Generalized Update: Belief Change in Dynamic Settings

1 Introduction

An underlying premise in much work addressing the design of intelligent agents or programs is that such agents should bold beliefs about the true state of the world. Typically, these beliefs are incomplete, for there is much an agent will not know about its environment. In realistic settings one must also expect an agent's beliefs to be incorrect from time to time. If an agent is in a position to make observations and detect such errors, a mechanism is required whereby the agent can change its beliefs to incorporate new information.

Theories of belief change have received considerable at­ tention in recent years in the AI community. One crucial distinction that has come to light in this work is that between belief revision and belief update. The distinction can be best understood as one pertaining to the source of incorrect beliefs. On the one hand, an agent's beliefs about the world may sim­ ply be mistaken or incomplete, for instance, in the case where it adopts some default belief. If an agent observes that this belief is mistaken, it must take steps to correct the miscon­ ception. Such a process is know as belief revision, of which the theory of Alchourr6n, Giirdenfors and Makinson (1985; 1988) is the best-known characterization. On the other hand, an agent's beliefs, while correct at one time, may have become inaccurate due to changes in the world. As events occur and other agents act, certain facts become true and others false. An agent observing such processes or their results must take steps to ensure its state of belief reflects these changes. This process is known as belief update, as proposed by Winslett (1988) and Katsuno and Mendelzon (1991).

In this paper, we describe a semantic model for belief change that generalizes belief update to incorporate aspects of belief revision. The aim of this model is twofold: (a) to provide a unifying semantics for both revision and update that highlights the orthogonal roles both have to play in routine belief change; and (b) to provide a more compelling account of belief update to deal with observations of changes in the world that provide information about the prior world state.

There have been attempts to provide general semantics for belief change operators (e.g., (Friedman and Halpern 1994)); but often these models are such that under certain assumptions the change is a revision and under others it is an update. We argue that routine belief change should involve both update and revision, and develop a model that incorporates aspects of both; but we show that revision and update, as currently conceived, are special cases of our general operator.

The result of this union is a more robust and realistic notion of update in which observations of change can inform and agent's prior beliefs and expectations. Such observations are pervasive; consider the following example. A warehouse con­ trol agent believes it is snowing on Route 1 after yesterday's weather forecast, and expects the arrival of a number of trucks to be delayed. Now suppose a certain truck arrives, causing the agent to update its beliefs; furthermore, contrary to its expectations, the truck arrives on time. There are two possi­ ble explanations: either the truck was able to speed through the snow or it did not snow after all. If the latter explana­ tion is more plausible, current update theories cannot arrive at the desired update in a natural way. The observation of the change in the world's state (arrival of the truck) indicates that the agent's prior beliefs (e.g. , that it is snowing) were wrong. The update should not simply involve changes that reflect the evolution of the world, but should place these changes in the context of the corrected or revised prior beliefs. The agent should revise its beliefs to capture the fact that it is did not snow and adjust its expectations regarding the arrival of other trucks accordingly. Routine belief changes often involve as­ pects of revision (correcting or augmenting one's beliefs) and update (allowing beliefs about the world to "evolve").

The general model we present to capture such considera­ tions takes as a starting point the notion of ranked or structured belief sets. By ranking situations according to their degree of plausibility, we obtain a natural way of assessing degrees of belief and a very natural semantics for belief revision. Such models have been used extensively for revision (Grove 1988; Glirdenfors 1988; Boutilier 1994c). To this we add the notion of a transition or evolution from one world state to another. As proposed by Katsuno and Mendelzon (KM), updates re­ flect changes in the world, and transitions can be used to model such changes. However, in contrast to the KM model and following our earlier work (Boutilier 1994a), we assume that the relative plausibility of transitions (and hence possible updates) is not something that is judged directly; rather we assume that events or actions provide the impetus for change. The plausibility of a transition is a function of: (a) the plau­ sibility of possible causing events; and (b) the likelihood of that event having the specified outcome. In this way, we can model events or actions that have defeasible effects (which can be judged as more or less likely).

