Upload
duongnhan
View
219
Download
1
Embed Size (px)
Citation preview
0001
0002
0003
0004
0005
0006
0007
0008
0009
0010
0011
0012
0013
0014
0015
0016
0017
0018
0019
0020
0021
0022
0023
0024
0025
0026
0027
0028
0029
0030
0031
0032
0033
0034
0035
0036
0037
0038
0039
0040
0041
0042
0043
Yet More Modal Logics of Preference Change
and Belief Revision
Jan van Eijck
Centre for Mathematics and Computer Science (CWI)Kruislaan 4131098 SJ Amsterdam, The Netherlands
Abstract
We contrast Bonanno’s ‘Belief Revision in a Temporal Framework’(Bonanno, 2008) with preference change and belief revision from theperspective of dynamic epistemic logic (DEL). For that, we extend thelogic of communication and change of van Benthem et al. (2006b) withrelational substitutions (van Benthem, 2004) for preference change,and show that this does not alter its properties. Next we move to amore constrained context where belief and knowledge can be definedfrom preferences (Grove, 1988; Board, 2002; Baltag and Smets, 2006,2008b), prove completeness of a very expressive logic of belief revision,and define a mechanism for updating belief revision models using acombination of action priority update (Baltag and Smets, 2008b) andpreference substitution (van Benthem, 2004).
1 Reconstructing AGM style belief revision
Bonanno’s paper offers a rational reconstruction of Alchourron GardenforsMakinson style belief revision (AGM belief revision) (Alchourron et al.,1985; see also Gardenfors, 1988 and Gardenfors and Rott, 1995), in a frame-work where modalities B for single agent belief and I for being informedare mixed with a next time operator © and its inverse ©−1.
Both the AGM framework and Bonanno’s reconstruction of it do notexplicitly represent the triggers that cause belief change in the first place.Iϕ expresses that the agent is informed that ϕ, but the communicativeaction that causes this change in information state is not represented. Also,ϕ is restricted to purely propositional formulas. Another limitation thatBonanno’s reconstruction shares with AGM is that it restricts attention toa single agent: changes of the beliefs of agents about the beliefs of otheragents are not analyzed. In these respects (Bonanno, 2008) is close toDynamic Doxastic Logic (DDL), as developed by Segerberg (1995, 1999).
AGM style belief revision was proposed more than twenty years ago, andhas grown into a paradigm in its own right in artificial intelligence. In the
Krzysztof R. Apt, Robert van Rooij (eds.). New Perspectives on Games and Interaction.
Texts in Logic and Games 4, Amsterdam University Press 2008, pp. 81–104.
0044
0045
0046
0047
0048
0049
0050
0051
0052
0053
0054
0055
0056
0057
0058
0059
0060
0061
0062
0063
0064
0065
0066
0067
0068
0069
0070
0071
0072
0073
0074
0075
0076
0077
0078
0079
0080
0081
0082
0083
0084
0085
0086
82 J. van Eijck
meanwhile rich frameworks of dynamic epistemic logic have emerged thatare quite a bit more ambitious in their goals than AGM was when it wasfirst proposed. AGM analyzes operations +ϕ for expanding with ϕ, −ϕfor retracting ϕ and ∗ϕ for revising with ϕ. It is formulated in a purelysyntactic way, it hardly addresses issues of semantics, it does not proposesound and complete axiomatisations. It did shine in 1985, and it still shinesnow, but perhaps in a more modest way.
Bonanno’s paper creates a nice link between this style of belief revisionand epistemic/doxastic logic. While similar in spirit to Segerberg’s work,it addresses the question of the rational reconstruction of AGM style beliefrevision more explicitly. This does add quite a lot to that framework: clearsemantics, and a sound and complete axiomatisation. Still, it is fair to saythat this rational reconstruction, nice as it is, also inherits the limitationsof the original design.
2 A broader perspective
Meanwhile, epistemic logic has entered a different phase, with a new focuson the epistemic and doxastic effects of information updates such as publicannouncements (Plaza, 1989; Gerbrandy, 1999). Public announcements areinteresting because they create common knowledge, so the new focus oninformation updating fostered an interest in the evolution of multi-agentknowledge and belief under acts of communication.
Public announcement was generalized to updates with ‘action models’that can express a wide range of communications (private announcements,group announcements, secret sharing, lies, and so on) in (Baltag et al., 1998)and (Baltag and Moss, 2004). A further generalization to a complete logicof communication and change, with enriched actions that allow changingthe facts of the world, was provided by van Benthem et al. (2006b). Thetextbook treatment of dynamic epistemic logic in (van Ditmarsch et al.,2007) bears witness to the fact that this approach is by now well established.
The above systems of dynamic epistemic logic do provide an account ofknowledge or belief update, but they do not analyse belief revision in thesense of AGM. Information updating in dynamic epistemic logic is mono-tonic: facts that are announced to an audience of agents cannot be unlearnt.Van Benthem (2004) calls this ‘belief change under hard information’ or‘eliminative belief revision’. See also (van Ditmarsch, 2005) for reflectionon the distinction between this and belief change under soft information.
Assume a state of the world where p actually is the case, and where youknow it, but I do not. Then public announcement of p will have the effectthat I get to know it, but also that you know that I know it, that I knowthat you know that I know it, in short, that p becomes common knowledge.But if this announcement is followed by an announcement of ¬p, the effect
0087
0088
0089
0090
0091
0092
0093
0094
0095
0096
0097
0098
0099
0100
0101
0102
0103
0104
0105
0106
0107
0108
0109
0110
0111
0112
0113
0114
0115
0116
0117
0118
0119
0120
0121
0122
0123
0124
0125
0126
0127
0128
0129
Yet More Modal Logics of Preference Change and Belief Revision 83
will be inconsistent knowledge states for both of us.It is clear that AGM deals with belief revision of a different kind: ‘be-
lief change under soft information’ or ‘non-eliminative belief revision’. In(van Benthem, 2004) it is sketched how this can be incorporated into dy-namic epistemic logic, and in the closely related (Baltag and Smets, 2008b)a theory of ‘doxastic actions’ is developed that can be seen as a further stepin this direction.
Belief revision under soft information can, as Van Benthem observes, bemodelled as change in the belief accessibilities of a model. This is differentfrom public announcement, which can be viewed as elimination of worldswhile leaving the accessibilities untouched.
Agent i believes that ϕ in a given world w if it is the case that ϕ istrue in all worlds t that are reachable from w and that are minimal for asuitable plausibility ordering relation ≤i. In the dynamic logic of belief revi-sion these accessibilities can get updated in various ways. An example from(Rott, 2006) that is discussed by van Benthem (2004) is ⇑A for so-called‘lexicographic upgrade’: all A worlds get promoted past all non-A worlds,while within the A worlds and within the non-A worlds the preference re-lation stays as before. Clearly this relation upgrade has as effect that itcreates belief in A. And the belief upgrade can be undone: a next updatewith ⇑¬A does not result in inconsistency.
Van Benthem (2004) gives a complete dynamic logic of belief upgradefor the belief upgrade operation ⇑A, and another one for a variation onit, ↑A, or ‘elite change’, that updates a plausibility ordering to a new onewhere the best A worlds get promoted past all other worlds, and for therest the old ordering remains unchanged.
