Business & Finance

10 pages
6 views

A Knowledge Base for Capturing Comprehensive Mission Experience

of 10
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Share
Description
A Knowledge Base for Capturing Comprehensive Mission Experience
Transcript
  A Knowledge Base for Capturing Comprehensive Mission Experience Dennis AnderssonSwedish Defense Research Agencydennis.andersson@foi.se Abstract Crisis management, first responders and militaryorganizations are often deployed in dynamicenvironments where standard operating proceduresand set routines are insufficient to control theunfolding events. In such operations, controllers mayrely on experience from scenario-based training or  prior missions to make well-grounded decisions. Thisarticle proposes a knowledge base for capturing and retaining information pertaining to these operations, to support sharing of mission experience. The knowledgebase captures contextual information, such as decisionmaking cues, in the form of heterogeneous multimediarecordings. The proposed model allows multiple streams to be replayed simultaneously, synchronized via a Hierarchical Time Stream Petri Nets-model, tocreate an explorable hypermedia model that stimulatesaural and visual perception, enabling internalizationand thus generating new tacit knowledge. As acomplement to training, the proposed approach canhelp decision makers add to their pool of knowledgeand increase their ability of making recognition- primed decisions. 1. Introduction  Many organizations base their existence aroundstandardized procedures and predefined business models, and require improvisation mainly “to adjust to change and to consistently move products and services out the door” [ 12, p.33]. Improvisation can even be asign of organizational dysfunction, i.e. the lack of  planning [6] and poor design of the organization [38].In some organizations however, improvisation may berequired to adapt to unforeseen events. Firstresponders, crisis management and military, for instance, are often challenged with scenarios that maynot be globally unique, but that they have not beenencountered before themselves. These tacticalorganizations are regularly forced to improvise, makerapid decisions and take action based on unreliable andincomplete information. Failure for these organizationscan sometimes be detrimental to society; they thereforerequire a high degree of resilience [13, 72] to the newenvironments and situations they face. Such resilienceis often tacitly implemented by individual or organizational improvisation; as noted by e.g.Mendonça and Wallace: “improvisation is creative action in context, where the context is provided by theneed to meet goals, such as economic or human losses, while satisfying time constraints” [ 43]. Weick addsthat under certain conditions, such as the lifethreatening situation in the Mann-Gulch disaster, teamssometimes force their own conception of the situationinto one that they have experienced before [70]. On thesame line of thinking is Klein, who proposed recognition-primed decision making  (RPD) as a modelof how decision-makers use their experience to assesssituations, create plans and implement them [34].Recognition is based on goals, cues and expectancies[34], and supports organizational traits such asflexibility, adaptability, improvisation, resilience,innovation, and creativity; which are all useful for resolving dynamic and unforeseen events in tacticaloperations [5]. An implication of RPD is that thecollective experience of members in an organizationconstitutes an important portion of its organizationalmemory (OM), and that the level of experience mayhave a direct impact on their ability to perform.Argote and Miron-Spektor proposed that “Experience is what transpires in the organization as it   performs its task. Experience can be measured in termsof the cumulative number of task performances ” [ 7, p.1124]. That is, they suggest that experience isacquired by doing, and that the main mechanism for adecision maker to boost his/her pool of missionexperience is by being exposed to more missions.Although their statement may be true, Huber has notedthat the storage and representation of non-standard procedures and routines can be useful for futuredecision-making together with retrieval mechanismslike smart indexing or artificial intelligence [28] andthat that the acquisition and digitization of it can beachieved through automated computerized processes[28], i.e. a representation of experience can be digitallystored in an OM and browsed or queried by operatorsthat seek to learn to recognize situations which theyhave never encountered before. This has strong  implications to the management society since fieldexperience can be expensive and hard to acquire e.g. asconcluded from the Mann-Gulch disaster [70].Following, a model is proposed for capturingexperience from tactical operations in a format thatmay stimulate RPD when internalized, while allowingindexing mechanisms for querying and retrieval, suchas Huber described [28]. 