Finally, in response to an observation, an agent attempts to explain the observation by postulating conditions under which that observation is expected. An explanation consists of three components: an initial condition, an event (or action), and an outcome of that event. The key aspect of our model is the ranking of such explanations- an explanation is more or less plausible depending on the plausibility of the initial condition, the plausibility of the event given that starting point, and the plausibility of the event's outcome. The belief change that results provides the essence of the generalized update (GU) operator: an agent believes the consequences of the most plausible explanations of the observation.

Unlike other theories of update, our model allows an agent to trade off the likelihood of possible events, outcomes and prior beliefs in coming up with plausible explanations of an observation. Of course, by allowing prior beliefs to be "changed" during update, we are essentially folding belief re­ vision into the update process (as we elaborate below). We thus generalize the KM update model to work on structured (rather than flat) belief sets. Furthermore, the information required to generate such explanations is very natural.

In Section 2 we present the AGM theory of revision and the KM theory of update, emphasizing the semantic models that have been proposed and adopting the qualitative probabilistic model of (1987; 1992). In Section 3 we present our model of generalized update, with an emphasis on semantics, and contrast it with the "flat" KM model. We describe two exam­ ples to illustrate the key features of the model. In Section 4 we describe the formal relationship between revision, update and GU. We show that under certain assumptions GU satisfies the KM postulates. In addition we show that both "flat" KM update and AGM revision are special cases of GU. In partic­ ular, the connection formally verifies the intuition that AGM revision is due to changes in belief about a static world, while update reflects belief change about an evolving world.

2 Classical Belief Revision and Belief Update

Throughout, we assume that an agent has a deductively closed belief set K, a set of sentences drawn from some logical lan­ guage reflecting the agent's beliefs about the current state of the world. For ease of presentation, we assume a logically finite, classical propositional language, denoted LcPL. and consequence operation Cn. The belief set K will of­ ten be generated by some finite knowledge base KB (i.e., K = Cn(KB)). The identically true and false propositions are denoted T and .l, respectively. Given a set of possible worlds (or valuations over LcpL) W and A E LCPL. we denote by II All the set of A-worlds, the elements of W satisfying A. The worlds satisfying all sentences in a set K is denoted II K 11.

2.1 Belief Revision

Given a belief set K, an agent will often obtain information A not present in K . In this case, K must be revised to incorporate A. If A is consistent with K, one expects A to simply be added to K: we call K1 = Cn( K U {A}) the expansion of K by A. More problematic is the case when K 1- -,A; certain beliefs must be given up before A is adopted. The AGM theory provides a set of guidelines, in the form of the following postulates, governing this process. We use K.A to denote the revision of K by A.

(Rl) K~ is a belief set (i.e. deductively closed).

• (R2) A E K~.
• (R3) K~ ~ K1.
• (R4) If...,A ¢ K then K1 ~ K~.
• (R5) K~ = Cn( .l) iff f= ..,A.
• (R6) Iff= A = B then K~ = Ki:J.
• (R7) K~AB ~ (K~)~ .
• (RS) If -,B ¢ K~ then (K~)~ ~ K~AB·

Unfortunately, while the postulates constrain possible revi­ sions, they do not dictate the precise beliefs that should be re­ tracted when A is observed. An alternative model of revision, based on the notion of epistemic entrenchment (Glirdenfors 1988), has a more constructive nature. Given a belief set K , we can characterize the revision of K by ordering beliefs according to our willingness to give them up. If one of two beliefs must be retracted in order to accommodate some new fact, the less entrenched belief will be relinquished, while the more entrenched persists.

