behavior [TL08]1

“Two hypothesis regarding the relationship between theory of mind understanding and lie-telling” First hypothesis (TOM1):

Relationship between children’s lie-telling and their first-order belief understanding: telling a lie requires deliberately creating a false belief in the mind of another.

Second hypothesis (TOM2):

Relationship between children’s ability to maintain their lies and their second-order belief understanding: maintaining a lie in follow-up questions may be related to a child’s ability to represent others beliefs and what the other will infer from any knowledge revealed by the child.

1Victoria Talwar and Kang Lee, Social and cognitive correlates of childrens lying

behavior, Child development 79 (2008), no. 4, 866881.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 17 / 55

slide-22
SLIDE 22

Lying and cognition

Whether lying requires cognition?

Lying is quite frequent!

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 18 / 55

slide-23
SLIDE 23

Lying and cognition

Whether lying requires cognition?

Lying is quite frequent! But, challenging !!

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 18 / 55

slide-24
SLIDE 24

Lying and cognition

Whether lying requires cognition?

Lying is quite frequent! But, challenging !! It usually takes longer than telling the truth!

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 18 / 55

slide-25
SLIDE 25

Lying and cognition

Whether lying requires cognition?

Lying is quite frequent! But, challenging !! It usually takes longer than telling the truth! Demands a great deal of cognitive resources...

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 18 / 55

slide-26
SLIDE 26

Lying and cognition

Whether lying requires cognition?

Lying is quite frequent! But, challenging !! It usually takes longer than telling the truth! Demands a great deal of cognitive resources... Why?

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 18 / 55

slide-27
SLIDE 27

Lying and cognition

Whether lying requires cognition?

Lying is quite frequent! But, challenging !! It usually takes longer than telling the truth! Demands a great deal of cognitive resources... Why? Because it involves two stages:

The decision to lie The synthesis of the lie

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 18 / 55

slide-28
SLIDE 28

Outline

1

Motivation

2

Lying and Cognition

3

Lies and Deception Defining-Modelling Lies (Existing work) Engineering Lies

4

Epistemic Planning Classical Planning Our approach State-of-the-art: Epistemic Planning

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 19 / 55

slide-29
SLIDE 29

Defining Lies

We investigate the role of intentionionality in lying. A person x deceives another person y if and only if x intentionally causes y to believe p, where p is false or x does not believe that p is true [Car06]2.

2Thomas L Carson, The definition of lying, Nos 40 (2006), no. 2, 284306 Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 20 / 55

slide-30
SLIDE 30

Defining Lies

A detailed account on the definition of lying and deception can be found at plato.stanford.edu/The Definition of Lying and Deception

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 21 / 55

slide-31
SLIDE 31

Modelling Lies

Related Work

Alexandru Baltag[Bal02]3, David Steiner [Ste06]4, Ditmarsch et al.[VDVESW12]5, Hans Van Ditmarsch [VD14]6 use event model to model the act of lying “lying that p” : “communicating p in the belief that ¬p is the case”

3A logic for suspicious players: Epistemic actions and beliefupdates in games,

Bulletin of Economic Research (2002)

4David Steiner, A system for consistency preserving belief change, Proceedings of the

ESSLLI Workshop on Rationality and Knowledge

5Hans Van Ditmarsch, Jan Van Eijck, Floor Sietsma, and Yanjing Wang, On the

logic of lying, Games, actions and social software, Springer, 2012

6Hans Van Ditmarsch, Dynamics of lying, Synthese 191 (2014) Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 22 / 55

slide-32
SLIDE 32

Modelling Lies

Related Work

Alexandru Baltag[Bal02]3, David Steiner [Ste06]4, Ditmarsch et al.[VDVESW12]5, Hans Van Ditmarsch [VD14]6 use event model to model the act of lying “lying that p” : “communicating p in the belief that ¬p is the case” Project lying as an epistemic action that induces a transformation on an epistemic state (pointed Kripke model).