2. Background A tactical operation, or  mission , typically consistsof one or more teams working towards the same goal,constricted by time and space, and restricted e.g. byorganizational rules of engagement. The concept of tactical operations can include a wide variety of different operations, e.g. crisis management, firstresponse, military missions, cyber operations, financialoperations, etc. Weick and Sutcliffe emphasized thatsuch operations are very sensitive to context and allfunctions therefore must be implemented to specificcontexts. They add further that even minor contextualdifferences may imply some degree of adaptation [72]. This idea is further strengthened by Weick’s findingsfrom his review of the Mann-Gulch firefighter mission[70]. On the same line, Mendonça and Wallace proposed improved training for decision makers inemergency response organizations by focusing oncreative thinking rather than standard procedures tostimulate improvisation [43]. According to their model,improvisation is based on adaptation of declarative and procedural knowledge. Thus, controllers in suchdynamic environments often need to make quick decisions based on partial understanding of their surroundings. Scenario-based training and fieldexperience are common methods of improving the controllers’ pool of knowledge, increasing their abilityof recognizing scenarios and making better decisionsunder such difficult conditions.Alternative methods of acquiring missionexperience include e.g. live, virtual and constructive(LVC) simulations [48] and serious gaming [26]. TheReconstruction & Exploration (R&E) approach [3, 4]models tactical operations as a chain of events for command and control analysis, allowing organizationsto back-track system breakdowns to find the root causeand adjust procedures accordingly to avoid repeatingerrors. Hypermedia story-telling systems have also been implemented to facilitate training, based onlessons learned from past missions. Users of suchsystems have reported that they must be integrated inday-to-day operative work and that the lessons must bedelivered at the right time since the users themselvesclaim not to be able to realize what experience theyneed to acquire until they encounter a real situation[30]. 2.1. Organizational memory Weick states that “ if an organization is to learnanything, then the distribution of its memory, theaccuracy of that memory, and the conditions under which that memory is treated as a constraint become crucial characteristics of organizing” [ 69, p.206]. Asmission experience is an integral part of the performing organization’ s OM, Weick  ’s claims show thatmaintaining accurate mission experience allows anorganization to be better prepared for future missions.Walsh and Ungson defined a framework for analyzing organizational memory in which storedinformation is contained in retention facilities (bins)distributed across the organization [68]. The bins are:individuals, culture, transformation, structures andecology [68]. The bin analogy makes it clear that thememory of an organization is decentralized anddependent on each individual in the organization.By relying on storing some knowledge inindividuals alone, an organization becomes vulnerableto corporate amnesia [37, 73]. As an alternative tonetworking and encouraged collaboration betweenemployees, organizational memory systems (OMS)have been proposed as the solution to reduce the extentof this amnesia [37, 54]. An OMS must capture, store,disseminate and facilitate the use of context dependentknowledge [1]. This knowledge can be either explicit(business rules, guidelines, manuals, databases, etc.) or tacit (individual or group experiences, e.g. ideas, facts,assumptions, meanings, questions, decisions, guesses,etc.) [15].Integrating the tacit knowledge into computerizedOM models is a known challenge that this paper addresses, by proposing a knowledge base for missionexperience. 2.2. Case-based knowledge representations Knowledge representation (KR) should not beconfused with knowledge base, nor should it beconfused with data structure [20]. The knowledgerepresentation defines how to view and reason aboutthe knowledge while the knowledge base is arepository of information stored in data structures, e.g.in a database. Some of the most common knowledgerepresentations are Expressions (Equations), Rule- based, Regular Grammars (Automata), Semantic Networks [53], Object-orientation (OO), Scripts [10,58, 59], Frames [45] and Case-based [36, 64]. In thefollowing sections, these KRs are briefly presented to provide a baseline for the proposed model.A case-based  KR stores the information in groups  related to cases, or scenarios. A case is defined as atuple of problem, solution and outcome [36, 64]. Theunderlying idea behind case-based reasoning (CBR) isthat there is more to knowledge than just semanticmemory, there is also an episodic memory [60, 66].Storing information about the problems, their solutionsand outcomes enables the users to match their current situations to previous “cases” and find the most suitable solution and