Semantically, an entrenchment relation (hence a revision function) can be modeled using an ordering on possible worlds reflecting their relative plausibility (Grove 1988; Boutilier 1994c). However, rather than use a qualitative ranking re­ lation, we adopt the presentation of (Spohn 1987; Gold­ szmidt and Pearl 1992) and rank all possible worlds using a K-ranking. Such a ranking K : W -+ N assigns to each world a natural number reflecting its plausibility or degree of believability. If K( w) < K( v) then w is more plausible than v or "more consistent" with the agent's beliefs. We insist that K- 1 (0) # 0, so that maximally plausible worlds are assigned rank 0. These maximally plausible worlds are exactly those consistent with the agent's beliefs; that is, the epistemically possible worlds according to K are those deemed most plausi­ ble in K (see (Spohn 1987) for further details). We sometimes assume K is a partial function, and loosely write K( w) = oo to mean K( w) is not defined (i.e., w is not in the domain of K, or is impossible).

Rather than modeling an agent's epistemic state with a "flat" unstructured belief set K, we use a K-ranking to capture objective beliefs K as well as entrenchment information that determines how an agent will revise K. An epistemic state,.. induces the (objective) belief set

The ranking ,.. also induces a revision function: to revise by A an agent adopts the most plausible A-worlds as epistemically possible. Thus, using min( A, ,.. ) to denote this set, we have

If IIAII n W = 0, we set min( A,,..)= 0 andKA = LcPL (the inconsistent belief set). It is normally assumed that II All n W :f. 0 for every satisfiable A - thus every proposition is accorded some degree of plausibility. It is well-known that this type of model induces the class of revision functions sanctioned by the AGM postulates (Grove 1988; Boutilier 1994c; Goldszmidt andPearl1992). 1

The ranking function ,.. can naturally be interpreted as char­ acterizing the degree to which an agent is willing to accept certain alternative states of affairs as epistemically possible. As such it seems to be appropriate for modeling changes in be­ lief about an unchanging world. The most plausible A -worlds in our assessment of the current state of affairs are adopted when A is observed.

As an example, consider the ranking shown in Figure 1 (a), which reflects the epistemic state of someone who believes her book and glasses are on the patio. If she were to learn that in fact her book is inside, she would also believe her glasses are inside, for the most plausible Inside(B)-world (,.. = 1) also satisfies Inside( G): she strongly believes she left her book and glasses in the same place.

2.2 Belief Update

Katsuno and Mendelzon ( 1991) have proposed a general char­ acterization of belief update that seems appropriate when an agent wishes to change its beliefs to reflect changes in, or evolution of, the world. The KM theory is also captured by a set of postulates and an equivalent semantic model. We describe update in terms of a knowledge base KB rather than a deductively closed belief set K.

If some new fact A is observed in response to some (un­ specified) change in the world (i.e., some action or event occurrence), then the fonnula KB o A denotes the new belief set incorporating this change. The KM postulates governing admissible update operators are:

• (Ul) KBoA I= A
• (U2) If KB I= A then KB o A is equivalent to KB
• (U3) If KB and A are satisfiable, then KB o A is satisfiable
• (U4) If I= A::: B, KB1 ::: KB2 then KB1 o A::: KB2 o B
• (US) (KBoA) 1\ B I= KBo (A 1\ B)
• (U6) If KB o A I= B and KB o B I= A then KB o A ::: KB o B
• (U7) If KB is complete then (KB o A) 1\ (KB o B) I= KB o (AV B)

(US) (KB1 V KB2) o A::: (KB1 o A) V (KB2 o A)

