Actions, functions and failures in dynamic environments MORTEN LIND Center for Human-Machine Interaction. Report CHMI-8-2000 1.0 INTRODUCTION The overall purpose of this report is to analyse relations between action concepts and concepts of function and failure. The investigation is based on the theory of action developed by VonWright (1963) and explore the formalized framework offered by this theory to define the relations between action concepts and concepts of function and failure. The work reported here is part of the authors research on formalization of means-end concepts for the analysis of complex industrial automation systems (see also (Lind, 1994) and (Lind, 2001)). It should be noted that the work is of preliminary nature and may therefore contain material that is inconclusive. 2.0 A THEORY OF ACTION This section introduces the basic ideas of VonWrights theory of action. This theory is gradually extended in the report in order to accommodate the needs of the analysis. 2.1 Elementary Changes The starting point of VonWrights analysis of action is the observation that actions can be characterized by the change they cause in the environment. A typification of actions can then be obtained if changes can be typed. If we consider a state of affairs described by the proposition p, VonWright distinguish between the four types of elementary changes shown in table 1 where ∼p means that p is not true and the formula ∼pTp should be read ‘∼p then p’. The four types are the combinations that are logically possible. It should be noted that the notion of change implies a notion of time because the formula pTq means that p precedes q in time. VonWright actually distinguish between two interpretations of the T namely ”and then” and ”and next” referring to performances and processes respectively (VonWright, 1965)(VonWright, 1966). These distinctions will not concern us here. Note also that the state of affairs p or q are generic and may have a variety of interpretations depending on the nature of the environment and the aspect considered. In other words the meaning of p depends on how we choose to characterize state of affairs in the environment. This choice is obviously dependent on the domain and the problem under investigation. The types of elementary actions can therefore be used to describe dynamic environments where multiple representations are required in order to characterize the changes caused by the interventions of an agent. 2.2 Elementary Actions To each of the four elementary change types there are corresponding elementary action types, which can be divided into elementary interventions and elementary omissions (table 1). The distinction between interventions and omissions is used to describe actions where the agent decide to influence the environment in order to change its state into a desired end state (i.e. to intervene), and actions where the agent decide to be passive i.e. omissions where the natural dynamics of the environment bring about the desired end state. This means that not doing is 2 considered and action if it is intentional. It should however be noted that an action can only be considered an omission if the agent is capable and has the opportunity to action. In order to distinguish interventions from omissions, the formula representing actions need to be extended to the form pT[qIs] as proposed by VonWright. The formula is read as follows: the environment is initially in state p and then (T) in state q instead (I) of being in state s. The distinction between an elementary change and an elementary action is accordingly related to the fact that the environment would be in state s due to its natural dynamics (or to the intervention of another agent) instead of the desirable state q if the intervention was not done. This means that the state q cannot be the same as the state s in formulae defining the four elementary interventions. Furthermore, in the state of affairs s and q should be the same in an omission. This is indeed so in table 1. 1 TABLE 1. Elementary changes, interventions and omissions (adapted from VonWright ) Elementary change Elementary intervention Elementary omission ∼pTp p happens ∼pT[pI∼p] produce p ∼pT[pIp] let p happen pTp p remains pT[pI∼p] maintain p pT[pIp] let p remain pT∼p p disappear pT[∼pIp] destroy p pT[∼pI∼p] let p disappear ∼pT∼p p remains absent ∼pT[∼pIp] suppress p ∼pT[∼pI∼p] let p be absent It is this intentional nature of intervention and omissions that is the basis for the definition of failures introduced later. If the agent had no intentions, changes in the environment would have no significance i.e. it would not be meaningful to talk about failures2 In order to explain the nature of the four types of intervention we will assume that p represents the state ’the valve is open.’ We will first consider ~pT[pI~p] describing the action of changing a world where ∼p is true into a world where p is true. Thus ~pT[pI~p] whose generic meaning is ’the production of p’ is in our example represented by the sentence ’the valve is being opened.’ The expression pT[~pIp] describes the action ’the destroying of p.’ With the example it describes the action of closing the valve. The action pT[pI~p] describes that the world does 1 2 The term ’maintain’ is used here for convenience instead of the term ’preserve’ used by VonWright. It will be shown later that this is not entirely correct because an agent can fail by expecting something to be the case, which is in fact not turning out to be true. It is therefore necessary to distinguish between failure in intention and failure in expectation. The expectations could be about the future state of the physical environment or be about the intentions of another agent. The agent could also fail by not being capable of bringing about the end state se (defeated by the antagonist). 3 not change in the feature described by p on two successive occasions, i.e., it means ’the maintenance of p.’ In our example the intervention pT[pI~p] therefore would describe the action ’keeping the valve open.’ Finally ~pT[~pIp]describes the world unchanged in the feature described by ~p. This action therefore represent ’the suppression of p i.e. a situation where the valve is closed but will open unless an agent does not keep it closed. The valve could be, e.g., a pneumatic valve designed to open in the event of loss of air. If there is air loss the operator could manually suppress the opening of the valve by keeping it closed. 2.3 Concurrent Actions Actions can also be dependent through their conditions such that the following pairs of actions are concurrently present: Two agents affecting the same state of affairs (who win?). (maintain p, destroy p) (maintain p, let p disappear) (maintain p, p disappear) (produce p, suppress p) (produce p, p remains absent) (produce p, let p be absent) (destroy p, p remains) (destroy p, let p remain) (suppress p, p happens) (suppress p, let p happen) These are related to the counterfactual aspects of intentional actions. These dependencies, that assume the presence of two agents, mean that the action maintain p is always present concurrently with destroy p and produce p is always present together with suppress p. The reason for these constraints on the combination of the elementary actions is of course that we are only considering a single state p. Note that an agent may not know whether another agent or mother nature is responsible for the dynamics of the environment. As an example, the agent may not know that the destroying of a state p he is trying to maintain is caused by an agent having the intention to destroy p or he is fighting against the workings of natural forces3. The conditions for the actions derived from VonWright’s analysis are important because they define conditions for the mutual fitness of a set of composite actions comprising a plan or performance. If a condition is not fulfilled the action cannot be accomplished. Satisfaction of the conditions therefore ensure a smooth unbroken flow of activities. The structures obtained by the composition of elementary actions could therefore called activity flow structures4. 3 4 The natural forces in play may be devised by a design agent that has modified the natural environment according to his intentions. There is here a link to the concepts of Multilevel Flow Modeling (MFM) developed by the present author (Lind, 1994). The analysis of action concepts presented in the present report is part of a research effort to develop a more general approach to the modeling problems addressed by MFM. It is expected that MFM can be embedded in a more general framework where concepts of action (and function) play an essential role. 4 The analysis above does not tell which of the agents that succeed in realizing his intention. In order to address this question we need to know more than the theory of action of VonWright can give us. The theory only provides a catalog of the possible outcomes of the interaction. In order to determine which of the outcomes that are realized we need to know more about the causal factors involved. This is the type of problem addressed by theories of dynamics e.g. from the natural sciences, but the force dynamics of Talmy (1988) from the field of cognitive semantics should also be mentioned. Talmy’s concept of force is a metaphor to physical forces and the validity of his theory is therefore questionable when it is used to describe the interaction between humans and artifacts of even moderate complexity. Here it is often inappropriate to characterize interactions between humans and physical artifacts entirely through concepts of force. It would for example be strange to describe the starting of a motor only in terms of forces. To describe the dynamics of pressing the start button we obviously require the concept of force from physics because the operator is interacting with a spring designed to keep the button either off or on. However, this level of description, that may be relevant for analyzing problems with the switch, is quite irrelevant to the intent of the operator, which is to start the motor. Admittedly, it would be possible to go through an analysis in terms of (physical) forces to follow the causal chain from the pressing of he button to the acceleration of the motor shaft. But we are in this way led to descriptions on a too high level of detail. VonWrights analysis does not suffer from these problems. However, Talmy’s force dynamics may be used to report on how humans express themselves through natural language about their experience with the physical world and could in this way contribute to the engineering of human machine interfaces. However, the serious problem of validity is still there to be solved in order to avoid misrepresentation of the artifact in the interface. Misrepresentation is a source of failure because it increases the complexity of the operator’s task. 2.3.1 Conditions for intervention and antagonists The four elementary interventions can, as pointed out by VonWright, only be done provided some conditions are fulfilled. As an example; it is not possible to do the maintenance of p if p not is and will vanish unless maintained. Thus the action keeping the valve open is only meaningful if the valve is already open and the valve will close if the action is not done. These conditions are of a logical nature as they are intrinsic to the definition of an intervention i.e. through the meaning of the terms si, se and sh in the formula siT[seIsh] and through the constraints that are imposed on these terms. Thus the unequality of se and sh implies that the intervening agent is counteracted by nature or by other agents in the environment. The concept of an intervention therefore implies the notion of a antagonist i.e. some agency or disposition of the environment that may hinder the agent in bringing about the state se. However, according to the present definition of an intervention, the antagonist is always defeated i.e. fail because the agent obtain the intended end state of affairs. The nature of the counteraction depends on the type of intervention. Correspondences between interventions, conditions and the (defeated) antagonists are shown in the table below. 