possibly adapt it to better fit the problem at hand. This KR scheme gives a high degreeof freedom in what information to store and how; mostcurrent systems use textual descriptions, althoughhypermedia variants have been proposed, e.g. [18, 57].One of the benefits of CBR systems is that they donot assume complete domain knowledge; instead theystart with what information there is, and thencontinuously improve as the knowledge base is populated [17]. However, case-based reasoning does place a larger cognitive burden on the inference engine(often the user) as the task of mapping a currentsituation to the stored problems can be cumbersomeand ad-hoc, especially if the data stored in them isunstructured. In this sense Case-based reasoning isvery much like recognition-primed decision making[34]. Automating this inference is a difficult matter of defining valid distance metrics for the case problems,which is a research topic of its own.Mille and colleagues commented that CBR systemsexploiting the temporal dimension are not sonumerous, and that contexts surrounding the problemsare disregarded when reasoning about solutions andtherefore proposed trace-based reasoning  (TBR) as anextension to CBR [17, 44]. They define a trace as whatremains of a phenomenon after it has ended. Lessformal than CBR and TBR is  storytelling  [62], anarrative form of case histories and explanations [31]. 2.3. Alternative knowledge representations  Rule-based, expressions and automata KRs are alllexical in the sense that they model knowledge as a setof rules based on overarching idea that knowledge can be described by logical or mathematical expressions.These KRs are useful for modeling static knowledge,more resembling formal knowledge such as standard procedures. Although representing and managing thisknowledge can certainly be useful for tacticalorganizations, it is out of scope for this paper. Semantic networks group together different topicsusing relationships. It is generally a static model of knowledge, where inference is a matter of resolvingrelationships such as has-color or the highly debated is-a relationship [11]. Semantic networks are useful todescribe static chunks of memory, and may be usefulfor modeling formal knowledge [14, 35, 53, 56, 74].These relationships could be useful to represent higher order knowledge deduced by subject matter experts,such as cause and effect relationships between differentevents within a mission knowledge base.Modeling knowledge by use of   frames assumessome knowledge about the structure of the entities thatare to be modeled to create the slots that should befilled with information and was proposed by Minsky[45]. Frames are related to object-orientation , as aframe plays the same role as a class, i.e. representing astereotype. OO extends that idea by introducinginheritance and hierarchy. The idea of structuringknowledge like this is appealing as it enables indexingand searching which is useful for retrievinginformation. However, the diversity of operationslimits the level of detail that can be put into the framesor classes and therefore a knowledge base based onframes or OO will be more likely to be successful if theinformation stored is uniform to the inference engine. Scripts were proposed in the 70s as knowledgeabout autobiographical and stereotypical events. Scriptknowledge is generated by being repeatedly exposed toan event, and as such resembles experience nicely.There is a risk though, as has been proved, thatsubjects confuse events with similar scripts [10]. Thereis also an inherent problem that it assumesstereotypical events, which would typically moreresemble the routine tasks of an organization that thisresearch is not concerned with.A related idea is the  Memory Organization Packets  (MOP) [58, 59], that can be regarded as metascripts[64] to account for episodic knowledge of heterogeneous events. The MOPs can be regarded asgeneralized scripts where they are lifted out of context.MOPs can probably represent mission knowledgefairly well; however the lack of context isdiscouraging. 3. Story-telling of mission experience A knowledge base for missions requires both tacitand explicit knowledge pertaining to the mission, i.e.the experience of the individual controllers in theoperation. Naturalistic decision making theory is basedon the concept that cues, goals and expectancies allenable successful recognition-primed decision making[34]. Information relevant to describe, or simulate,these aspects, together with actions and the outcome of that particular operation, are integral elements of aknowledge base for mission histories. Further, a needto easily search and retrieve information from such proposed knowledge bases is anticipated [28].Existing solutions are often based on networkingand encouraging interaction and collaboration betweenemployees [23, 61]. This paper proposes an alternative  model that enables capturing of such knowledge,specifically for dynamic tactical operations in whichcontrollers depend on flexibility, adaptability,improvisation, resilience, innovation, and creativity[5]. Operations ’ sensitive to context [72] implies that aknowledge representation of mission histories must beable to represent context. This section presents firsthow context can be captured and how that affects thedecision of which knowledge representation to choose.Thereafter a solution is proposed both of how to storethe knowledge and reassemble it into entities that can be interpreted, or  internalized  [47], by the learner togenerate new tacit knowledge. 