The equivalent semantic model of KM sheds more light on the intuitions underlying update. IIKBII represents the set of possibilities we are prepared to accept as the actual state of affairs. Since observation A is the result of some change in the actual world, we ought to consider, for each possibility w E IIKBII, the most plausible way (or ways) in which w might have changed in order to make A true. To capture this intuition, Katsuno and Mendelzon propose a family of preorders {::;w: w E W}, where each ::;w is a reflexive, transitive relation over W. We interpret each such relation as follows: if u ::;w v then u is at least as plausible a change relative tow as is v. Finally, afaithfulness condition is imposed: for every world w, the preorder ::;w has w as a minimum element; that is, w <w v for all v :f. w. Naturally, the most plausible candidate changes in w that result in A are those worlds v satisfying A that are minimal in the relation ::;w. The set of such minimal A-worlds for each relation ::;w. and each w E IIKBII, intuitively capture the situations we ought to accept as possible when updating KB with A. In other words,

1 We refer to (Boutilier l994c; Friedman and Halpern 1994) for a discussion of languages with which one can express properties of belief sets and revision functions.

where min( A, ::;w) is the set of minimal elements in IIAII (w.r.t. ::;w).

If the orderings ::;w are total preorders (so that all ele­ ments are comparable), then update operators are charac­ terized by (U1}-(U9) (see (Katsuno and Mendelzon 1991; Boutilier 1994a)):

We assume for the most part that we are dealing with such total update operators (but we discuss this further in Section 4). It should be clear how this (total) model can be recast in tenns of ,..-rankings: we simply associate a ranking "'w with each world w (such that ,..;;; 1 (0) = {w}) and use min(A,,..w) to update by A.

As a concrete example, suppose that someone observes that the grass in front of her house is wet. Prior to the observation she believed that she left her book outside on the patio and that the grass and book were dry (see KBin Figure 1 (b)). As shown in the figure, the most plausible evolution of the epistemically possible world w, given the wet grass, is v; hence she believes her book got wet too. This may be due to the fact that the most likely cause of wet grass is rain, which dampens things on the patio as well. A less plausible transition (world u) is caused by the sprinkler being activated. However, had she observed dryB in addition to wetG, she would have accepted this explanation (and its consequences, such as her glasses being dry if they are with her book).

3 Generalized Update

One difficulty with the KM theory of update is that it does not allow an observation to force revision of an agent's beliefs about the state of the world prior to the observation. This is a crucial drawback, for even though one may not care about outdated beliefs directly, information gained about one's prior state ofbelief can influence updated beliefs. Even simple tasks such as modeling information gathering actions are beyond the scope of KM update. Consider, for example, Moore's (1985) litmus test: the contents of a beaker are unknown and one dips litmus paper into it to determine if it is an acid or a base. The prior state of belief consists of two possible worlds (acid and base) and the color of the paper after the test action should rule out one of the possibilities. Unfortunately, the KM theory does not allow this to take place; the semantics of update requires that both prior possibilities be updated to reflect the observed color (e.g., blue). One is forced to accept that, if the contents were acidic (in which case it should tum red), some extraordinary change occurred (the test failed, the contents of the beaker were switched, etc.). 2

We can relax the KM update model to allow certain KB­ worlds to be ruled out if the observation is not reachable through any reasonable transition from that world. But we must go further. It may be that an observation "conflicts" with all KB-worlds. To continue the example, imagine the contents of the beaker are not unknown, but are believed to be acidic. If the test result is blue the agent should revise its beliefs about the contents of the beaker. In order to do this, we must extend the model of update to deal with structured or ranked belief sets so that we have some guidance for the revision of our beliefs. In general, belief change will involve certain aspects of both revision and update.

Rather than generalizing the KM update semantics directly, we adopt the approach of (Boutilier 1994a), where we argued that evolutions or changes in the world should not be ranked directly. We suppose that events or actions provide the im­ petus for change, and the plausibility of a given evolution is determined by the plausibility of the event that caused the change. The motivation for this approach is that users can often more readily assess the relative plausibility of an event (in a given context) and the effects of that event, as opposed to directly assessing the plausibility of an evolution. We extend this idea further by supposing that events are nondeterminis­ tic and that their possible outcomes can also be ranked. For example, an attempt to pick up a block will likely result in a world where the block is held, but occasionally will fail, leaving the agent empty-handed.