5 TABLE 2. Elementary interventions and types of antagonists Elementary intervention Condition for intervention Type of antagonist produce p p is not and will remain absent unless produced suppressor (suppress p) maintain p p is and will disappear unless maintained destroyer (destroy p) destroy p p is and will remain unless destroyed maintainer (maintain p) suppress p p is not and will happen unless suppressed producer (produce p) 2.3.2 Conditions for omission and substitutes As the four elementary omissions describe actions where the natural dynamics (disposition) of the environment by itself or by the intervention of another agent bring about the desired end state, there is no difference between the states q(se) and s(sh)in Table 1. There is then clearly a distinction between interventions and omissions. However, the difference between a mere change and an omission is not completely clarified. The difference can only be expressed if the belief or expectations behind the agent decision not to act is introduced in some way in the formulae. The concept of an omission implies the notion of a proxy because the dynamics of the environment bring about the state that the agent intended. The change is caused by mother nature or by another agency that substitute the agent. The substitute bring about the intended state instead of the agent. Note that a substitute should not be considered as a helper since this would imply concurrent intervention by the agent and that the agent was failing when trying to bring about the state intended. Failures (and helpers) will be discussed later. The conditions for omission are expressed in the formula siT[seIsh] through the equality of se and sh meaning that the end state intended is the state that will occur if the agent was not interacting with the environment. An omission then imply a decision not to act grounded in the belief that the environment will bring about the desired state or that an agent in the environment have the intention to produce or maintain it. The formula does not express these beliefs in its present form and extensions are required in order to characterize certain types of failure. These extensions are introduced later. 6 TABLE 3. Elementary omissions and types of substitute Elementary omission Condition for omission Type of substitute let p happen p is not and will happen producer (produce p) let p remain p is and will remain maintainer (maintain p) let p disappear p is and will disappear destroyer (destroy p) let p be absent p is not and will remain absent suppressor (suppress p) 2.4 A Possible Reduction The eight elementary actions may be reduced to four by using the following equivalences: produce ∼p maintain ∼p destroy ∼p suppress ∼p let ∼p happen let ∼p remain let ∼p disappear let ∼p be absent ⇔ ⇔ ⇔ ⇔ ⇔ ⇔ ⇔ ⇔ destroy p suppress p produce p maintain p let p disappear let p be absent let p happen let p remain These equivalences give a certain freedom of expression. They may be useful in modeling situations of conflict, e.g., where a state of affairs p is maintained by one agent and another agent is suppressing p. Logically the same situation could also be described as a situation where the second agent is maintaining ∼p. In the latter case the description of the situation would include both p and ∼p and it would not be immediately obvious from the description that the agents are in conflict. The existence of a conflict would also depend on whether both p and ∼p can be true at the same time. This is not possible in binary valued logic but is perfectly possible in multi-valued logics or in fuzzy logic that allows degrees of truth. A reduction of the four action types to two as shown in Table 4, seems on these grounds to be problematic5. 5 A firm conclusion on this point is not attempted at the moment as more analysis of the problem is required. The analysis of failure types later in the paper take advantage of the reduction and may therefore later turn out to be incomplete. 7 TABLE 4. Elementary changes, interventions and omissions (reduced) Elementary changes Elementary intervention Elementary omission ∼pTp p happens ∼pT[pI∼p] produce p ∼pT[pIp] let p happen pTp p remains pT[pI∼p] maintain p pT[pIp] Let p remain 2.5 Two Agents VonWright extend the analysis of actions to include also the description of interaction between two or more agents. The description of an interaction requires a deal more state descriptions in order to properly define the intentionality. We need to know the following when we have two agents A and B interacting: si : the intial state of affairs sAB : the actual subsequent end-state of affairs with the two agents in it sh : the (hypothetical) end state of affairs had there been no agent at all sA : the (hypothetical) end-state of affairs obtained had A been alone sB : the (hypothetical) end-state of affairs obtained had B been alone Four situations can be distinguished as shown in Table 5. TABLE 5. Four situations with none, one and two agents Agents Formula None siTsh Explanation Illustration sh The environment changes state from si to sh by its own dynamics. si sh siT[sAIsh] A si The agent A intervene and the state of the environment changes from si to sA instead of sh sA A sh B A and B siT[sBIsh] siT[sABIsh] si The agent B intervene and the state of the environment changes from si to sB instead of sh Both agents A and B intervene and the state of the environment changes from si to sAB instead of sh sB B sh si A sAB B 8 The formula introduced earlier describing the actions of a single agent siT[seIsh] is here extended in order to include the hypothetical states sA and sB to the following formula siT[sABIsh|sA,sB].We get then the elementary situations shown in Table 6. TABLE 6. The elementary situations with two agents. A alone: siT[sAIsh] (A and B): siT[sABIsh] Interventions produce p maintain p destroy p suppress p ∼pT[pI∼p] ∼pT[pI∼p] B let A produce