3.1. Capturing the dynamic situation A model is a simplified representation or description of a system or complex entity. It is thusinherent in the definition of models that someinformation is lost when it is created. A model of mission experience for the purpose of sharing it can, byreferring to how naturalistic decisions are made, berationalized as cues, goals, expectancies, actions,outcome and context [34]. Cues for decision makingcan be based on any sensory function, i.e. what the person sees, hears, feels, smells and tastes; as well as previous knowledge, or history. Although it is certainlyfeasible to capture data on cues for all these senses, inmost tactical operations vision and sound provide the primary cues, which are easily captured by audio andvideo recordings. Organizational goals andexpectancies are often communicated and can then becaptured in written form, while individual goals andexpectancies are harder to capture as they may be tacit.Depending on how explicit the individual goals andexpectancies are, interviews or surveys may proveuseful to capture those, as has been tried for instance inthe field of economy [41].There are multiple methods of capturing decisionsand actions, e.g. as written text observations or usingvideo. The relevant context on the other hand is muchmore challenging, as it is not always often feasible tocapture everything. Deciding what context is relevantto a future decision may be very tricky on beforehand.However, to support the visualization of an operational picture, the context information must be included [32].Context is sometimes defined as information used to characterize the situation of an entity [22]. Alsohistory may be part of context [39] which makes iteven more complicated to define what data to captureand what part of the context is relevant in a givensituation. A knowledge base with too little contextinformation is undesirable as missing information onthe stimuli may render it impossible to deduce why acontroller made certain decisions, and thus theexperience obtained by the decision-maker cannot betransferred to the learner. The other extreme, of toomuch information, can be troublesome as it will requireheavy processing for the learner to filter out therelevant cues, and make relevant conclusions. Incomplex missions it is highly unlikely that all stimulito every decision and action can be captured by anyman or system, especially considering the fact thathistory may be part of the stimuli. Consequently, thereis a high risk of producing large amounts of unnecessary information, but still not enough to cover all relevant cues. It is therefore an important designdecision whether to keep a minimalistic representationof the context and risk losing valuable cues, or to keepan extensive representation where a lot of informationmay be useless and require extra processing. As cueshave been recognized as important to the proposedknowledge base, a mission experience KR thatsupports heavy processing of large amounts of contextdata is arguably the preferred option. 3.2. Case-based mission knowledge base Story-telling KRs are suitable for this application because the narrative format lends itself well tocapturing the cues and context of actions and decisionswithin a mission [31]. Stories are also independent of data format, can focus on arbitrary unit of analysis, e.g.one controller or a team, and maintain causal ordering[49]. While the free-form story telling may very well be used to capture the experiences from controllers in amission, it lacks a mechanism for indexing andcategorizing the knowledge as is required to make itaccessible to the organization [28]. The case-based KR adds these mechanisms and is closely related to story-telling, in fact case problems and solutions are oftenrepresented by textual stories in many KB systems, e.g.[16, 42]. The problem, solution, and outcome can all bedescribed by the story of cues, expectancies, goals,decisions and actions.A case in such a knowledge base represents anepisode [32] of the operation where decisions andaction are taken. Goals and expectancies outline thesituation at hand, as well as the recorded cues andcontext. The course of events that unfold representssolution and the outcome. Since all parts can becontained within a chronological story, the case-basedknowledge base allows a user to review and tag eachcase based on his/her interpretations of the story, bothfor the purpose of indexing the OM to simplify futurequerying and for the purpose of maintainingannotations. Figure 1 shows how multiple cases, (withtheir  p roblems, s olutions and o utcomes) relate to thestory model of a mission, i.e. the mission history [5].  