We assume a set of events E . An event maps each world into a partial ,..-ranking over worlds, e : W -+ (W -+ N) . We use "'w,e to denote the ranking e(w). Intuitively, "'w,e(v) describes the plausibility that world v results when event e occurs at world w. We say v is a possible outcome of e at w iff Kw,e ( v) is defined; thus "'w,e only ranks the possible outcomes of e. We call this evolution of w into v a transition, which we write w ~ v. We occasionally assume the existence of the null eventn, such that "'w,n(w) = 0 and "'w,n(v) = oo if w =f:. v. The null event ensures (with certainty) that the world does not change.

2 Note that one cannot escape the dilemma by supposing there is 110 such transition, for postulate (U3) ensures that updating acid by blue is consistent (Boutilier 1994a).

Since an agent making an observation will often not know a priori what event caused an observation, we assume that each world has associated with it an event ordering p,( w) that describes the plausibility of various event occurrences at that world. Formally, p, : W -+ ( E -+ N); we write "'w to denote the ranking p,( w ). Intuitively, "'w( e) captures the plausibility of the occurrence of event e at world w. Again, we assume "'w is a partial function over E, with "'w(e) = oo taken to mean that e cannot occur at w.

We now describe generalized update.

Def. A generalized update model has the form M = {W, ,.., E , p,), where W is a set of worlds,,.. is a ,..-ranking over W (the agent's epistemic state), Eisa set of events (mappings ,..w,e over W), and J.L is an event ordering (a set of mappings "'w over E). We assume that K is the belief set induced by "'·

In summary, an agent must have information about the nature of the world(,..), what is likely to happen or not (J.L), and the effects of those occurrences (E). Such models contain the in­ formation necessary to update Kin response to an observation A; we denote the resulting belief set K.4.

To begin, we suppose that one "tick of the clock" has passed and that the agent must update its ranking ,.. to reflect the possible occurrence of certain events, without the benefit of observation. Intuitively, the posteriorplausibility of a world v depends on the plausibility of the transitions that lead to v. The plausibility of a transition w ~ v depends on the plausibility of w, the likelihood that e occurred, and the likelihood of outcome v given w, e. In other words: 3

With this in hand, an updated ranking ,._o can be given by

This epistemic state essentially captures the notion that the world has evolved one "step" but that the agent has no in­ formation about the nature of this transition (other than that contained in the model M). We note that the agent's actual beliefs are determined by the minimal worlds in K 0 (i.e., those v such that K 0 (v) = 0).

3 We note that this formula is the qualitative analog of the prob­ abilistic equation Pr( w ..:. v) = Pr( vjw, e) · Pr( elw) · Pr( w ). We refer to (Spohn 1987; Goldszmidt and Pearl1992) for details on the relationship between qualitative and quantitative probabilities.

As with KM update, updates usually occur in response to some observation, with the assumption that something oc­ curred to cause this observation. After observing A an agent should adjust its beliefs by considering that only the most plausible transitions leading to A actually occurred. The set of possible A-transitions is:

The most plausible A-transitions, denoted min(Tr(A)), are those possible A -transitions with the minimal K-ranking. Given that A has actually been observed, an agent should as­ sume that one of these transitions describes the actual course of events. The worlds judged to be epistemically possible are those that result from these most plausible transition:

De f. Let K be the belief set determined by update model M. The generalized update of K by A (w.r.t M) is

In other words, an agent updating by observation A believes what is true at the states that result from the most plausible A-transitions. We also have the following:

This conforms to our intuitions about the updating process: the direct update of K by A, K~, determines the same belief set as the process of first updating one's entire epistemic state K to get K 0 , and then performing belief revision of ,.,o by the observation A. Loosely, we might say (K 0 ):4 = K~.