p ∼pT[∼pI∼p] B prevents A from producing p * pT[pI∼p] B let A maintain p pT[∼pI∼p] B prevents A from maintaining p * pT[∼pIp] B let A destroy p pT[pIp] B prevents A from destroying p * ∼pT[∼pIp] B let A suppress p ∼pT[pIp] B prevents A from suppressing p * ∼pT[pIp] B let A let p happen ∼pT[∼pIp] B prevents A from letting p happen * pT[pIp] B let A let p remain pT[∼pIp] B prevents A from letting p remain * pT[∼pI∼p] B let A let p disappear pT[pI∼p] B prevents A from letting p disappear * pT[∼pI∼p] B let A let p be absent pT[pI∼p] B prevents A from letting p be absent * pT[pI∼p] pT[∼pIp] ∼pT[∼pIp] Omissions let p happen let p remain let p disappear let p be absent ∼pT[pIp] pT[pIp] pT[∼pI∼p] pT[∼pI∼p] It should be noted that all cases above marked with ”*” are changes where A fail due to the intervention of B. A fails because either because B has a conflicting 9 interest or because the consequences of B’s action prevents A from succeeding (influences the opportunities or capability for action). Examples: A and B:(p&q)T[(~p&~q)I(p&q)|(~p&q),(p&~q)] A alone:pT[~pIp] B alone:qT[~qIq] A destroys p and let B destroy q B destroys q and let A destroy p These two descriptions represent two perspectives on the same situation 1) B is a substitute and alternatively 2) A is a substitute. A and B:(p&q)T[(~p&q)I(p&q)|(~p&q),(p&~q)] A alone:pT[~pIp] B alone:qT[~qIq] A destroys p and prevents B from destroying q B let q remain and let A destroy p Note that A cannot distinguish between the vanishing of q due to the intervention by B and the disappearance of q due to natural causes. 10 2.6 Examples In order to illustrate the use of the action types developed above we will consider the manipulation of a physical object in space. The examples have a general interest because the dynamics of physical domains are described by so-called state spaces. These spaces are abstract but are usually explained and represented in terms of a physical space metaphor. In state spaces a change, and accordingly also an action, is defined by the transition between two locations. The examples are obviously also relevant for other domains where the space metaphor can be used. 6 TABLE 7. Example1: Translocation of an object Intervention move-to(x,y) Omission ? ? let-move-to(x,y) y ? y The object y move by itself to location x The object y is moved to location x x x let-stay-at(x,y) keep-at(x,y) x The object y is kept y at location x. x The object y stay by itself at location x y y ? move-from(x,y) ? let-move-from(x,y) The object y is moved away from location x keep-away-from(x,y) The object y is kept away from location x y ? The object y move by itself away from location x . x let-stay-away-from(x,y) y x ? The object y stay away by itself from location x ? y ? x y x ? Note that the eight types of action in Table 7 implicitly assume two roles, a spatial location and a movable object. Most actions include additional roles for their definition. The roles define opportunities for objects and agents. The fulfillment of a role may be conditional on a state of affairs (see also Lind, 2001). 6 Full lines show the result of an intervention. Dotted lines show natural dynamics of the object 11 TABLE 8. Example 2: Putting in and getting out an object from an enclosed space. Intervention put-in(x,y) Omission let-get-in(x,y) y The object y is put into x The object y is put into x by itself keep-in(x,y) let-stay-in(x,y) The object y is kept in x y get-out(x,y) The object y is kept out of x The object y stay inside x by itself y let-get-out(x,y) The object y is taken out of x keep-out(x,y) y y y The object y get out of x by itself . let-stay-out(x,y) y y The object y stay out of x by itself Note, as above, that the eight types of action in Table 8 implicitly assumes the introduction of two roles, here a container and a contained-object. 2.7 Sequences of Elementary Actions The elementary actions can be combined into sequences. However, due to strong logical relations between the elementary actions the number of possible sequences is restrained. The number is extremely limited because each action has a condition to be satisfied and because that condition should be the result of the preceding action in the sequence. This can be derived from the so-called ‘praxis triangle’ shown above. Since we have eight possible actions according to VonWrights theory we can expand and translate the triangle into the graph shown below (each arc label corresponds to the top of a triangle). It is seen that the possible sequences may include both interventions and omissions. 12 produce p (~p) let p happen (p) let p be absent (~p) suppress p (p) ~p p let p rem ain (p) m aintain p (~p) let p disappear (~p) destroy p (p) FIGURE 1. State transformations created by the elementary actions. The following two examples demonstrate the combination of elementary actions into sequences. ~p A A produce p p m aintain p p let p rem ain A p = ’location of A is x’ move-to keep-at ~p A A A table table move-to produce p p = ’location of A is x’ let-stay-at FIGURE 2. Example sequences of elementary actions from the blocks world. 2.8 Contradictory and antonymic states Transitions between states are in VonWrights formulation contradictory. However states can also be defined as alternative antonymic states. Thus a transition between a states p and q will be described by the formula pTq 13 which is equivalent to (p&~q)T(~p&q) 2.9 State Spaces State spaces can be constructed by composing atomic states. There may be states that cannot be obtained at the same time or that are logically related. States may also be decomposed into state descriptions on a more fine grained level. Note that state and action descriptions are interchangeable. Thus, omissions where an agent let the environment maintain (or suppress) a state can also be described as a state of the environment. Consider the blocks world example where a put-down action (an intervention) is followed by a let-stay action (an omission). However, the omission let-stay is equivalent to the state on-table. The choice of description depends on the point of view or focus. If we describe the situation in state terms we focus on the block whereas an action description would focus on the agent. 