3A logic for suspicious players: Epistemic actions and beliefupdates in games,

Bulletin of Economic Research (2002)

4David Steiner, A system for consistency preserving belief change, Proceedings of the

ESSLLI Workshop on Rationality and Knowledge

5Hans Van Ditmarsch, Jan Van Eijck, Floor Sietsma, and Yanjing Wang, On the

logic of lying, Games, actions and social software, Springer, 2012

6Hans Van Ditmarsch, Dynamics of lying, Synthese 191 (2014) Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 22 / 55

slide-33
SLIDE 33

Modelling Lies

Related Work

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 23 / 55

slide-34
SLIDE 34

Outline

1

Motivation

2

Lying and Cognition

3

Lies and Deception Defining-Modelling Lies (Existing work) Engineering Lies

4

Epistemic Planning Classical Planning Our approach State-of-the-art: Epistemic Planning

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 24 / 55

slide-35
SLIDE 35

Engineering Lies

Questions we are interested in.. The decision to lie

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 25 / 55

slide-36
SLIDE 36

Engineering Lies

Questions we are interested in.. The decision to lie The synthesis of a lie

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 25 / 55

slide-37
SLIDE 37

Engineering Lies

We seek a solution by formulating this problem as an epistemic planning problem...

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 26 / 55

slide-38
SLIDE 38

Engineering Lies

We seek a solution by formulating this problem as an epistemic planning problem... The decision to lie: motivated by a goal The synthesis of a lie: a plan to achieve the goal

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 26 / 55

slide-39
SLIDE 39

Outline

1

Motivation

2

Lying and Cognition

3

Lies and Deception Defining-Modelling Lies (Existing work) Engineering Lies

4

Epistemic Planning Classical Planning Our approach State-of-the-art: Epistemic Planning

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 27 / 55

slide-40
SLIDE 40

Classical Planning

Task of finding a sequence of deterministic actions with known effects such that, when applied in the initial, fully- known state, it results in a state where the goal is satisfied.

Definition

The classical planning model S is defined as the tuple S, s0, SG, A, f , c where:

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 28 / 55

slide-41
SLIDE 41

Classical Planning

Task of finding a sequence of deterministic actions with known effects such that, when applied in the initial, fully- known state, it results in a state where the goal is satisfied.

Definition

The classical planning model S is defined as the tuple S, s0, SG, A, f , c where: S is a finite and discrete set of states

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 28 / 55

slide-42
SLIDE 42

Classical Planning

Task of finding a sequence of deterministic actions with known effects such that, when applied in the initial, fully- known state, it results in a state where the goal is satisfied.

Definition

The classical planning model S is defined as the tuple S, s0, SG, A, f , c where: S is a finite and discrete set of states s0 ∈ S is the known initial state

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 28 / 55

slide-43
SLIDE 43

Classical Planning

Task of finding a sequence of deterministic actions with known effects such that, when applied in the initial, fully- known state, it results in a state where the goal is satisfied.

Definition

The classical planning model S is defined as the tuple S, s0, SG, A, f , c where: S is a finite and discrete set of states s0 ∈ S is the known initial state SG ⊆ S is a non-empty set of goal states

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 28 / 55

slide-44
SLIDE 44

Classical Planning

Task of finding a sequence of deterministic actions with known effects such that, when applied in the initial, fully- known state, it results in a state where the goal is satisfied.

Definition

The classical planning model S is defined as the tuple S, s0, SG, A, f , c where: S is a finite and discrete set of states s0 ∈ S is the known initial state SG ⊆ S is a non-empty set of goal states A(s) ⊆ A the set of actions in A that are applicable in each state s ∈ S,

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 28 / 55

slide-45
SLIDE 45

Classical Planning

Task of finding a sequence of deterministic actions with known effects such that, when applied in the initial, fully- known state, it results in a state where the goal is satisfied.

Definition

The classical planning model S is defined as the tuple S, s0, SG, A, f , c where: S is a finite and discrete set of states s0 ∈ S is the known initial state SG ⊆ S is a non-empty set of goal states A(s) ⊆ A the set of actions in A that are applicable in each state s ∈ S, s0 = f (a, s) is a deterministic transition function which, given a state s ∈ S and an action a ∈ A(s), returns the resulting state s0 c(a, s) is the positive cost of applying action a in the state s

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 28 / 55

slide-46
SLIDE 46

Classical Planning

Representation: STRIPS (Fikes and Nilsson, 1971)[FN71]

Definition

A classical planning problem P in STRIPS is defined as the tuple F, I, A, G where:

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 29 / 55

slide-47
SLIDE 47

Classical Planning

Representation: STRIPS (Fikes and Nilsson, 1971)[FN71]