This is taken one step further in a general logic for changing prefer-ences, in (van Benthem and Liu, 2004), where upgrade as relation changeis handled for (reflexive and transitive) preference relations ≤i, by meansof a variation on product update called product upgrade. The idea is tokeep the domain, the valuation and the epistemic relations the same, butto reset the preferences by means of a substitution of new preorders for thepreference relations.
Treating knowledge as an equivalence and preference as a preorder, with-out constraining the way in which they relate, as is done by van Benthemand Liu (2004), has the advantage of generality (one does not have to spec-ify what ‘having a preference’ means), but it makes it harder to use thepreference relation for modelling belief change. If one models ‘regret’ aspreferring a situation that one knows to be false to the current situation,then it follows that one can regret things one cannot even conceive. And us-ing the same preference relation for belief looks strange, for this would allowbeliefs that are known to be false. Van Benthem (private communication)
0130
0131
0132
0133
0134
0135
0136
0137
0138
0139
0140
0141
0142
0143
0144
0145
0146
0147
0148
0149
0150
0151
0152
0153
0154
0155
0156
0157
0158
0159
0160
0161
0162
0163
0164
0165
0166
0167
0168
0169
0170
0171
0172
84 J. van Eijck
advised me not to lose sleep over such philosophical issues. If we follow thatadvice, and call ‘belief’ what is true in all most preferred worlds, we can stilltake comfort from the fact that this view entails that one can never believeone is in a bad situation: the belief-accessible situations are by definitionthe best conceivable worlds. Anyhow, proceeding from the assumption thatknowledge and preference are independent basic relations and then study-ing possible relations between them has turned out very fruitful: the recenttheses by Girard (2008) and Liu (2008) are rich sources of insight in whata logical study of the interaction of knowledge and preference may reveal.
Here we will explore two avenues, different from the above but related toit. First, we assume nothing at all about the relation between knowledge onone hand and preference on the other. We show that the dynamic logic ofthis (including updating with suitable finite update models) is complete anddecidable: Theorem 3.1 gives an extension of the reducibility result for LCC,the general logic of communication and change proposed and investigatedin (van Benthem et al., 2006b).
Next, we move closer to the AGM perspective, by postulating a closeconnection between knowledge, belief and preference. One takes prefer-ences as primary, and imposes minimal conditions to allow a definition ofknowledge from preferences. The key to this is the simple observation inTheorem 4.1 that a preorder can be turned into an equivalence by takingits symmetric closure if and only if it is weakly connected and converselyweakly connected. This means that by starting from weakly and converseweakly connected preorders one can interpret their symmetric closures asknowledge relations, and use the preferences themselves to define conditionalbeliefs, in the well known way that was first proposed in (Boutilier, 1994)and (Halpern, 1997). The multi-agent version of this kind of conditionalbelief was further explored in (van Benthem et al., 2006a) and in (Baltagand Smets, 2006, 2008b). We extend this to a complete logic of regular dox-astic programs for belief revision models (Theorem 4.3), useful for reasoningabout common knowledge, common conditional belief and their interaction.Finally, we make a formal proposal for a belief change mechanism by meansof a combination of action model update in the style of (Baltag and Smets,2008b) and plausibility substitution in the style of (van Benthem and Liu,2004).
3 Preference change in LCC
An epistemic preference model M for set of agents I is a tuple (W,V,R, P )where W is a non-empty set of worlds, V is a propositional valuation, R isa function that maps each agent i to a relation Ri (the epistemic relationfor i), and P is a function that maps each agent i to a preference relationPi. There are no conditions at all on the Ri and the Pi (just as there are
0173
0174
0175
0176
0177
0178
0179
0180
0181
0182
0183
0184
0185
0186
0187
0188
0189
0190
0191
0192
0193
0194
0195
0196
0197
0198
0199
0200
0201
0202
0203
0204
0205
0206
0207
0208
0209
0210
0211
0212
0213
0214
0215
Yet More Modal Logics of Preference Change and Belief Revision 85
no constraints on the Ri relations in LCC (van Benthem et al., 2006b)).We fix a PDL style language for talking about epistemic preference mod-
els (assume p ranges over a set of basic propositions Prop and i over a setof agents I):
ϕ ::= ⊤ | p | ¬ϕ | ϕ1 ∧ ϕ2 | [π]ϕ
π ::= ∼i | ≥i |?ϕ | π1;π2 | π1 ∪ π2 | π∗
This is to be interpreted in the usual PDL manner, with [[[π]]]M giving therelation that interprets relational expression π in M = (W,V,R, P ), where∼i is interpreted as the relation Ri and ≥i as the relation Pi, and wherethe complex modalities are handled by the regular operations on relations.We employ the usual abbreviations: ⊥ is shorthand for ¬⊤, ϕ1 ∨ ϕ2 isshorthand for ¬(¬ϕ1∧¬ϕ2), ϕ1 → ϕ2 is shorthand for ¬(ϕ1∧ϕ2), ϕ1 ↔ ϕ2
is shorthand for (ϕ1 → ϕ2)∧ (ϕ2 → ϕ1), and 〈π〉ϕ is shorthand for ¬[π]¬ϕ.The formula [π]ϕ is true in world w of M if for all v with (w, v) ∈ [[[π]]]M
it holds that ϕ is true in v. This is completely axiomatised by the usualPDL rules and axioms (Segerberg, 1982; Kozen and Parikh, 1981):
Modus ponens and axioms for propositional logicModal generalisation From ⊢ ϕ infer ⊢ [π]ϕ
Normality ⊢ [π](ϕ→ ψ)→ ([π]ϕ→ [π]ψ)Test ⊢ [?ϕ]ψ ↔ (ϕ→ ψ)Sequence ⊢ [π1;π2]ϕ↔ [π1][π2]ϕChoice ⊢ [π1 ∪ π2]ϕ↔ ([π1]ϕ ∧ [π2]ϕ)Mix ⊢ [π∗]ϕ↔ (ϕ ∧ [π][π∗]ϕ)Induction ⊢ (ϕ ∧ [π∗](ϕ→ [π]ϕ))→ [π∗]ϕ
In (van Benthem et al., 2006b) it is proved that extending the PDL languagewith an extra modality [A, e]ϕ does not change the expressive power of thelanguage. Interpretation of the new modality: [A, e]ϕ is true in w in M ifsuccess of the update of M with action model A to M ⊗ A implies that ϕis true in (w, e) in M ⊗ A. To see what that means one has to grasp thedefinition of update models A and the update product operation ⊗, whichwe will now give for the epistemic preference case.
An action model (for agent set I) is like an epistemic preference modelfor I, with the difference that the worlds are now called events, and thatthe valuation has been replaced by a precondition map pre that assigns toeach event e a formula of the language called the precondition of e. Fromnow on we call the epistemic preference models static models.
Updating a static model M = (W,V,R, P ) with an action model A =(E,pre,R,P) succeeds if the set
{(w, e) | w ∈W, e ∈ E,M, w |= pre(e)}
0216
0217
0218
0219
0220
0221
0222
0223
0224
0225
0226
0227
0228
0229
0230
0231
0232
0233
0234
0235
0236
0237
0238
0239
0240
0241
0242
0243
0244
0245
0246
0247
0248
0249
0250
0251
0252
0253
0254
0255
0256
0257
0258
86 J. van Eijck
is non-empty. The result of this update is a new static model M ⊗ A =(W ′, V ′, R′, P ′) with
• W ′ = {(w, e) | w ∈W, e ∈ E,M, w |= pre(e)},
• V ′(w, e) = V (w),
• R′i is given by {(w, e), (v, f)) | (w, v) ∈ Ri, (e, f) ∈ Ri},
• P ′i is given by {(w, e), (v, f)) | (w, v) ∈ Pi, (e, f) ∈ Pi}.