14 3 INTERPRETATIONS The definition of action types presented above was based on four so-called elementary changes, which were distinguished exclusively on the basis of logical form. Further distinctions are possible and necessary for describing interactions in complex systems. The action types have many interpretations depending how states are defined i.e. the semantics of the terms that enter into the proposition p. Any statement of a proposition must contain words for terms and at least for one relation (Langer(1953): a logical picture of a state of affairs). Depending of the meaning of the terms and the relation various types of states can be defined. For each type of state a set of corresponding elementary actions (and changes) can be defined. 3.1 Types of Change The distinction between changes that are caused by an action and a change in the action itself is important. Watzlawick (1974) distinguish between first and second order changes. By a change in the action we mean for example the disappearance of an opportunity to action. The disappearance of the opportunity may be due to an intervention, or to the omission of an agent intentionally letting the opportunity disappear. As another example of changes in the action we could mention a change in its state of performance. These distinctions are of fundamental importance for the understanding of the functional organization of complex artifacts like industrial production plants and play a vital role in e.g. diagnosis of malfunction and planning of control actions. It is obvious that a logic of changes in the action can be described within the framework of VonWrights analysis of elementary actions. However, an application of the analysis require a detailed analysis of what we actually mean by ’state of affairs’, a notion that is left unexplored by VonWright. 3.2 State of Affairs This subject is a complex one. However, the few selected distinctions discussed below illustrate the possibility of a variety of state types. 3.2.1 States of being existence (p = x exist: x is an object) produce ’x exist’(provide x) maintain ’x exist’(conserve x) destroy ’x exist’(eliminate x) suppress ’x exist’ let ’x exist’ happen let ’x exist’ remain let ’x exist’ disappear let ’x exist’ be absent presence (p = x is at s: s is a location in space) produce ‘x is at s’ maintain ‘x is at s’ destroy ‘x is at s’ suppress ‘x is at s’ 15 let let let let ‘x ‘x ‘x ‘x is is is is at at at at s’ s’ s’ s’ appearance (p = happen remain disappear be absent x appears at t: t is a point in time) produce ’x appears at t’ maintain ’x appears at t’ 3.2.2 States of action doing (p = x is done by a: x an action, a an agent) produce ’x is done by a’(initiate) maintain ’x is done’ (perform) destroy ’x is done’ (terminate) suppress ’x is done’(inhibit) let ’x is done’ happen let ’x is done’ remain let ’x is done’ disappear let ’x is done’ be absent capability (p = ’a is capable of x’: a an agent, x an action) produce ’x can be done by a’ (enable)) maintain ’x can be done by a’ (ensure) destroy ’x can be done by a’ (disable) suppress ’x can be done by a’ (impede) let ’x can be done by a’ happen let ’x can be done by a’ remain let ’x ca be done by a’ disappear let ’x can be done by a ’ be absent opportunity(p =’x is opportune for a’:a an agent, x an action) produce ’x is opportune for a´ maintain ’x is opportune for a’ destroy ’x is opportune for a’ suppress ’x is opportune for a’ let ’x is opportune for a’ happen let ’x is opportune for a’ remain let ’x is opportune for a’ disappear let ’x is opportune for a’ be absent The distinctions between the states of capability ’x can be done by a’ and opportunity ’x is opportune for a’ and the state ´x is done’ are important because the two former states provide the necessary conditions for the latter (the doing). Note that the conditions are not sufficient. The decision to act and the initiation of the action are missing. These logical relations are crucial for the ordering of actions into activity sequences. The distinctions can actually account for the overall logic structure of e.g. start-up plans of industrial process plants. 16 4.0 MULTIPLE DESCRIPTIONS The results of the analysis of action types are generic since we have not given the state of affairs p a specific interpretation. The interpretation can be chosen in each particular case depending on the problem. Multiple description of an activity can be obtained in his way. Descriptions may be organized in various ways. One interesting possibility is to order descriptions in means-end structures. In ship operation we can use the categories on several task levels. A description of ship operations on several task levels is shown in Table 9. It is seen that the ship can be described on the levels of actuation, maneuvering, track following and navigation. The levels are distinguished by the goals to be obtained, the disturbances to be counteracted and by the time horizon. The world model on each level can be used to propose plans (sequences of actions). The goal to be obtained is defined in terms of the level above. TABLE 9.The elementary actions applied on ship operations (tentative). Action type move-to Actuation Obtain rudder angle or propeller speed. Task Maneuvering Tracking Obtain given Reach given track position, course or speed. stay-at Keep rudder angle or propeller speed. Keep position, speed or course. Keep on track Stay at landmark move-from Change from actual rudder angle or propeller speed. Avoid rudder angle or propeller speed. Change from actual position, speed or course. Leave actual track Move from landmark Avoid position, speed or course. Avoid track Avoid landmarks and obstacles. Drifting to desired position, speed or course. Move to desired track by maneuvering or drifting.* Move to desired landmark by tracking or drifting.