Definition

A classical planning problem P in STRIPS is defined as the tuple F, I, A, G where: F is the set of fluent symbols in the problem.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 29 / 55

slide-48
SLIDE 48

Classical Planning

Representation: STRIPS (Fikes and Nilsson, 1971)[FN71]

Definition

A classical planning problem P in STRIPS is defined as the tuple F, I, A, G where: F is the set of fluent symbols in the problem. I the set of atoms over F which are true initially,

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 29 / 55

slide-49
SLIDE 49

Classical Planning

Representation: STRIPS (Fikes and Nilsson, 1971)[FN71]

Definition

A classical planning problem P in STRIPS is defined as the tuple F, I, A, G where: F is the set of fluent symbols in the problem. I the set of atoms over F which are true initially, A is the set of actions, where every α ∈ A is a tuple pre(α), add(α), del(α) , where pre(α), add(α), del(α) ⊆ F ,

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 29 / 55

slide-50
SLIDE 50

Classical Planning

Representation: STRIPS (Fikes and Nilsson, 1971)[FN71]

Definition

A classical planning problem P in STRIPS is defined as the tuple F, I, A, G where: F is the set of fluent symbols in the problem. I the set of atoms over F which are true initially, A is the set of actions, where every α ∈ A is a tuple pre(α), add(α), del(α) , where pre(α), add(α), del(α) ⊆ F , G is a set of atoms over F that define the goal

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 29 / 55

slide-51
SLIDE 51

Classical Planning

Solution: A plan

Definition

A plan π for a classical planning model S is a sequence of actions π = [a0, a1, ..., an] such that, when applied to the initial state s0 , it results in a sequence of states [s0, s1, ..., sn] where sn ∈ SG .

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 30 / 55

slide-52
SLIDE 52

Epistemic Planning

Enrichment of automated planning with epistemic notions: Knowledge, Belief.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 31 / 55

slide-53
SLIDE 53

Epistemic Planning

Enrichment of automated planning with epistemic notions: Knowledge, Belief. Single-agent: given an agent’s current state of knowledge, and a desirable state of knowledge, how does it get from one to the other?

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 31 / 55

slide-54
SLIDE 54

Epistemic Planning

Enrichment of automated planning with epistemic notions: Knowledge, Belief. Single-agent: given an agent’s current state of knowledge, and a desirable state of knowledge, how does it get from one to the other? Multi-agent: the current and desirable states of knowledge might also refer to the states of knowledge of other agents, including higher-order knowledge like ensuring that agent A doesn’t get to know that agent B knows P.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 31 / 55

slide-55
SLIDE 55

Epistemic Planning

an epistemic planning domain

Definition

An epistemic planning domain is then defined as a state transition system Σ = (S, A, λ), where S is a set of epistemic states of L(F, AG), A is a finite set of epistemic actions of L(F, AG), and λ(s, a) = s ⊗ a if s ⊗ a is defined and ⊗ is the above operation.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 32 / 55

slide-56
SLIDE 56

Epistemic Planning

an epistemic planning problem

Definition

An epistemic planning problem is a triple (Σ, s0, φg) where Σ = (S, A, λ) is an epistemic planning domain on (F, AG), s0 ∈ S is the initial state, and φg is a formula in L(F, AG). An action sequence a1, ..., an where s0 ⊗ a1 ⊗ ... ⊗ an | = φg is a solution of the problem.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 33 / 55

slide-57
SLIDE 57

Lying and deception

From epistemic planning perspective..

The world of lies...

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 34 / 55

slide-58
SLIDE 58

Lying and deception

From epistemic planning perspective..

The world of lies... Doxastic world: an agent reasoning about lying cannot work with knowledge modality; but with its own belief set.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 34 / 55

slide-59
SLIDE 59

Lying and deception

From epistemic planning perspective..

The world of lies... Doxastic world: an agent reasoning about lying cannot work with knowledge modality; but with its own belief set. Partially observable environment: Uncertainity about the initial situation

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 34 / 55

slide-60
SLIDE 60

Lying and deception

From epistemic planning perspective..

The world of lies... Doxastic world: an agent reasoning about lying cannot work with knowledge modality; but with its own belief set. Partially observable environment: Uncertainity about the initial situation Talkative agents: restricted only to epistemic actions- speech acts (announcements)

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 34 / 55

slide-61
SLIDE 61

Lying and deception

From epistemic planning perspective..