If the static model has a set of distinguished states W0 and the action modela set of distinguished events E0, then the distinguished worlds of M ⊗ Aare the (w, e) with w ∈ W0 and e ∈ E0.
Figure 1. Static model and update.
Figure 1 gives an example pair of a static model with an update action.The distinguished worlds of the model and the distinguished event of theaction model are shaded grey. Only the R relations are drawn, for threeagents a, b, c. The result of the update is shown in Figure 2, on the left.This result can be reduced to the bisimilar model on the right in the samefigure, with the bisimulation linking the distinguished worlds. The result ofthe update is that the distinguished “wine” world has disappeared, withoutany of a, b, c being aware of the change.
In LCC, action update is extended with factual change, which is handledby propositional substitution. Here we will consider another extension, withpreference change, handled by preference substitution (first proposed byvan Benthem and Liu, 2004). A preference substitution is a map fromagents to PDL program expressions π represented by a finite set of bindings
{i1 7→ π1, . . . , in 7→ πn}
where the ij are agents, all different, and where the πi are program ex-pressions from our PDL language. It is assumed that each i that does
0259
0260
0261
0262
0263
0264
0265
0266
0267
0268
0269
0270
0271
0272
0273
0274
0275
0276
0277
0278
0279
0280
0281
0282
0283
0284
0285
0286
0287
0288
0289
0290
0291
0292
0293
0294
0295
0296
0297
0298
0299
0300
0301
Yet More Modal Logics of Preference Change and Belief Revision 87
Figure 2. Update result, before and after reduction under bisimulation.
not occur in the lefthand side of a binding is mapped to ≥i. Call the set{i ∈ I | ρ(i) 6= ≥i} the domain of ρ. If M = (W,V,R, P ) is a preferencemodel and ρ is a preference substitution, then Mρ is the result of changingthe preference map P of M to P ρ given by:
P ρ(i) := Pi for i not in the domain of ρ,
P ρ(i) := [[[ρ(i)]]]M for i = ij in the domain of ρ.
Now extend the PDL language with a modality [ρ]ϕ for preference change,with the following interpretation:
M, w |= [ρ]ϕ :⇐⇒ Mρ, w |= ϕ.
Then we get a complete logic for preference change:
Theorem 3.1. The logic of epistemic preference PDL with preference changemodalities is complete.
Proof. The preference change effects of [ρ] can be captured by a set of re-duction axioms for [ρ] that commute with all sentential language constructs,and that handle formulas of the form [ρ][π]ϕ by means of reduction axiomsof the form
[ρ][π]ϕ ↔ [Fρ(π)][ρ]ϕ,
0302
0303
0304
0305
0306
0307
0308
0309
0310
0311
0312
0313
0314
0315
0316
0317
0318
0319
0320
0321
0322
0323
0324
0325
0326
0327
0328
0329
0330
0331
0332
0333
0334
0335
0336
0337
0338
0339
0340
0341
0342
0343
0344
88 J. van Eijck
with Fρ given by:
Fρ(∼i) := ∼i
Fρ(≥i) :=
{
ρ(i) if i in the domain of ρ,≥i otherwise,
Fρ(?ϕ) := ?[ρ]ϕ,Fρ(π1;π2) := Fρ(π1);Fρ(π2),Fρ(π1 ∪ π2) := Fρ(π1) ∪ Fρ(π2),Fρ(π
∗) := (Fρ(π))∗.
It is easily checked that these reduction axioms are sound, and that for eachformula of the extended language the axioms yield an equivalent formulain which [ρ] occurs with lower complexity, which means that the reductionaxioms can be used to translate formulas of the extended language to PDLformulas. Completeness then follows from the completeness of PDL. q.e.d.
4 Yet another logic. . .
In this section we look at a more constrained case, by replacing the epis-temic preference models by ‘belief revision models’ in the style of Grove(1988), Board (2002), and Baltag and Smets (2006, 2008b) (who call them‘multi-agent plausibility frames’). There is almost complete agreement thatpreference relations should be transitive and reflexive (preorders). But tran-sitivity plus reflexivity of a binary relation R do not together imply thatR ∪ Rˇ is an equivalence. Figure 3 gives a counterexample. The two extra
Figure 3. Preorder with a non-transitive symmetric closure.
conditions of weak connectedness for R and for Rˇ remedy this. A binaryrelation R is weakly connected (terminology of Goldblatt, 1987) if the fol-lowing holds:
∀x, y, z((xRy ∧ xRz)→ (yRz ∨ y = z ∨ zRy)).
Theorem 4.1. Assume R is a preorder. Then R ∪Rˇ is an equivalence iffboth R and Rˇ are weakly connected.
0345
0346
0347
0348
0349
0350
0351
0352
0353
0354
0355
0356
0357
0358
0359
0360
0361
0362
0363
0364
0365
0366
0367
0368
0369
0370
0371
0372
0373
0374
0375
0376
0377
0378
0379
0380
0381
0382
0383
0384
0385
0386
0387
Yet More Modal Logics of Preference Change and Belief Revision 89
Proof. ⇒: immediate.⇐: Let R be a preorder such that both R and Rˇ are weakly connected.
We have to show that R∪Rˇ is an equivalence. Symmetry and reflexivity areimmediate. For the check of transitivity, assume xR ∪ R y and yR ∪ R z.There are four cases. (i) xRyRz. Then xRz by transitivity of R, hencexR ∪ R z. (ii) xRyR z. Then yR x and yR z, and by weak connectednessof R , either (xR z or zR x), hence xR ∪ R z, or x = z, hence xRz byreflexivity of R. Therefore xR ∪ R z in all cases. (iii) xR yRz. Similar.(iv) xR yR z. Then zRyRx, and zRx by transitivity of R. ThereforexR ∪R z. q.e.d.
Call a preorder that is weakly connected and conversely weakly con-nected locally connected. The example in Figure 4 shows that locally con-nected preorders need not be connected. Taking the symmetric closure of
Figure 4. Locally connected preorder that is not connected.
this example generates an equivalence with two equivalence classes. Moregenerally, taking the symmetric closure of a locally connected preorder cre-ates an equivalence that can play the role of a knowledge relation definedfrom the preference order. To interpret the preference order as conditionalbelief, it is convenient to assume that it is also well-founded: this makes fora smooth definition of the notion of a ‘best possible world’.
A belief revision model M (again, for a set of agents I) is a tuple(W,V, P ) where W is a non-empty set of worlds, V is a propositional valua-tion and P is a function that maps each agent i to a preference relation ≤ithat is a locally connected well-preorder. That is, ≤i is a preorder (reflexiveand transitive) that is well-founded (in terms of <i for the strict part of≤i, this is the requirement that there is no infinite sequence of w1, w2, . . .
with . . . <i w2 <i w1), and such that both ≤i and its converse are weaklyconnected.