* stay-away-from let-move-to Drifting to desired rudder angle or propeller speed. let-stay-at Drifting on desired Drifting on desired Keep at track by rudder angle or position, speed or maneuvering or propeller speed. course. drifting* let-move-from Drifting away from actual rudder angle or propeller speed. let-stay-away-from Keep away from rudder angle or propeller speed by drifting. Navigation Reach landmark Keep on landmark by tracking or drifting* Drifting away from actual position, speed or course. Move from track Move from by maneuvering or landmark by drifting* tracking or drifting* Keep away from position, speed or course by drifting. Keep away from Keep away from track by drifting or landmark by maneuvering.* drifting or tracking navigating.* Drifting means here to let the system behave according to its own dynamics (which may involve the dynamics of the underlying levels of control). 17 5.0 FUNCTIONS 5.1 Functions and Indirect Agency When an agent intervenes in a dynamic environment we talk about direct agency. When a designer create and object with the intention to let the object perform a function under suitable conditions we talk about indirect agency. Concepts of action and function are therefore closely related. But functional concepts include both passive and active meanings. In the passive meaning we refer to entities that are objects of change (i.e. that let the changes happen to them). In the active meaning (implying the concept of indirect agency) we refer to objects that change the state of other objects. The functions of these objects are defined by the state transformation i.e. the action type. 5.2 Functions and Omissions Omissions play also an important role in understanding the concept of function. Consider the operation of an electronic switch. The switch is turned on by an operator by flipping the lever. By flipping the lever the operator intervenes in the environment. After flipping the lever the switch stays in the new position and the operator omit further action expecting that the spring in the switch to maintain the lever position. This expectation has a threefold justification: 1) the spring is actually capable of maintaining the switch position and 2) the spring has the opportunity to do it i.e. the operator has flipped the lever to the position where the spring can keep it in position and 3) the spring does it i.e. it has the proper disposition or tendency to exert force that will keep the lever in position. The expectation may be justified by the (social) fact that (design) function of the spring is ‘to maintain lever position’. These expectations may be shared by several agents. It is realized that functions are ascribed to objects in the environment in situations where the agent omit to act (let the environment do the work). Consider also the examples in Figure2. In this case we can say that the table has the function of keeping the block in a certain vertical position. This is actually the way the table is used (use function). This explains also that the gripper can release the block and omit further action. When omitting to grip the block the task of keeping the block in the vertical position is allocated to the table. 18 6.0 FAILURES The aspects of action and the distinction between capability and opportunity introduced earlier can be used to explore the ways an action can fail. For a start we can distinguish between two overall categories of failure, one relating to the possibility for action and the other to the performance of the action. Thus an action can fail because the agent had an intention to act, but did not posses the capability to do it or the environment did not offer the opportunity for action. In addition the realization of the intention i.e. the performance of the action can still fail in any of the aspects mentioned by Lind(2001) even though the required capability and opportunity is available. Thus an action can fail by being done by the wrong agent (WHO). It can fail in the result obtained (WHAT). It can fail in the manner it was done or by not using the proper means (HOW). An action can also fail by being done at the wrong point time, in a wrong place or under wrong circumstances (CONTEXT). Finally an action can fail by being initiated by a wrong cause, by being done with the wrong aim or in an improper state of mind (WHY). These types of performance failure may be relevant in different contexts. 6.1 Failure Types Action failures can be accordingly be categorized into three fundamental types as illustrated below. In the figures, s represent the actual state, G the set of goal states and O the set of possible states that can be obtained in the given environment - the opportunities. C are the states that the agent is capable of obtaining. The intersection R of O and C define all the states that can be reached by the agent. R is therefore called the reachability set. This distinction between capabilities and opportunities actually presupposes a separation of the agent and the environment by a boundary of interaction (see also Lind (2001)). We can refine our analysis of action failure types by introducing the set of goal states G. G is the set of states that are the intended consequences of the agent’s action. The resulting failure types are shown in Table 10. TABLE 10. Failure types Failure in performance The agent has the opportunity the capability to obtain all states in G but fail because actual state s obtained does belong to the goal set G. and goal the not R G⊂O∩C s G O C s∉G Failure in opportunity There is misfit between the goal set of the agent and the opportunities offered by the environment. The agent is capable of obtaining all states in G but the opportunities are not available. . R s O G C G⊂C/O s∉G 19 Failure in capability There is a misfit between the agents capabilities and intentions. The opportunities are there to obtain the states in G but they cannot be reached because the agent is not capable. G ⊂ O /C s∉G s G O R C 6.2 Interpretations The analysis of failure types presented above can be expanded tentatively in