The world of lies... Doxastic world: an agent reasoning about lying cannot work with knowledge modality; but with its own belief set. Partially observable environment: Uncertainity about the initial situation Talkative agents: restricted only to epistemic actions- speech acts (announcements) Non-reliable audience: speech acts have non-deterministic effects

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 34 / 55

slide-62
SLIDE 62

Lying and deception

From epistemic planning perspective..

The world of lies... Can we relax some conditions?

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 35 / 55

slide-63
SLIDE 63

Lying and deception

From epistemic planning perspective..

The world of lies... Can we relax some conditions? Like, non-deterministic effects of the speech act?

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 35 / 55

slide-64
SLIDE 64

Lying and deception

From epistemic planning perspective..

The world of lies... Can we relax some conditions? Like, non-deterministic effects of the speech act? (Dynamics of lying, van Ditmarsch, H. Synthese (2014)) distinguish three addressee perspectives:

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 35 / 55

slide-65
SLIDE 65

Lying and deception

From epistemic planning perspective..

The world of lies... Can we relax some conditions? Like, non-deterministic effects of the speech act? (Dynamics of lying, van Ditmarsch, H. Synthese (2014)) distinguish three addressee perspectives:

(Cred) the credulous agent who believes everything that it is told (even at the price of inconsistency)

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 35 / 55

slide-66
SLIDE 66

Lying and deception

From epistemic planning perspective..

The world of lies... Can we relax some conditions? Like, non-deterministic effects of the speech act? (Dynamics of lying, van Ditmarsch, H. Synthese (2014)) distinguish three addressee perspectives:

(Cred) the credulous agent who believes everything that it is told (even at the price of inconsistency) (Skep) the skeptical agent who only believes what it is told if that is consistent with its current beliefs

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 35 / 55

slide-67
SLIDE 67

Lying and deception

From epistemic planning perspective..

The world of lies... Can we relax some conditions? Like, non-deterministic effects of the speech act? (Dynamics of lying, van Ditmarsch, H. Synthese (2014)) distinguish three addressee perspectives:

(Cred) the credulous agent who believes everything that it is told (even at the price of inconsistency) (Skep) the skeptical agent who only believes what it is told if that is consistent with its current beliefs the belief revising agent who believes everything that it is told by consistently revising its current, possibly conflicting, beliefs

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 35 / 55

slide-68
SLIDE 68

Outline

1

Motivation

2

Lying and Cognition

3

Lies and Deception Defining-Modelling Lies (Existing work) Engineering Lies

4

Epistemic Planning Classical Planning Our approach State-of-the-art: Epistemic Planning

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 36 / 55

slide-69
SLIDE 69

Lying and deception

From epistemic planning perspective...

The world of lies... Doxastic world: an agent reasoning about lying cannot work with knowledge modality; but with its own belief set. Partially observable environment: Uncertainity about the initial situation Talkative agents: restricted only to epistemic actions - speech acts (announcements)

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 37 / 55

slide-70
SLIDE 70

Lying and deception

From epistemic planning perspective...

The world of lies... Doxastic world: an agent reasoning about lying cannot work with knowledge modality; but with its own belief set. Partially observable environment: Uncertainity about the initial situation Talkative agents: restricted only to epistemic actions - speech acts (announcements) Credulous audience

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 37 / 55

slide-71
SLIDE 71

Lying and deception

From epistemic planning perspective...

The world of lies... Doxastic world: an agent reasoning about lying cannot work with knowledge modality; but with its own belief set. Partially observable environment: Uncertainity about the initial situation Credulous audience

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 38 / 55

slide-72
SLIDE 72

Lying and deception

From epistemic planning perspective...

The world of lies... Doxastic world: an agent reasoning about lying cannot work with knowledge modality; but with its own belief set. Partially observable environment: Uncertainity about the initial situation Talkative agents: restricted only to epistemic actions - speech acts (announcements) < −− and we focus here Credulous audience

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 38 / 55

slide-73
SLIDE 73

On synthsizing lies

From epistemic planning perspective

Problem: In a (Adversarial) multi-agent epistemic planning setting...