0388
0389
0390
0391
0392
0393
0394
0395
0396
0397
0398
0399
0400
0401
0402
0403
0404
0405
0406
0407
0408
0409
0410
0411
0412
0413
0414
0415
0416
0417
0418
0419
0420
0421
0422
0423
0424
0425
0426
0427
0428
0429
0430
90 J. van Eijck
In what follows we will use <i with the meaning explained above, ≥i forthe converse of ≤i, >i for the converse of <i, and ∼i for ≤i ∪ ≥i.
The locally connected well-preorders≤i can be used to induce accessibil-ity relations→P
i for each subset P of the domain, by means of the followingstandard definition:
→Pi := {(x, y) | x ∼i y ∧ y ∈MIN≤i
P},
where MIN≤iP , the set of minimal elements of P under ≤i, is defined as
{s ∈ P : ∀s′ ∈ P (s′ ≤ s⇒ s ≤ s′)}.
This picks out the minimal worlds linked to the current world, accord-ing to ≤i, within the set of worlds satisfying [[ϕ]]M. The requirement ofwellfoundedness ensures that MIN≤P will be non-empty for non-empty P .Investigating these→P relations, we see that they have plausible propertiesfor belief:
Proposition 4.2. Let ≤ be a locally connected well-preorder on S and letP be a non-empty subset of S. Then→P is transitive, euclidean and serial.
Proof. Transitivity: if x→P y then y ∼ x and y ∈MIN≤P . If y →P z thenz ∼ y and z ∈ MIN≤P . It follows by local connectedness of ≤ that z ∼ x
and by the definition of →P that x→P z.Euclideanness: let x→P y and x→P z. We have to show y →P z. From
x →P y, y ∼ x and y ∈ MIN≤P . From x →P z, z ∼ x and z ∈ MIN≤P .From local connectedness, y ∼ z. Hence y →P z.
Seriality: Let x ∈ P . Since ≤ is a preorder there are y ∈ P withy ≤ x. The wellfoundedness of ≤ guarantees that there are ≤ minimal suchy. q.e.d.
Transitivity, euclideanness and seriality are the frame properties correspond-ing to positively and negatively introspective consistent belief (KD45 belief,Chellas, 1984).
Figure 5 gives an example with both the ≤ relation (shown as solidarrows in the direction of more preferred worlds, i.e., with an arrow from x
to y for x ≥ y) and the induced → relation on the whole domain (shown asdotted arrows). The above gives us in fact knowledge relations ∼i togetherwith for each knowledge cell a Lewis-style (1973) counterfactual relation: aconnected well-preorder, which can be viewed as a set of nested spheres, withthe minimal elements as the innermost sphere. Compare also the conditionalmodels of Burgess (1981) and Veltman (1985) (linked to Dynamic DoxasticLogic in Girard, 2007).
Baltag and Smets (2008b,a) present logics of individual multi-agent be-lief and knowledge for belief revision models, and define belief update for
0431
0432
0433
0434
0435
0436
0437
0438
0439
0440
0441
0442
0443
0444
0445
0446
0447
0448
0449
0450
0451
0452
0453
0454
0455
0456
0457
0458
0459
0460
0461
0462
0463
0464
0465
0466
0467
0468
0469
0470
0471
0472
0473
Yet More Modal Logics of Preference Change and Belief Revision 91
Figure 5. Preference (solid arrows) and belief (dotted arrows).
this as a particular kind of action update in the style of (Baltag et al., 1998),called action priority update. Here we sketch the extension to a system thatalso handles common knowledge and common conditional belief, and wherethe action update has belief change incorporated in it by means of relationalsubstitution.
The set-up will be less general than in the logic LCC: in LCC no assump-tions are made about the update actions, and so the accessibility relationscould easily deteriorate, e.g., as a result of updating with a lie. Since inthe present set-up we make assumptions about the accessibilities (to wit,that they are locally connected well-preorders), we have to ensure that ourupdate actions preserve these relational properties.
Consider the following slight modification of the PDL language (againassume p ranges over a set of basic propositions Prop and i over a set ofagents I):
ϕ ::= ⊤ | p | ¬ϕ | ϕ1 ∧ ϕ2 | [π]ϕ
π ::= ∼i | ≤i | ≥i | →ϕi | ←
ϕi | G |?ϕ | π1;π2 | π1 ∪ π2 | π
∗
Call this language LPref. This time, we treat ∼i as a derived notion, byputting in an axiom that defines ∼i as ≤i∪≥i. The intention is to let ∼i beinterpreted as the knowledge relation for agent i, ≤i as the preference rela-tion for i, ≥i as the converse preference relation for i,→ϕ
i as the conditionalbelief relation defined from ≤i as explained above, ←ϕ
i as its converse, andG as global accessibility. We use →i as shorthand for →⊤
i .We have added a global modalityG, and we will set up things in such way
that [G]ϕ expresses that everywhere in the model ϕ holds, and that 〈G〉ϕexpresses that ϕ holds somewhere. It is well-known that adding a globalmodality and converses to PDL does not change its properties: the logic re-
0474
0475
0476
0477
0478
0479
0480
0481
0482
0483
0484
0485
0486
0487
0488
0489
0490
0491
0492
0493
0494
0495
0496
0497
0498
0499
0500
0501
0502
0503
0504
0505
0506
0507
0508
0509
0510
0511
0512
0513
0514
0515
0516
92 J. van Eijck
mains decidable, and satisfiability remains EXPTIME-complete (Blackburnet al., 2001).
The semantics of LPref is given relative to belief revision models as indi-cated above. Formula meaning [[ϕ]]M and relational meaning [[[π]]]M are han-dled in the usual way. The interpretation of the knowledge relation of agenti is given by [[[∼i]]]
M := ≤M
i ∪ ≥M
i , that for the preference relation of agenti by [[[≤i]]]
M := ≤M
i , that for the converse preference relation of agent i by
its converse, that for the conditional belief of agent i by [[[→ϕi ]]]M :=→
[[ϕ]]M
i ,that for ←ϕ
i by its converse. The global modality is interpreted as the uni-versal relation, and test, sequential composition, choice and Kleene star areinterpreted as usual.
The interplay between the modalities [∼i] (knowledge) and [≥i] (safebelief) is analysed by Baltag and Smets (2008b), where they remark thatthe converse preference modality [≥i] in belief revision models behaves likean S4.3 modality (reflexive, transitive and not forward branching), and liveshappily together with the S5 modality for [∼i].
To see how this all works out, let us have a look at the truth conditionsfor [→ϕ
i ]ψ. This is true in a world w in model M if in all worlds v withv ∼i w and v minimal in [[[ϕ]]]M under ≤i it holds that ψ is true. This isindeed conditional belief, relative to ϕ. Compare this with [≥i]ψ. This istrue in a world w if in all worlds that are at least as preferred, ψ is true.Finally, [∼i]ψ is true in w if ψ is true in all worlds, preferred or not, that ican access from w.
As a further example, consider a situation where Alexandru is drinkingwine, while Jan does not know whether he is drinking wine or beer, andSonja thinks that he is drinking tea. The actual situation is shaded grey,Jan’s preference relation has solid lines, that of Sonja dotted lines. Reflexivearrows are not drawn, so Alexandru’s preferences are not visible in thepicture.