several directions by imposing different interpretations on the state of affairs s. The following distinctions can be proposed if s is interpreted as action aspects i.e. performance, modality, and circumstances representing features of an action (Lind, 2001). TABLE 11. Failure types and aspects of an action Aspect Performance Modality Interpretation O is the set of results that are given opportunity, G is the set of results aimed at, C is the set of results that is within the capability of the agent and s is the result obtained. manner means Circumstance timing location O is the set of manners that are given opportunity, G is the set of intended manners and C is the set of manners that is within the capability of the agent O is the set means that are given opportunity (available), G is the set of intended means and C is the set of means that the agent is capable of using O is the set of times that are given opportunity, G is the set of intended times for action and C is the set of times that the agent is capable of realizing O is the set of locations that are given opportunity, G is the set of intended locations and C is the set of locations that the agent is capable of using 6.2.1 An Example: Acts of communication Considering acts of communication such as sending and receiving a message we can in this way define a variety of failure types of communication acts: TABLE 12. Failures in sending a message. Aspect performance modality manner Interpretation O is the set of messages that can be received by the environment G is the set of messages that should be sent C is the set of messages that the agent is capable of sending s is the actual message sent O is the set of manners of receiving messages that are given opportunity 20 means circumstance timing location G is the set of manners of sending messages that are intended C is the set of manners of sending messages that is within the capability of the agent O is the set means for sending messages that are given opportunity (available) G is the set of intended means of sending messages C is the set of means of message sending that the agent is capable of using O is the set of times for sending messages that are given opportunity G is the set of intended times for sending messages C is the set of times that the agent is capable of sending messages O is the set of location for sending messages that are given opportunity G is the set of intended locations for sending messages C is the set of locations for sending messages that the agent is capable of using TABLE 13. Failures in receiving a message. Aspect performance modality Interpretation O is the set of messages that can be sent by the environment G is the set of messages that should be received C is the set of messages that the agent is capable of receiving s is the actual message received manner means circumstance timing location O is the set of manners of sending messages that are given opportunity G is the set of manners of receiving messages that are intended C is the set of manners of receiving messages that is within the capability of the agent O is the set means for receiving messages that are given opportunity (available) G is the set of intended means of receiving messages C is the set of means of message reception that the agent is capable of using O is the set of times for receiving messages that are given opportunity G is the set of intended times for receiving messages C is the set of times that the agent is capable of receiving messages O is the set of location for receiving messages that are given opportunity G is the set of intended locations for receiving messages C is the set of locations for receiving messages that the agent is capable of using 21 5.3 Failures of Performance The changes of state caused by an agent interacting with a dynamic environment are not always successful. Sometimes the action fails because it is not producing the result intended by the agent. In other situations an agent can decide not to intervene with the environment because he expects that the state of the environment will change into a desirable end state by its own dynamics. However, such an omission can also be considered a failure when the resulting state of environment did not turn out according to expectations. The evaluation of actions in dynamic environments is accordingly not a simple task. The need to consider differences in the states obtained in the environment increases the complexity of the task. Thus, an intervention where the agent counteract the natural dynamics of the environment so that its state is maintained, is quite different from an intervention that produces a new state. These problems will be approached in the following by the derivation of a set of elementary types of successes and failures. In the formula siT[seIsh] used above to define actions it is not possible to distinguish a successful intervention from one that has failed. Also we cannot define what is to be understood by omission failure. When omitting to act the agent is taking advantage of the natural dynamics of the environment to bring about the desired end state. The agent must therefore expect that the state will be obtained and this belief must be based on some model of the environment. These problems arise because VonWright’s analysis do not make sufficient distinctions. It has a basic limitation because the action always result in the desired end state i.e. that the actual state obtained se is the state intended. We need therefore to extend VonWrights analysis with this distinction. By introducing the expectations of the agent in the formula we will also be able to distinguish between success and failures of omissions. But since an agent intervene because of an intention to bring about a state in the environment it must also be assumed that the agent at the same time also has the expectation that the state will be obtained. If the agent act but does not expect to realize his intentions he is not rational. This means that in an intervention, the expected and the intended state