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 39 / 55

slide-74
SLIDE 74

On synthsizing lies

From epistemic planning perspective

Problem: In a (Adversarial) multi-agent epistemic planning setting... where an agent is a reasoner and a planner,

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 39 / 55

slide-75
SLIDE 75

On synthsizing lies

From epistemic planning perspective

Problem: In a (Adversarial) multi-agent epistemic planning setting... where an agent is a reasoner and a planner, given its initial belief set, a goal condition

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 39 / 55

slide-76
SLIDE 76

On synthsizing lies

From epistemic planning perspective

Problem: In a (Adversarial) multi-agent epistemic planning setting... where an agent is a reasoner and a planner, given its initial belief set, a goal condition and all possible speech acts,

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 39 / 55

slide-77
SLIDE 77

On synthsizing lies

From epistemic planning perspective

Problem: In a (Adversarial) multi-agent epistemic planning setting... where an agent is a reasoner and a planner, given its initial belief set, a goal condition and all possible speech acts, Can it come up with a deceiving plan?

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 39 / 55

slide-78
SLIDE 78

On synthsizing lies

From epistemic planning perspective

Problem: In a (Adversarial) multi-agent epistemic planning setting... where an agent is a reasoner and a planner, given its initial belief set, a goal condition and all possible speech acts, Can it come up with a deceiving plan? Or, said to have lied?

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 39 / 55

slide-79
SLIDE 79

On synthsizing lies

From epistemic planning perspective

Problem: In a (Adversarial) multi-agent epistemic planning setting... where an agent is a reasoner and a planner, given its initial belief set, a goal condition and all possible speech acts, Can it come up with a deceiving plan? Or, said to have lied? Does the agent choose ’what we call’ a deceptive speech act?

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 39 / 55

slide-80
SLIDE 80

On synthsizing lies

From epistemic planning perspective

The one where the mouse deceives the fox... We formulate our problem as: F, AG, A, s0, φg, where

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 40 / 55

slide-81
SLIDE 81

On synthsizing lies

From epistemic planning perspective

The one where the mouse deceives the fox... We formulate our problem as: F, AG, A, s0, φg, where F be the set of grounded fluents, F = {at(X, L, T), dang(X, Y ), threat(X, L, T), (at(X, L, T) ∩ at(Y , L, T) ∩ dang(X, Y )) → threat(Y , L, T) where X, Y ∈ AG

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 40 / 55

slide-82
SLIDE 82

On synthsizing lies

From epistemic planning perspective

The one where the mouse deceives the fox... We formulate our problem as: F, AG, A, s0, φg, where F be the set of grounded fluents, F = {at(X, L, T), dang(X, Y ), threat(X, L, T), (at(X, L, T) ∩ at(Y , L, T) ∩ dang(X, Y )) → threat(Y , L, T) where X, Y ∈ AG AG be the set of agents, AG = {m, f , g}

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 40 / 55

slide-83
SLIDE 83

On synthsizing lies

From epistemic planning perspective

The one where the mouse deceives the fox... We formulate our problem as: F, AG, A, s0, φg, where F be the set of grounded fluents, F = {at(X, L, T), dang(X, Y ), threat(X, L, T), (at(X, L, T) ∩ at(Y , L, T) ∩ dang(X, Y )) → threat(Y , L, T) where X, Y ∈ AG AG be the set of agents, AG = {m, f , g} A is the set of grounded actions, A = {announce(X, φ), quit(X, L, T)} where φ ∈ S and S is a set of epistemic states of L(F, AG) and X ∈ AG

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 40 / 55

slide-84
SLIDE 84

On synthsizing lies

From epistemic planning perspective

The one where the mouse deceives the fox... We formulate our problem as: F, AG, A, s0, φg, where F be the set of grounded fluents, F = {at(X, L, T), dang(X, Y ), threat(X, L, T), (at(X, L, T) ∩ at(Y , L, T) ∩ dang(X, Y )) → threat(Y , L, T) where X, Y ∈ AG AG be the set of agents, AG = {m, f , g} A is the set of grounded actions, A = {announce(X, φ), quit(X, L, T)} where φ ∈ S and S is a set of epistemic states of L(F, AG) and X ∈ AG s0 be the initial state, s0 = {at(f , l, t), at(m, l, t), ¬at(g, l, t), dang(f , m), dang(g, m), dang(g, f )}