This notion of update naturally gives rise to the notion of an explanation for observation A. We can view updating by A as a process of postulating the most likely explanations for A and adopting the consequences of these explanations as our new beliefs. Unlike update of unstructured belief sets, explanations must consider (and trade-oft) plausible initial conditions, events and event outcomes that lead to A. An explanation for A (given model M) is any triple (w, e, v) such that w ~ v E Tr(A) (which implies K(w ~ v) < oo). Thus it is possible that e occurred at w, leading to v and resulting in A. The most plausible explanations for A are those explanations with minimal K-ranking. If A is explainable (i.e., if the set of explanations is not empty), then the most plausible explanations correspond to the most plausible A-transitions: thus GU can be interpreted as an abductive process. Note, however, that Proposition 1 means we not generate explanations explicitly.

Before considering the formal properties of this model, we illustrate its nature with two examples. To keep the treatment simple, in the first example we use only deterministic events, while in the second we assume only one possible event.

Figure 2(a) illustrates the prior belief state of an agent who believes her book is on the patio (P) and that both the grass and her book are dry. However, if her book is not on the patio, she believes she has left it inside (K(In) = 1). We omit other less plausible worlds. We assume three events: it might rain, the sprinkler might be turned on, or nothing happens

(the null event). She judges Kw(null) = 0, Kw(rain) = 1 and Kw(sprinkler) = 2, so rain is more plausible than sprinkler (we assume a "global" ordering, suitable for all w ). The outcomes of these events are deterministic - in particular, both rain and the sprinkler will make the grass wet, but the book will only get wet if it rains and it is on the patio. Now, if wet grass is observed, our agent will update her beliefs to accept wetG. A consequence of this is that she will now believe her book is wet: the most likely explanation is simply that it rained. If wetG /1. dryB are both observed (for instance, if she is told the book is safe), there are two most plausible posterior worlds satisfying the observation (i.e., K( wetG /1. dry B) = 2). This corresponds to the existence of two plausible explanations: either the book is on the patio (K = 0) and the sprinkler turned on (K = 2); or the book is inside (K = 1) and it rained (K = 1). The result is that the agent is no longer sure where the book is. If we had instead set K(sprinkler) = 3, observing wetG /1. dryB would have caused the agent to believe that the book had been inside all along. The sprinkler explanation for the dry book becomes less plausible than having left the book inside. We see then that observing certain changes in the world can cause an agent to revise its beliefs about previous states of affairs. These revisions can impact on subsequent predictions and behavior (e.g., if the book is inside th.en so are her glasses). 4

A second example is shown in Figure 2(b). We assume only one possible event (or action), that of dipping litmus paper in a beaker. The beaker is believed to contain either an acid or a base (K = 0); little plausibility (K = r) is ac­ corded the possibility that it contains some other substance (say, kryptonite). The expected outcome of the test is a color change of the litmus paper: it changes from yellow to red if the substance is an acid, to blue if it is a base, and to green if it is kryptonite. However, the litmus test can fail some small percentage of the time, in which case the paper also turns green. This outcome is also accorded little plausibility (K = g). If the paper is dipped, and red is observed, the agent will adopt the new belief acid. Unlike KM update, generalized update permits observations to rule out possible transitions, or previously epistemically possible worlds. As such, it is an appropriate model for revision and expansion of beliefs due to information-gathering actions. If an out­ come of green presents two competing explanations: either the test failed (the substance is an acid or a base) or the beaker contains kryptonite. The most plausible explanation and the updated belief state depend on the relative magnitudes of g and r. The figure suggests that g < r, so the a test failure is most plausible and the belief acid V base is retained. If test failures are more rare (r < g), then this outcome would cause the agent to believe the beaker held kryptonite.