The equivocality of understanding context calls for a model with “rich media to resolve unanalyzableissues” [19, p.563]. It should be noted that low mediarichness technologies such as e-mail or SMS, have been found to be the communication medium of choicefor many individuals even when the equivocality ishigh [24]. Still, when it comes to learning; bothlearning score and learning satisfaction have beenreported to increase with media richness when thestudied content is of high uncertainty and equivocality[65].High media richness models can be reconstructedusing a combination of appropriate instrumentation for the foreseen cases in the mission, e.g., video, audio, photos, GPS tracks, and system logs [3, 50]. Relyingon these unfiltered data sources reduces the risk for rationalization of data, which often occurs whenspoken stories are told and retold. Many tacticalorganizations already employ such media-rich datacollection [29, 51, 52] for use in their debriefings or after action reviews (AAR) [46, 67]. Reconstructing astory model from this collected information is thus amatter of combining data feeds into a hypermediavisualization tool where it can be reviewed post hoc.As the model itself does not enforce any particular dataformat, this solution gives a high degree of freedom inhow to instrument and capture cues and context; whichis useful since tactical operations can be very diverse,even within the realms of one organization. Resourcesscarcity may also force organizations to limit theamount of data to collect and review, calling for careful thought on what data sources will yield the bestvalue for money in each particular case. This problemis a topic of its own, and a matter of balancingeconomy versus knowledge retention; andconsequently out of scope of this paper.The question remains of how to process, or explore,the reconstructed stories and cases. Literature has proposed several hypermedia visualization tools tocreate or show stories, e.g. a video-based story-tellingsystem [54] to capture the informal knowledge of anorganization, using video chunking and annotations tocreate stories that can be shared within theorganization.A similar solution is to design a multimedia playback tool capable of dealing with arbitraryformats, to visualize generic mission histories [3, 4,50]. Less generic, but arguably more powerful withintheir context, are the tailor-made AAR systems thathave previously been proposed, e.g. [8, 55]. Keel proposes a framework for grouping information andcollaborative sense-making [33], based on anIntelligent Agents representation [75]. He mentions theimportance of external representation and managementof task-related knowledge to reduce the cognitive burden for collaborating usersCroasdell, Paradice and Courtney propose ahypermedia model to support organizational memoryand learning based on high-level query and browsingcapabilities [18]. They conclude that such a modelcould protect organizations against losses of individualmemory through staff turnover [18]. 3.3. Building a story from samples While there is evidence that the theory of mediarichness does not tell the whole story of how to selectthe proper level of media richness [21], thecombination of all three of text, audio and video has been shown to yield the highest perceived usefulnessfor e-learning compared to any combination of one or two of those media types [40]. Although the referencedresearch was focused on e-learning, the proximity of that field to organizational memory and knowledgesharing makes the result interesting enough to consider the use of combined media types, hypermedia, in the pursuit of a good representation of informalknowledge, or experience, in tactical organizations.Having established that context can be represented by a model with high media richness, the questionremains how to combine media chunks into acomprehendible story. Research on such hypermediasystems is plentiful, with theories on how to link information together and allow the user to navigatethrough the dataset [18, 25, 27]. This concept looks promising to build computerized models of organizational memory from media chunks  –  however,the concept of temporal ordering was identified ascrucial for contextual awareness, and should beretained [49]. An easy and concise model providingthat synchronization is Hierarchical Time Stream Petri Nets (HTSPN) [63] where a story can be representedconsisting of three layers of synchronization: multimedia scenarios , atomic components (mediachunks), information units (video frames, audiosegments, etc). HTSPN assumes some timinginformation to trigger playback of each clip. Perhapsthe simplest and most obvious way of implementing Figure 1. A comprehensive storyrepresentation with multiple cases.   Chronological story (mission history) Case I  p s o Case II  p s o
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks
SAVE OUR EARTH

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!

x