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 40 / 55

slide-85
SLIDE 85

On synthsizing lies

From epistemic planning perspective

The one where the mouse deceives the fox... We formulate our problem as: F, AG, A, s0, φg, where F be the set of grounded fluents, F = {at(X, L, T), dang(X, Y ), threat(X, L, T), (at(X, L, T) ∩ at(Y , L, T) ∩ dang(X, Y )) → threat(Y , L, T) where X, Y ∈ AG AG be the set of agents, AG = {m, f , g} A is the set of grounded actions, A = {announce(X, φ), quit(X, L, T)} where φ ∈ S and S is a set of epistemic states of L(F, AG) and X ∈ AG s0 be the initial state, s0 = {at(f , l, t), at(m, l, t), ¬at(g, l, t), dang(f , m), dang(g, m), dang(g, f )} φg is the goal formula, φg = ¬threat(m, l, t′) ∩ t′ > t

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 40 / 55

slide-86
SLIDE 86

On synthsizing lies

From epistemic planning perspective

The action operators are further defined as: announce(X, φ) :

pre: Nil add: B{AG−X}φ del: Nil

quit(X, L, T) : (domain specific action operator)

pre: threat(X, L, T) add: ¬at(X, L, T + 1) del: Nil

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 41 / 55

slide-87
SLIDE 87

On synthsizing lies

From epistemic planning perspective

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 42 / 55

slide-88
SLIDE 88

On synthsizing lies

From epistemic planning perspective

Second order reasoning that the mouse does in S2 (on the fox’s behalf) Bf (at(f , l, t) ∩ at(g, l, t) ∩ dang(g, f ))∩ Bf ((at(X, L, T) ∩ at(Y , L, T) ∩ dang(X, Y )) → threat(Y , L, T)) → Bf threat(f , l, t) mouse believes that fox believes that there is a threat to its life.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 43 / 55

slide-89
SLIDE 89

On synthsizing lies

From epistemic planning perspective

Second order reasoning that the mouse does in S2 (on the fox’s behalf) according to which:

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 44 / 55

slide-90
SLIDE 90

On synthsizing lies

From epistemic planning perspective

Second order reasoning that the mouse does in S2 (on the fox’s behalf) according to which: State S2 entails the preconditions of Quit(f , l, t),

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 44 / 55

slide-91
SLIDE 91

On synthsizing lies

From epistemic planning perspective

Second order reasoning that the mouse does in S2 (on the fox’s behalf) according to which: State S2 entails the preconditions of Quit(f , l, t), hence, Quit(f , l, t) is applicable in it,

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 44 / 55

slide-92
SLIDE 92

On synthsizing lies

From epistemic planning perspective

Second order reasoning that the mouse does in S2 (on the fox’s behalf) according to which: State S2 entails the preconditions of Quit(f , l, t), hence, Quit(f , l, t) is applicable in it, leading to the addition of the post-effect: ¬at(f , l, t + 1), which led the mouse anticipate that the fox would flee away.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 44 / 55

slide-93
SLIDE 93

On synthsizing lies

From epistemic planning perspective

Second order reasoning that the mouse does in S2 (on the fox’s behalf) according to which: State S2 entails the preconditions of Quit(f , l, t), hence, Quit(f , l, t) is applicable in it, leading to the addition of the post-effect: ¬at(f , l, t + 1), which led the mouse anticipate that the fox would flee away. and due to ¬at(f , l, t + 1), threat(m, l, t + 1) wouldn’t hold in S3

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 44 / 55

slide-94
SLIDE 94

On synthsizing lies

From epistemic planning perspective

The one where the mouse deceives the fox... One solution (a plan) could be: announce(m, at(g, l, t)) ⊗ quit(f , l, t) such that: s0 ⊗ announce(m, at(g, l, t)) ⊗ quit(f , l, t) | = ¬threat(m, l, t + 1) which consists of a deceptive speech act...

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 45 / 55

slide-95
SLIDE 95

On synthsizing lies

From epistemic planning perspective

The one where the mouse deceives the gruffalo...

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 46 / 55

slide-96
SLIDE 96

On synthsizing lies

From epistemic planning perspective

The one where the mouse deceives the gruffalo... The gruffalo is not a credulous agent

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 46 / 55

slide-97
SLIDE 97

On synthsizing lies

From epistemic planning perspective

The one where the mouse deceives the gruffalo... The gruffalo is not a credulous agent Belief revision mechanism required.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 46 / 55

slide-98
SLIDE 98

On synthsizing lies

From epistemic planning perspective

The one where the mouse deceives the gruffalo... The gruffalo is not a credulous agent Belief revision mechanism required. Abduction involved!