4 Relationship to Revision and Update

The analysis of the update postulates is similar to that pre­ sented in (Boutilier 1994a). There we described a model of update that used plausible events to explain the occurrence of observations, giving rise to an update operator. Only under certain assumptions does this operator satisfy the KM postu­ lates, and we argued that these assumptions are not always appropriate. The key difference here is that the abductive approach has been generalized to allow ranked outcomes of events, and more importantly, ranked belief structures. Sur­ prisingly, this has little bearing on the update postulates: the same assumptions are required. We describe these briefly, and refer to (Boutilier 1994a) for further discussion. We first note that our model satisfies a number of the KM postulates.

4 The world In-DryB-WetG at K = 3 is shown for illustration. Technically, that world has rank 1 since it occurs below, and the explanation "sprinkler and book inside" will never be adopted, unless further propositions and observations can distinguish the two worlds.

Prop. 2 If o is the GU operator induced by some GU model then o satisfies postulates (Ul), (U4), (U6), (U7) and (U9).

One key difference between the GU model and the KM model is reflected in (U2), which asserts that KB o A is equiv­ alent to KB whenever KB entails A. This cannot be the case in general, for even if KB F A, the most plausible event oc­ currence may be something that changes another proposition while leaving A true. Observing A may simply mean that the change proceeded as expected. (U2) is appropriate only if we are willing to assume persistence of propositions, that changes (are believed to) occur only if evidence for them is observed. While appropriate in some settings, this is not a universal principle suitable for belief change. Nevertheless, we can model it by assuming centered update models:

Def. AGUmodelM = (W,~,E,p}iscenteredifEcontains the null event nand ~~ 1 (0) = { n} for all w E W .

Prop. 3 lfo is induced by a centered GU model then o satisfies (U2).

The second key difference is reflected in the failure of (U3) (both (U5) and (U8) fail for related reasons), which asserts thatKBoA is satisfiable if A is. In our model, this corresponds to every A being explainable no matter what beliefs are held. GU models need not satisfy (U3). Consider the case where no event can result in an A-world (i.e., where Tr(A) = 0): the observation of A is then unexplainable, and KX = LCPL. the inconsistent belief set To prevent this, we can simply insist that every satisfiable sentence A is explainable.

Def. A GU model (W, ~. E, p} is complete iff for any satis­ fiable A E LcPL. there are w, v E W, e E E such that ~(w) < oo, ~w(e) < oo, ~w, e(v) < oo and v FA.

Prop. 4 If o is induced by a complete GU model then o satis­ fies (U3), (U5) and (U8).

In (Boutilier 1994a) we criticized (U3) as inappropriate for the update of flat belief sets. For example, if our beliefs corre­ sponded to a single world where acid is believed, (U3) forces the observation of blue to behave quite poorly (as described above). However, such a maxim is much more reasonable in generalized update. It does not force one to propose wildly im­ plausible transitions from prior epistemically possible states; instead one can revise one's beliefs to account for the obser­ vation. In this case, we simply give up the belief acid.

There are a number of systematic ways in which one can enforce the condition of completeness such as requiring the existence of "miraculous" events that can cause anything (Boutilier 1994a). In our setting, one quite reasonable condi­ tion we might impose is that all worlds have some plausibility (i.e., ~ is a total function on W) and that the null event is possible (not necessarily plausible) at each of those. The first requirement is usually assumed of epistemic states, and the second simply ensures that all worlds persist with some de­ gree of plausibility. Thus while explanations of A may be implausible they will not be impossible.

Finally, putting Propositions 3 and 4 together we have:

Thm. 5 If o is induced by a complete, centered GU model then o satisfies (UIHU9).

We note that the converse of this theorem and the preceding propositions is easy to verify, though not especially interest­ ing. Primarily, we are interested in determining the nature of belief change given information about beliefs, events and event orderings, rather than the construction of models that corroborate arbitrary operators satisfying the postulates. We also note that our characterization theorem includes (U9) be­ cause of our use of ~-rankings, which totally order events and worlds. One of the main reasons for using such rank­ ings is that they allow the scales of plausibility used to rank worlds, events and outcomes to be compared and added. In general, the use of qualitative ranking relations does not admit this flexibility unless one is willing to postulate a "metric" by which a combination of preorders can be compared. This is not a difficult task, but is somewhat more cumbersome than the approach provided here. Equivalent results should be obtainable in the more general setting however.