In the actual world it is true that Jan knows that Alexandru knows whatAlexandru is drinking: [∼j ]([∼a]w∨ [∼a]b), and that Sonja believes Alexan-dru is drinking tea and that Alexandru knows it: [s][∼a]t. Under condi-tion ¬t, however, Sonja has the belief in the actual world that Alexandruis drinking beer: [→¬t
s ]b. Moreover, Jan and Sonja have a common be-lief under condition ¬t that Alexandru is drinking wine or beer: [→¬t
j ∪→¬ts ; (→¬t
j ∪→¬ts )∗](w ∨ b). As a final illustration, note that [←s]⊥ is true
0517
0518
0519
0520
0521
0522
0523
0524
0525
0526
0527
0528
0529
0530
0531
0532
0533
0534
0535
0536
0537
0538
0539
0540
0541
0542
0543
0544
0545
0546
0547
0548
0549
0550
0551
0552
0553
0554
0555
0556
0557
0558
0559
Yet More Modal Logics of Preference Change and Belief Revision 93
in a world if this is not among Sonja’s most preferred worlds. Notice thatif Sonja conditionalizes her belief to these worlds, she would believe that
Alexandru is drinking beer: [→[sˇ]⊥s ]b is true in the actual world.
It should be clear from the example that this language is very expressive.To get at a complete logic for it, we need axioms and rules for propositionallogic, S5 axioms for the global modality (Goranko and Passy, 1992), axiomsfor forward connectedness of ≥ and of ≤ (see Goldblatt, 1987), axioms forconverse, relating ≤ to ≥ and → to ←, as in temporal logic (Prior, 1967),and the general axioms and rules for PDL (Segerberg, 1982). Finally, addthe following definition of conditional belief in terms of knowledge and safebelief that can already be found in (Boutilier, 1994) as an axiom:
[→ϕi ]ψ :≡ 〈∼i〉ϕ→ 〈∼i〉(ϕ ∧ [≥i](ϕ→ ψ)).
This definition (also used in Baltag and Smets, 2008b) states that condi-tional to ϕ, i believes in ψ if either there are no accessible ϕ worlds, orthere is an accessible ϕ world in which the belief in ϕ→ ψ is safe. The fullcalculus for LPref is given in Figure 6.
Theorem 4.3. The axiom system for LPref is complete for belief revisionmodels; LPref has the finite model property and is decidable.
Proof. Modify the canonical model construction for modal logic for the caseof PDL, by means of Fischer-Ladner closures (Fischer and Ladner, 1979)(also see Blackburn et al., 2001). This gives a finite canonical model withthe properties for ≤i and ≥i corresponding to the axioms (since the axiomsfor ≤i and ≥i are canonical). In particular, each≥i relation will be reflexive,transitive and weakly connected, each relation ≤i will be weakly connected,and the ≤i and ≥i relations will be converses of each other. Together thisgives (Theorem 4.1) that the ≤i ∪ ≥i are equivalences. Since the canonicalmodel has a finite set of nodes, each ≤i relation is also well-founded. Thus,the canonical model is in fact a belief revision model. Also, the →i and←i relations are converses of each other, and related to the ≥i relationsin the correct way. The canonical model construction gives us for eachconsistent formula ϕ a belief revision model satisfying ϕ with a finite setof nodes. Only finitely many of the relations in that model are relevantto the satisfiability of ϕ, so this gives a finite model (see Blackburn et al.,2001, for further details). Since the logic has the finite model property it isdecidable. q.e.d.
Since the axiomatisation is complete, the S5 properties of ∼i are deriv-able, as well as the principle that knowledge implies safe belief:
[∼i]ϕ→ [≥i]ϕ.
0560
0561
0562
0563
0564
0565
0566
0567
0568
0569
0570
0571
0572
0573
0574
0575
0576
0577
0578
0579
0580
0581
0582
0583
0584
0585
0586
0587
0588
0589
0590
0591
0592
0593
0594
0595
0596
0597
0598
0599
0600
0601
0602
94 J. van Eijck
Modus ponens and axioms for propositional logic
Modal generalisation From ⊢ ϕ infer ⊢ [π]ϕ
Normality ⊢ [π](ϕ→ ψ)→ ([π]ϕ→ [π]ψ)
Inclusion of everything in G ⊢ [G]ϕ→ [π]ϕ
Reflexivity of G ⊢ [G]ϕ→ ϕ
Transitivity of G ⊢ [G]ϕ→ [G][G]ϕ
Symmetry of G ⊢ ϕ→ [G]〈G〉ϕ
Knowledge definition ⊢ [∼i]ϕ↔ [≤i ∪ ≥i]ϕ
Truthfulness of safe belief ⊢ [≥i]ϕ→ ϕ
Transitivity of safe belief ⊢ [≥i]ϕ→ [≥i][≥i]ϕ
≥ included in ≤ˇ ⊢ ϕ→ [≥i]〈≤i〉ϕ
≤ included in ≥ˇ ⊢ ϕ→ [≤i]〈≥i〉ϕ
Weak connectedness of ≤ ⊢ [≤i]((ϕ ∧ [≤i]ϕ)→ ψ) ∨ [≤i]((ψ ∧ [≤i]ψ)→ ϕ)
Weak connectedness of ≥ ⊢ [≥i]((ϕ ∧ [≥i]ϕ)→ ψ) ∨ [≥i]((ψ ∧ [≥i]ψ)→ ϕ)
Conditional belief definition ⊢ [→ϕi ]ψ ↔ (〈∼i〉ϕ→ 〈∼i〉(ϕ ∧ [≥i](ϕ→ ψ)))
→ included in ←ˇ ⊢ ϕ→ [→ψi ]〈←ψ
i 〉ϕ
← included in →ˇ ⊢ ϕ→ [←ψi ]〈→ψ
i 〉ϕ
Test ⊢ [?ϕ]ψ ↔ (ϕ→ ψ)
Sequence ⊢ [π1;π2]ϕ↔ [π1][π2]ϕ
Choice ⊢ [π1 ∪ π2]ϕ↔ ([π1]ϕ ∧ [π2]ϕ)
Mix ⊢ [π∗]ϕ↔ (ϕ ∧ [π][π∗]ϕ)
Induction ⊢ (ϕ ∧ [π∗](ϕ→ [π]ϕ))→ [π∗]ϕ
Figure 6. Axiom system for LPref.
0603
0604
0605
0606
0607
0608
0609
0610
0611
0612
0613
0614
0615
0616
0617
0618
0619
0620
0621
0622
0623
0624
0625
0626
0627
0628
0629
0630
0631
0632
0633
0634
0635
0636
0637
0638
0639
0640
0641
0642
0643
0644
0645
Yet More Modal Logics of Preference Change and Belief Revision 95
The same holds for the following principles for conditional belief given in(Board, 2002):
Safe belief implies belief ⊢ [≥i]ϕ→ [→ψi ]ϕ
Positive introspection ⊢ [→ψi ]ϕ→ [∼i][→ψi]ϕ
Negative introspection ⊢ ¬[→ψi ]ϕ→ [∼i]¬[→
ψi ]ϕ
Successful revision ⊢ [→ϕi ]ϕ
Minimality of revision ⊢ 〈→ϕi 〉ψ → ([iϕ∧ψ]χ↔ [→ϕ
i ](ψ → χ))
We end with an open question: is →ϕi definable from ≥i and ≤i using only
test, sequence, choice and Kleene star?