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 46 / 55

slide-99
SLIDE 99

On synthsizing lies

From epistemic planning perspective

The one where the mouse deceives the gruffalo... The gruffalo is not a credulous agent Belief revision mechanism required. Abduction involved! A harder problem indeed!!

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 46 / 55

slide-100
SLIDE 100

Outline

1

Motivation

2

Lying and Cognition

3

Lies and Deception Defining-Modelling Lies (Existing work) Engineering Lies

4

Epistemic Planning Classical Planning Our approach State-of-the-art: Epistemic Planning

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 47 / 55

slide-101
SLIDE 101

State-of-the-art: Epistemic Planning

Syntactic approach

Compilation to Classical Planning problem (Muise et al.[MBF+15]7, Kominis and Geffner [KG15]8, [KG17]9

7Muise, Christian, et al. ”Planning over multi-agent epistemic states: A classical

planning approach.” Twenty-Ninth AAAI Conference on Artificial Intelligence. 2015.

8Kominis, Filippos, and Hector Geffner. ”Beliefs in multiagent planning: From one

agent to many.” Twenty-Fifth International Conference on Automated Planning and

  • Scheduling. 2015.

9Kominis, Filippos, and Hector Geffner. ”Multiagent online planning with nested

beliefs and dialogue.” Twenty-Seventh International Conference on Automated Planning and Scheduling. 2017.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 48 / 55

slide-102
SLIDE 102

State-of-the-art: Epistemic Planning

Syntactic approach

Compilation to Classical Planning problem (Muise et al.[MBF+15]7, Kominis and Geffner [KG15]8, [KG17]9 Take a restricted fragment of DEL and use a STRIPS-style encoding

  • f action operators over epistemic formulas instead of event models

and product updates.

7Muise, Christian, et al. ”Planning over multi-agent epistemic states: A classical

planning approach.” Twenty-Ninth AAAI Conference on Artificial Intelligence. 2015.

8Kominis, Filippos, and Hector Geffner. ”Beliefs in multiagent planning: From one

agent to many.” Twenty-Fifth International Conference on Automated Planning and

  • Scheduling. 2015.

9Kominis, Filippos, and Hector Geffner. ”Multiagent online planning with nested

beliefs and dialogue.” Twenty-Seventh International Conference on Automated Planning and Scheduling. 2017.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 48 / 55

slide-103
SLIDE 103

State-of-the-art: Epistemic Planning

Syntactic approach

Compilation to Classical Planning problem (Muise et al.[MBF+15]7, Kominis and Geffner [KG15]8, [KG17]9 Take a restricted fragment of DEL and use a STRIPS-style encoding

  • f action operators over epistemic formulas instead of event models

and product updates. Compile the given problem into an instance of classical planning - leverage the performance of state-of-the-art PDDL planners for solving it efficiently.

7Muise, Christian, et al. ”Planning over multi-agent epistemic states: A classical

planning approach.” Twenty-Ninth AAAI Conference on Artificial Intelligence. 2015.

8Kominis, Filippos, and Hector Geffner. ”Beliefs in multiagent planning: From one

agent to many.” Twenty-Fifth International Conference on Automated Planning and

  • Scheduling. 2015.

9Kominis, Filippos, and Hector Geffner. ”Multiagent online planning with nested

beliefs and dialogue.” Twenty-Seventh International Conference on Automated Planning and Scheduling. 2017.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 48 / 55

slide-104
SLIDE 104

State-of-the-art: Epistemic Planning

Semantic approach

DEL based planning with a plan as sequence of update/event models applicable over the initial epistemic state (a pointed Kripke model) to reach the desired epistemic state (characterised by the goal conditions) (Bolander and Andersen [BA11] and Lowe et al. [LPW11])

10Le, Tiep, et al. ”EFP and PG-EFP: Epistemic forward search planners in

multi-agent domains.” Twenty-Eighth International Conference on Automated Planning and Scheduling. 2018.