There are two special cases of GU that are worth men­ tioning in passing. First, we note that "plain" KM update of unstructured belief sets is easily captured in our model by the simple restriction of "' to rank worlds only as plausible (K- = 0) or impossible ("- = oo). Second, reasoning about agent-controlled action (and observations) is also possible, as indicated in the litmus example. To do so, we simply view an agent's actions as events: we associate with each action a a K--ranking "-w,a that ranks outcomes of action a at world w. We take the key difference between actions and events (at least, as far as belief change is concerned) to be that actions are within the agent's control so that it has direct knowledge of their occurrence. As such, actions need not be ranked ac­ cording to their plausibility of occurrence, nor do they need to be postulated as part of an explanation. Observations can only be explained by supposing the action had a particular (perhaps unexpected) outcome, or by revising beliefs about the initial conditions, or both. 5

We wrap up by considering how AGM belief revision can be modeled in our framework. The common folklore states that belief revision is a form of belief change suitable when the world is static or unchanging. To verify this intuition, we propose static update models.

De f. An update model M = (W, "'· E, fl) is staticif E = { n} where n is the null event n . 6

Thm. 6 If¢ is induced by a static GU model then ¢ satisfies (RIHR8).

Static event models have as the only possible transitions those of the form w ~ w with plausibility "-( w). Thus, the informal intuition about belief revision (and the AGM model} can be verified formally: AGM revision is a particular form of GU suitable for a "static" system. (The converse of Theorem 6 is easily verified.)

5 Concluding Remarks

We have provided a model for generalized belief update that extends both the classical update and revision models, com­ bining the crucial aspects of both, and retaining both as special cases. The main feature of GU is its insistence one be allowed to both revise and update one's beliefs about the world in re­ sponse to an observation.

In this paper, we have focussed exclusively on the semantics of generalized update. Appropriate representation languages for the concise expression of events (with defeasible effects), defeasible beliefs and other aspects of the model must still be developed. However, the many components of such lan­ guages are already in place, based primarily on conditional and dynamic logics, and other action languages.

One issue that has remained unexplored to a large extent is that of revising beliefs about system dynamics (event and outcome plausibilities). The GU model supposes that events and outcomes are specified independently of an agent's beliefs and are static. In general, however, one might expect an agent to have beliefs about these entities which are subject to revision. While not inconsistent with our model, a more elaborate treatment requires a language in which (defeasible) . beliefs about events, outcomes, and so on can be expressed.

5 Concurrent events and actions require special attention, how­ ever, and are beyond the scope of this paper.

6 As above we assume K is a total function on W .

Another crucial issue is that iterated updates that arise with sequences of events and observations; this introduces several complications. One is how to revise an epistemic state "' (rather than a belief set K) in response to an obser­ vation; several proposals exist for iterated revision (Spohn 1987; Boutilier 1994b; Williams 1994) but their applicability to this problem remains to be verified. A related problem is that the plausibility of a sequence of transitions need not be a function of the individual transitions; as discussed in (Fried­ man and Halpern 1994), more sophisticated update criteria are required, including judging the plausibility of sequences of transitions as a whole. If such a general semantic picture is fitted with a language with which to reason about events, we should be able to recast the GU model as a form of be­ lief revision about such "histories." Thus, the general view of explanation as a form of belief revision (Glirdenfors 1988; Boutilier and Becher 1994) can be extended to the explanation of observations in dynamic systems.

Acknowledgements: Thanks to Nir Friedman, Mois~s Goldszmidt, Joe Halpern and David Poole for helpful discus­ sions on this topic. This research was supported by NSERC Research Grant OGP0121843.