must be the same. By introducing the state expected by the agent we can solve the problems with failures of omissions and intervention. We can accordingly summarize the extended problem formulation in the following way: We have an agent acting in a dynamic environment. The initial state of the environment is si, the actual state after the intervention is sa instead of sh that would happen if the agent did not act. The agent has the expectation that the state after the change would be se (the result intended) This can be expressed in the modified formulae siT[sAEseIsh]. A successful intervention can then be expressed by the formula siT[pEpIsh] i.e. the state of the environment after the change was p as expected by the agent. A failed action has the form siT[pEqIsh] where p and q are not the same states. The agent failed since the state q was expected instead of sh but p happened. The intervention failed either because the agent was unable to bring about q or because his model of the environment was incorrect. A successful omission be expressed by the formula siT[pEpIp] i.e. the state p expected to obtain is the same as the state after the change (obtained without intervention). A failed omission would be expressed by siT[pEqIp] i.e. the state q 22 expected to obtain is not the same as the state p after the change (obtained without intervention). TABLE 14.Successes and failures of elementary interventions and omissions. Elementary intervention Elementary omission ∼pT[pEpI∼p] produce p ∼pT[pEpIp] let p happen pT[pEpI∼p] maintain p pT[pEpIp] let p remain pT[∼pE∼pIp] destroy p pT[∼pE∼pI∼p] let p disappear ∼pT[∼pE∼pIp] suppress p ∼pT[∼pE∼pI∼p] let p be absent ∼pT[∼pEpI∼p] failing to produce p ∼pT[∼pEpI∼p] failing to let p happen pT[∼pEpI∼p] failing to maintain p pT[∼pEpI∼p] failing to let p remain pT[pE∼pIp] failing to destroy p pT[pE∼pIp] failing to let p disappear ∼pT[pE∼pIp] failing to suppress p ∼pT[pE∼pIp] failing to let p be absent Success Failure Note that in all failed interventions the expected state is complement to the state that actually is brought about i.e. we have the patterns ~pEp or pE~p in the action formula. In all failed omissions the state expected by the agent after the change is the same as the state that would obtain if the action were not done. 5.4 Turning a Failure Into a Success An agent can turn a failure into a success by revising his expectations. In retrospect the agent may realize that his expectations were not fulfilled and change his mind to accept the new situation. Note that the failures are then changed into successful omissions. TABLE 15. Failures turned into successes by revision of expectations. ∼pT[∼pEpI∼p] failing to produce p pT[∼pEpI∼p] failing to maintain p => => ∼pT[∼pE∼pI∼p] let p be absent pT[∼pE∼pI∼p] let p disappear 23 pT[pE∼pIp] failing to destroy p => pT[pEpIp] let p remain ∼pT[pE∼pIp] failing to suppress p => ∼pT[pEpIp] let p happen ∼pT[∼pEpI∼p] failing to let p happen => ∼pT[∼pE∼pI∼p] let p be absent pT[∼pEpI∼p] failing to let p remain => pT[∼pE∼pI∼p] let p disappear pT[pE∼pIp] failing to let p disappear => pT[pEpIp] let p remain ∼pT[pE∼pIp] failing to let p be absent => ∼pT[pEpIp] let p happen 5.5 Help as Failure Avoidance An agent B can sometimes intervene in a situation with the intention of preventing another agent A from failing. In such situations we can say that the agent B help the agent A. Different types of help can be derived systematically from the failure types in Table 14. The result is shown in Table 16. TABLE 16. Types of help. A alone: siT[sAIsh] (A and B): siT[sABIsh] Interventions failing to produce p failing to maintain p failing to destroy p failing to suppress p ∼pT[∼pEpI∼p] ∼pT[∼pEpI∼p] ∼pT[pEpI∼p] pT[∼pEpI∼p] pT[∼pEpI∼p] pT[pEpI∼p] pT[pE∼pIp] pT[pE∼pIp] pT[∼pE∼pIp] ∼pT[pE∼pIp] ∼pT[pE∼pIp] ∼pT[∼pE∼pIp] B let A fail to produce p B help A to produce p B let A fail to maintain p B help A to maintain p B let A fail to destroy p B help A to destroy p B let A fail to suppress p B help A to suppres p Omissions failing to let p happen failing to ∼pT[∼pEpI∼p] pT[∼pEpI∼p] ∼pT[∼pEpI∼p] B let A fail to let p happen ∼pT[pEpI∼p] B help A to let p happen pT[∼pEpI∼p] B let A fail to let p remain 24 let p remain failing to let p disappear failing to let p be absent pT[pEpI∼p] B help A to let p remain pT[pE∼pIp] B let A fail to let p disappear pT[pE∼pIp] pT[∼pE∼pIp] ∼pT[pE∼pIp] ∼pT[pE∼pIp] ∼pT[∼pE∼pIp] B help A to let p disappear B let A fail to let p be absent B help A to let p be absent Note that control actions can be seen as means of helping the designer to realize his intentions in an uncertain world. There is a direct mapping between action types and types of control action as shown below (see also Lind, 1994). However, control actions cannot be defined exclusively as instances of VonWright’s action types, it is also necessary to include acts of observation, decision and intervention i.e. the means of control. Otherwise it is not possible to distinguish between ordinary help and the kind of help provided by control agents. TABLE 17. Mapping between elementary action and elementary control actions. Elementary action Elementary control action produce steer maintain regulate destroy trip suppress interlock 6.0 REFERENCES Langer, S. K. (1953). An Introduction to Symbolic Logic. New York:Dover Publications. Lind, M. (1994). Modeling Goals and Functions of Industrial Plant. Applied Artificial Intelligence, Vol 7. Lind, M. (2001). Possibilties for Action. Report from Center of Human Machine Interaction CHMI-7-2000. Talmy, L. (1988). Force Dynamics in Language and Cognition. Cognitive Science, vol 12, 49-100. Watzlawick, P. (1974). Change - Principles of Problem Formation and Problem Resolution. New York: W. W. Norton & Company. VonWright, G. H. (1963). Norm and Action. London: Routledge & Kegan Paul. VonWright, G. H. (1965). ”And Next”. Acta Philosophica Fennica, 18, pp. 293-304. VonWright, G. H. (1966). ”And Then”. Commentationes Physico-Mathematicae, Vol. 32, Nr. 7, pp.1-11. 25
© Copyright 2026 Paperzz