11Son, Tran Cao, et al. ”Finitary S5-theories.” European Workshop on Logics in

Artificial Intelligence. Springer, Cham, 2014.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 49 / 55

slide-105
SLIDE 105

State-of-the-art: Epistemic Planning

Semantic approach

DEL based planning with a plan as sequence of update/event models applicable over the initial epistemic state (a pointed Kripke model) to reach the desired epistemic state (characterised by the goal conditions) (Bolander and Andersen [BA11] and Lowe et al. [LPW11]) Le, Tiep, et al.[LFSP18]10 proposed presents prototypical epistemic forward planner which use language mA to specify multiagent planning domain and finitary S5-theory given by Son, Tran Cao, et al.[SPBG14]11 to represent belief states.

10Le, Tiep, et al. ”EFP and PG-EFP: Epistemic forward search planners in

multi-agent domains.” Twenty-Eighth International Conference on Automated Planning and Scheduling. 2018.

11Son, Tran Cao, et al. ”Finitary S5-theories.” European Workshop on Logics in

Artificial Intelligence. Springer, Cham, 2014.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 49 / 55

slide-106
SLIDE 106

State-of-the-art: Epistemic Planning

Semantic approach

Both approaches have their strength and weaknesses. Baral, Chitta et al. [BBvDM17]12 discuss these approaches in detail.

12Baral, C., Bolander, T., van Ditmarsch, H. and McIlrath, S., 2017. Epistemic

planning (Dagstuhl seminar 17231). In Dagstuhl Reports (Vol. 7, No. 6). Schloss Dagstuhl-Leibniz-Zentrum fuer Informatik

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 50 / 55

slide-107
SLIDE 107

We are in state of exploring the suitable planning formalism for our problem. Thank you.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 51 / 55

slide-108
SLIDE 108

References I

Thomas Bolander and Mikkel Birkegaard Andersen, Epistemic planning for single-and multi-agent systems, Journal of Applied Non-Classical Logics 21 (2011), no. 1, 9–34. Alexandru Baltag, A logic for suspicious players: Epistemic actions and belief–updates in games, Bulletin of Economic Research 54 (2002), no. 1, 1–45. Chitta Baral, Thomas Bolander, Hans van Ditmarsch, and Sheila McIlrath, Epistemic planning (dagstuhl seminar 17231), Dagstuhl Reports, vol. 7, Schloss Dagstuhl-Leibniz-Zentrum fuer Informatik, 2017. Thomas L Carson, The definition of lying, Noˆ us 40 (2006), no. 2, 284–306.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 52 / 55

slide-109
SLIDE 109

References II

Richard E Fikes and Nils J Nilsson, Strips: A new approach to the application of theorem proving to problem solving, Artificial intelligence 2 (1971), no. 3-4, 189–208. Filippos Kominis and Hector Geffner, Beliefs in multiagent planning: From one agent to many., ICAPS, 2015, pp. 147–155. , Multiagent online planning with nested beliefs and dialogue. Tiep Le, Francesco Fabiano, Tran Cao Son, and Enrico Pontelli, Efp and pg-efp: Epistemic forward search planners in multi-agent domains., ICAPS, 2018, pp. 161–170. Benedikt L¨

  • we, Eric Pacuit, and Andreas Witzel, Del planning and

some tractable cases, International Workshop on Logic, Rationality and Interaction, Springer, 2011, pp. 179–192.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 53 / 55

slide-110
SLIDE 110

References III

Christian J Muise, Vaishak Belle, Paolo Felli, Sheila A McIlraith, Tim Miller, Adrian R Pearce, and Liz Sonenberg, Planning over multi-agent epistemic states: A classical planning approach., AAAI, 2015,

  • pp. 3327–3334.

Tran Cao Son, Enrico Pontelli, Chitta Baral, and Gregory Gelfond, Finitary s5-theories, European Workshop on Logics in Artificial Intelligence, Springer, 2014, pp. 239–252. David Steiner, A system for consistency preserving belief change, Proceedings of the ESSLLI Workshop on Rationality and Knowledge, Citeseer, 2006, pp. 133–144. Victoria Talwar and Kang Lee, Social and cognitive correlates of childrens lying behavior, Child development 79 (2008), no. 4, 866–881.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 54 / 55

slide-111
SLIDE 111

References IV

Hans Van Ditmarsch, Dynamics of lying, Synthese 191 (2014), no. 5, 745–777. Hans Van Ditmarsch, Jan Van Eijck, Floor Sietsma, and Yanjing Wang, On the logic of lying, Games, actions and social software, Springer, 2012, pp. 41–72.

Shikha Singh, Deepak Khemani Deception: An epistemic planned event? 55 / 55