5 Combining update and upgrade
The way we composed knowledge and belief by means of regular operationsmay have a dynamic flavour, but appearance is deceptive. The resultingdoxastic and epistemic ‘programs’ still describe what goes on in a staticmodel. Real communicative action is changing old belief revision modelsinto new ones. These actions should represent new hard information thatcannot be undone, but also soft information like belief changes that canbe reversed again later on. For this we can use update action by meansof action models, with soft information update handled by means of actionpriority update (Baltag and Smets, 2008b,a), or preference substitution asin (van Benthem and Liu, 2004). Here we will propose a combination ofthese two.
Action models for belief revision are like belief revision models, butwith the valuation replaced by a precondition map. We add two extraingredients. First, we add to each event a propositional substitution, to beused, as in LCC, for making factual changes to static models. Propositionalsubstitutions are maps represented as sets of bindings
{p1 7→ ϕ1, . . . , pn 7→ ϕn}
where all the pi are different. It is assumed that each p that does not occur inthe lefthand side of a binding is mapped to p. The domain of a propositionalsubstitution σ is the set {p ∈ Prop | σ(p) 6= p}. If M = (W,V, P ) is a beliefrevision model and σ is an LPref propositional substitution, then V σ
Mis the
valuation given by λwλp · w ∈ [[pσ]]M. In other words, V σM
assigns to w theset of basic propositions p such that pσ is true in world w in model M. Mσ
is the model with its valuation changed by σ as indicated. Next, we addrelational substitutions, as defined in Section 3, one to each event. Thus,an action model for belief revision is a tuple A = (E,pre,P,psub, rsub)with E a non-empty finite set of events, psub and rsub maps from E topropositional substitutions and relational substitutions, respectively, andwith rsub subject to the following constraint:
0646
0647
0648
0649
0650
0651
0652
0653
0654
0655
0656
0657
0658
0659
0660
0661
0662
0663
0664
0665
0666
0667
0668
0669
0670
0671
0672
0673
0674
0675
0676
0677
0678
0679
0680
0681
0682
0683
0684
0685
0686
0687
0688
96 J. van Eijck
If e ∼i f in the action model, then rsub(e) and rsub(f) have thesame binding for i.
This ensures a coherent definition of the effect of relational substitution ona belief structure. The example in Figure 7 illustrates this. But note thatthe substitutions are subject to further constraints. In the action model in
Figure 7. Unconstrained relational substitution creates havoc.
the example, a single agent has a preference for ¬p over p. In the updatemodel, a substitution reverses the agent’s preferences, but the agent cannotdistinguish this from an action where nothing happens. What should theresult of the update look like? E.g., is there a preference arrow from (2, 3)to (1, 4)? This is impossible to answer, as action 3 asks us to reverse thepreference and action 4 demands that we keep the initial preference. Theconstraint on substitutions rules out such dilemmas.
The relational substitution ρ = rsub(e) at event e in action model A ismeant to to be interpreted ‘locally’ at each world w in input model M. IfP is the preference map of M, then let P ρw be given by:
P ρw(i) := Pi ∩ |w|2[[[∼i]]]M
for i not in the domain of ρ,
P ρw(i) := [[[ρ(i)]]]M ∩ |w|2[[[∼i]]]M
for i = ij in the domain of ρ.
Thus, P ρw is the result of making a change only to the local knowledge cellat world w of agent i (which is given by the equivalence class |w|[[[∼i]]]M).Let
P ρ(i) :=⋃
w∈W
P ρw(i)
Then P ρ(i) gives the result of the substitution ρ on P (i), for each knowledgecell |w|[[[∼i]]]M for i, and P ρ gives the result of the substitution ρ on P , foreach agent i.
0689
0690
0691
0692
0693
0694
0695
0696
0697
0698
0699
0700
0701
0702
0703
0704
0705
0706
0707
0708
0709
0710
0711
0712
0713
0714
0715
0716
0717
0718
0719
0720
0721
0722
0723
0724
0725
0726
0727
0728
0729
0730
0731
Yet More Modal Logics of Preference Change and Belief Revision 97
Now the result of updating belief revision model M = (W,V, P ) with ac-tion model A = (E,pre,P,psub, rsub) is given by M⊗A = (W ′, V ′, P ′),where
• W ′ = {(w, e) | w ∈W, e ∈ E,M, w |= pre(e)},
• V ′(w, e) = V σ(w),
• P ′(i) is given by the anti-lexicographical order defined from P ρ(i) andP(i) (see Baltag and Smets, 2008b,a).
With these definitions in place, what are reasonable substitutions? A pos-sible general form for a preference change could be a binding like this:
≥i 7→ [ϕ1, ϕ2, . . . , ϕn].
This is to be interpreted as an instruction to replace the belief preferencesof i in the local knowledge cells by the new preference relation that prefersthe ϕ1 states above everything else, the ¬ϕ1 ∧ ϕ2 above the ¬ϕ1 ∧ ¬ϕ2
states, and so on, and the ¬ϕ1 ∧ ¬ϕ2 ∧ · · · ∧ ¬ϕn−1 ∧ ϕn states above the¬ϕ1 ∧ ¬ϕ2 ∧ · · · ∧ ¬ϕn states. Such relations are indeed connected well-preorders.
Note that we can take [ϕ1, ϕ2, . . . , ϕn] as an abbreviation for the followingdoxastic program:
(∼i; ?ϕ1) ∪ (?¬ϕ1;∼i; ?¬ϕ1; ?ϕ2)
∪ (?¬ϕ1; ?¬ϕ2;∼i; ?¬ϕ1; ?¬ϕ2; ?ϕ3)
∪ · · ·
∪ (?¬ϕ1; · · · ; ?¬ϕn;∼i; ?¬ϕ1; · · · ; ?¬ϕn−1; ?ϕn)
In general we have to be careful (as is also observed in (van Benthem andLiu, 2004)). If we have a connected well-preorder then adding arrows toit in the same knowledge cell may spoil its properties. Also, the union oftwo connected well-preorders need not be connected. So here is a question:what is the maximal sublanguage of doxastic programs that still guaranteesthat the defined relations are suitable preference relations? Or should beliefrevision models be further constrained to guarantee felicitous preferencechange? And if so: how?
0732
0733
0734
0735
0736
0737
0738
0739
0740
0741
0742
0743
0744
0745
0746
0747
0748
0749
0750
0751
0752
0753
0754
0755
0756
0757
0758
0759
0760
0761
0762
0763
0764
0765
0766
0767
0768
0769
0770
0771
0772
0773
0774
98 J. van Eijck
6 Examples
Global amnesia: the event of agent a (Jason Bourne) forgetting all hisbeliefs, with everyone (including himself) being aware of this, is representedby the following action model (for the case of three agents a, b, c):
Alzheimer: the event of agent a forgetting everything, with the othersbeing aware of this, while a wrongly believes that nothing has happened. Itis tempting to model this with the following update model:
Note however that this does not satisfy the constraint on relation update(the two actions are connected, but the substitution for a is not the same),so it may result in incoherent models.
Lacular amnesia (specific forgetting): forgetting everything about p.One way to model this is by means of an action model with a single action,accessible to all, with the relational substitution
≥i 7→ (≥i ∪ ∼i; ?¬p)∗
This will effectively add best-world arrows from everywhere in the knowl-edge cell to all ¬p worlds.
Confession of faith in p, or publicly accepting p: an action modelwith a single action, accessible to all, with the relational substitution
≥i 7→ (≥i ∪ (?¬p;∼i; ?p))∗.
This will make the p worlds better than the ¬p worlds everywhere.
Submission to a guru: the act of adopting the belief of someone else,visible to all. A problem here is that the guru may know more than I do, sothat the guru’s preferences within my knowledge cell may not be connected.This means that the substitution ≤i 7→ ≤j—the binding that expressesthat i takes over j’s beliefs—may involve growth or loss of knowledge for i.Consider the example of the wine-drinking Alexandru again: if Jan were totake over Sonja’s beliefs, he would lose the information that Alexandru isdrinking an alcoholic beverage.
0775
0776
0777
0778
0779
0780
0781
0782
0783
0784
0785
0786
0787
0788
0789
0790
0791
0792
0793
0794
0795
0796
0797
0798
0799
0800
0801
0802
0803
0804
0805
0806
0807
0808
0809
0810
0811
0812
0813
0814
0815
0816
0817
Yet More Modal Logics of Preference Change and Belief Revision 99
Conformism: adopting the common beliefs of a certain group, visible toall: an action model with a single action accessible to all, with the followingsubstitution for conformist agent i:
≥i 7→ (≥i ∪ ≥j); (≥i ∪ ≥j)∗.
Belief coarsening: the most preferred worlds remain the most preferred,the next preferred remain the next preferred, and all further distinctionsare erased. An action model with a single action accessible to all, and thefollowing substitution for agent i:
≥i 7→ →i ∪ ?⊤.
The union with the relation ?⊤ has the effect of adding all reflexive arrows,to ensure that the result is reflexive again.
Van Benthem’s ⇑ϕ is handled by a substitution consisting of bindingslike this:
≥i 7→ (?ϕ;∼i?¬ϕ) ∪ (?ϕ;≥i?ϕ) ∪ (?¬ϕ;≥i?¬ϕ).
This is an alternative for an update with an action model that has ¬ϕ <B ϕ.The example shows that conservative upgrade is handled equally well byaction priority updating and by belief change via substitution. But beliefchange by substitution seems more appropriate for ‘elite change’. For thiswe need a test for being in the best ϕ world that i can conceive, by meansof the Panglossian formula 〈←ϕ
i 〉⊤. The negation of this allows us to defineelite change like this:
≥i 7→ →ϕi ∪ (≥i; ?[←
ϕi ]⊥).
This promotes the best ϕ worlds past all other worlds, while leaving therest of the ordering unchanged. Admittedly, such an operation could alsobe performed using action priority updating, but it would be much morecumbersome.
7 Further connections
To connect up to the work of Bonanno again, what about time? Notethat perceiving the ticking of a clock can be viewed as information update.A clock tick constitutes a change in the world, and agents can be awareor unaware of the change. This can be modelled within the frameworkintroduced above. Let t1, . . . , tn be the clock bits for counting ticks inbinary, and let C := C + 1 be shorthand for the propositional substitutionthat is needed to increment the binary number t1, . . . , tn by 1. Then publicawareness of the clock tick is modelled by:
0818
0819
0820
0821
0822
0823
0824
0825
0826
0827
0828
0829
0830
0831
0832
0833
0834
0835
0836
0837
0838
0839
0840
0841
0842
0843
0844
0845
0846
0847
0848
0849
0850
0851
0852
0853
0854
0855
0856
0857
0858
0859
0860
100 J. van Eijck
Thus, perception of the ticking of a clock can be modelled as ‘being intune with change in the world’. Still, this is not quite the same as the ‘nextinstance’ operator©, for the DEL framework is specific about what happensduring the clock tick, while © existentially quantifies over the change thattakes place, rather in the spirit of (Balbiani et al., 2007).
In belief revision there is the AGM tradition, and its rational reconstruc-tion in dynamic doxastic logic a la Segerberg. Now there also is a modalversion in Bonanno style using temporal logic. It is shown in (van Benthemand Pacuit, 2006) that temporal logic has greater expressive power thanDEL, which could be put to use in a temporal logic of belief revision (al-though Bonanno’s present version does not seem to harness this power). Asan independent development there is dynamic epistemic logic in the Amster-dam/Oxford tradition, which was inspired by the logic of public announce-ment, and by the epistemic turn in game theory, a la Aumann. Next to this,and not quite integrated with it, there is an abundance of dynamic logicsfor belief change based on preference relations (Spohn, Shoham, Lewis), andagain the Amsterdam and Oxford traditions. I hope this contribution hasmade clear that an elegant fusion of dynamic epistemic logic and dynamiclogics for belief change is possible, and that this fusion allows to analyzeAGM style belief revision in a multi-agent setting, and integrated within apowerful logic of communication and change.
Acknowledgements
Thanks to Alexandru Baltag, Johan van Benthem and Floor Sietsma forhelpful comments on several earlier versions of the paper, and for construc-tive discussions. Hans van Ditmarsch sent me his last minute remarks. Iam grateful to the editors of this volume for their display of the right mixof patience, tact, flexibility and firmness (with Robert van Rooij exhibitingthe first three of these qualities and Krzysztof Apt the fourth). The sup-port of the NWO Cognition Programme to the ‘Games, Action and SocialSoftware’ Advanced Studies project is gratefully acknowledged.
References
Alchourron, C.E., Gardenfors, P., & Makinson, D. (1985). On the Logicof Theory Change: Partial Meet Contraction and Revision Functions. TheJournal of Symbolic Logic, 50(2):510–530.
Balbiani, P., Baltag, A., van Ditmarsch, H., Herzig, A., Hoshi, T., & SantosDe Lima, T. (2007). What can we achieve by arbitrary announcements?
0861
0862
0863
0864
0865
0866
0867
0868
0869
0870
0871
0872
0873
0874
0875
0876
0877
0878
0879
0880
0881
0882
0883
0884
0885
0886
0887
0888
0889
0890
0891
0892
0893
0894
0895
0896
0897
0898
0899
0900
0901
0902
0903
Yet More Modal Logics of Preference Change and Belief Revision 101
a dynamic take on Fitch’s knowability. In Samet, D., ed., Proceedings ofthe 11th Conference on Theoretical Aspects of Rationality and Knowledge(TARK-2007), Brussels, Belgium, June 25–27, 2007.
Baltag, A. & Moss, L.S. (2004). Logics for Epistemic Programs. Synthese,139(2):165–224.
Baltag, A., Moss, L.S., & Solecki, S. (1998). The Logic of Common Knowl-edge, Public Announcements, and Private Suspicions. In Gilboa, I., ed.,Proceedings of the 7th Conference on Theoretical Aspects of Rationality andKnowledge, (TARK’98), pp. 43–56.
Baltag, A. & Smets, S. (2006). Conditional Doxastic Models: A Qualita-tive Approach to Dynamic Belief Revision. Electronic Notes in TheoreticalComputer Science, 165:5–21.
Baltag, A. & Smets, S. (2008a). The logic of conditional doxastic actions.This volume.
Baltag, A. & Smets, S. (2008b). A qualitative theory of dynamic interactivebelief revision. In Bonanno, G., van der Hoek, W., & Wooldridge, M., eds.,Logic and the Foundations of Game and Decision Theory (LOFT 7), vol. 3of Texts in Logic and Games, pp. 11–58. Amsterdam University Press.
van Benthem, J. (2004). Dynamic Logic for Belief Revision. Journal ofApplied Non-Classical Logics, 14(2):129–155.
van Benthem, J. & Liu, F. (2004). Dynamic Logic of Preference Upgrade.Journal of Applied Non-Classical Logics, 14:157–182.
van Benthem, J., van Otterloo, S., & Roy, O. (2006a). Preference logic,conditionals, and solution concepts in games. In Lagerlund, H., Lindstrom,S., & Sliwinski, R., eds., Modality Matters: Twenty-Five Essays in Honourof Krister Segerberg, vol. 53 of Uppsala Philosophical Studies, pp. 61–76,Uppsala.
van Benthem, J. & Pacuit, E. (2006). The tree of knowledge in action: To-wards a common perspective. In Governatori, G., Hodkinson, I., & Venema,Y., eds., Advances in Modal Logic, Vol. 6, pp. 87–106. College Publications.
van Benthem, J., van Eijck, J., & Kooi, B. (2006b). Logics of Communica-tion and Change. Information and Computation, 204(99):1620–1666.
Blackburn, P., de Rijke, M., & Venema, Y. (2001). Modal Logic, vol. 53 ofCambridge Tracts in Theoretical Computer Science. Cambridge UniversityPress.
0904
0905
0906
0907
0908
0909
0910
0911
0912
0913
0914
0915
0916
0917
0918
0919
0920
0921
0922
0923
0924
0925
0926
0927
0928
0929
0930
0931
0932
0933
0934
0935
0936
0937
0938
0939
0940
0941
0942
0943
0944
0945
0946
102 J. van Eijck
Board, O. (2002). Dynamic Interactive Epistemology. Games and EconomicBehaviour, 49(1):49–80.
Bonanno, G. (2008). Belief revision in a temporal framework. This volume.
Boutilier, C. (1994). Toward a logic of qualitative decision theory. In Doyle,J., Sandewall, E., & Torasso, P., eds., Proceedings of the 4th InternationalConference on Principle of Knowledge Representation and Reasoning (KR-94), pp. 75–86. Morgan Kaufmann.
Burgess, J.P. (1981). Quick completeness proofs for some logics of condi-tionals. Notre Dame Journal of Formal Logic, 22(1):76–84.
Chellas, B. (1984). Modal Logic: an Introduction. Cambridge UniversityPress.
van Ditmarsch, H. (2005). Belief change and dynamic logic. In Delgrande,J., Lang, J., Rott, H., & Tallon, J.-M., eds., Belief Change in Ratio-nal Agents: Perspectives from Artificial Intelligence, Philosophy, and Eco-nomics, number 05321 in Dagstuhl Seminar Proceedings, Dagstuhl. IBFI,Schloss Dagstuhl.
van Ditmarsch, H., van der Hoek, W., & Kooi, B. (2007). Dynamic Epis-temic Logic, vol. 337 of Synthese Library. Springer.
Fischer, M.J. & Ladner, R.E. (1979). Propositional dynamic logic of regularprograms. Journal of Computer and System Sciences, 18(2):194–211.
Gardenfors, P. (1988). Knowledge in Flux: Modeling the Dynamics of Epis-temic States. MIT Press, Cambridge MA.
Gardenfors, P. & Rott, H. (1995). Belief revision. In Gabbay, D.M., Hogger,C.J., & Robinson, J.A., eds., Handbook of Logic in Artificial Intelligence andLogic Programming, Vol. 4. Oxford University Press, Oxford.
Gerbrandy, J. (1999). Dynamic Epistemic Logic. In Moss, L.S., Ginzburg,J., & de Rijke, M., eds., Logic, Language and Computation, Vol. 2. CSLIPublications, Stanford University.
Girard, P. (2007). From onions to broccoli: Generalizing Lewis’s counterfac-tual logic. Journal of Applied Non-Classical Logics, 17(3):213–229. SpecialIssue on Belief Revision and Dynamic Logic.
Girard, P. (2008). Modal Logic for Preference and Belief Change. PhD the-sis, Universiteit van Amsterdam and Stanford University. ILLC PublicationDS-2008-04.
0947
0948
0949
0950
0951
0952
0953
0954
0955
0956
0957
0958
0959
0960
0961
0962
0963
0964
0965
0966
0967
0968
0969
0970
0971
0972
0973
0974
0975
0976
0977
0978
0979
0980
0981
0982
0983
0984
0985
0986
0987
0988
0989
Yet More Modal Logics of Preference Change and Belief Revision 103
Goldblatt, R. (1992 (first edition 1987)). Logics of Time and Computation,Second Edition, Revised and Expanded, vol. 7 of CSLI Lecture Notes. CSLI,Stanford. Distributed by University of Chicago Press.
Goranko, V. & Passy, S. (1992). Using the Universal Modality: Gains andQuestions. Journal of Logic and Computation, 2(1):5–30.
Grove, A. (1988). Two Modellings for Theory Change. Journal of Philo-sophical Logic, 17(2):157–170.
Halpern, J.Y. (1997). Defining relative likelihood in partially-ordered pref-erential structures. Journal of Artificial Intelligence Research, 7:1–24.
Kozen, D. & Parikh, R. (1981). An elementary proof of the completenessof PDL. Theoretical Computer Science, 14(1):113–118.
Lewis, D. (1973). Counterfactuals. Blackwell Publishing, Oxford.
Liu, F. (2008). Changing for the Better: Preference Dynamics and AgentDiversity. PhD thesis, University of Amsterdam. ILLC Publications DS-2008-02.
Plaza, J.A. (1989). Logics of public communications. In Emrich, M.L.,Pfeifer, M.S., Hadzikadic, M., & Ras, Z.W., eds., Proceedings of the 4thInternational Symposium on Methodologies for Intelligent Systems, pp. 201–216.
Prior, A. (1967). Past, Present and Future. Clarendon Press, Oxford.
Rott, H. (2006). Shifting Priorities: Simple Representations for Twenty-seven Iterated Theory Change Operators. In Lagerlund, H., Lindstrom, S.,& Sliwinski, R., eds., Modality Matters: Twenty-Five Essays in Honour ofKrister Segerberg, Uppsala Philosophical Studies Volume 53, pp. 359–384.Uppsala Universitet.
Segerberg, K. (1982). A completeness theorem in the modal logic of pro-grams. In Traczyck, T., ed., Universal Algebra and Applications, pp. 36–46.Polish Science Publications.
Segerberg, K. (1995). Belief Revision from the Point of View of DoxasticLogic. Bulletin of the IGPL, 3(4):535–553.
Segerberg, K. (1999). Two Traditions in the Logic of Belief: BringingThem Together. In Ohlbach, H.J. & Reyle, U., eds., Logic, Language andReasoning: Essays in Honour of Dov Gabbay, vol. 5 of Trends in Logic, pp.135–147. Kluwer Academic Publishers.
0990
0991
0992
0993
0994
0995
0996
0997
0998
0999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
104 J. van Eijck
Veltman, F. (1985). Logics for Conditionals. PhD thesis, University ofAmsterdam.