Automation
in Crisis management
Automation
in Crisis management
An explorative research
Final version
Author: Ing. Joost Olieroock Studentnumber: 1240978
Education: RUG, Technische Bedrijfswetenschappen 1st supervisor: Dr. T.W. de Boer
2nd supervisor: Prof. dr. R.J.J.M. Jorna
Company: TNO, Human Factors
Company unit: Informatieverwerking Company supervisor: Dr. J.M.C. Schraagen
“The author is responsible for the content of the paper;
the author has full copy rights on this paper ”
Preface
This research is a last phase in obtaining a masters degree in Technical Management Science at the University of Groningen.
During my former study at the HTS (polytechnic college), I found that automation had many faces that shared my interest. Not only the technical part, but also the implementation in organisations and the relation to the human users. This interest was further instigated in my present study, Technical Management Science. As one probably knows, the rapid technological development these days has a great impact, both on organisations and the people working in organisations.
In relation to the point of contact between technology and human factors I chose some optional subjects in this direction. With this background I approached TNO Human Factors for an assignment.
In February I started the research.
My gratitude goes to TNO Human Factors for giving me the opportunity to accomplish an assignment of my interest. Especially Jan-Maarten Schraagen, Kees van Dongen and Guido te Brake.
Furthermore I would like to thank Thomas de Boer and Rene Jorna of the University of Groningen for being my supervisors. Also I want to show my gratitude to the people participating in the project of Combined Systems, especially to Stijn Oomes and Paul Burghardt.
Rob and Willy Olieroock and Jolanda Janssens thank you for reading and rectifying stuff during the process. I found that participating in a scientific environment is both challenging and demanding.
Especially when this environment consists of different professional disciplines.
Joost Olieroock.
Groningen, 5th of November 2004.
Abstract
The domain of crisis management happens to have some specific problems that challenges the development of automated systems. The organisation of all parties in a crisis situation is difficult to manage. Information obtained from these parties may be contradicting. It was stated that in crisis management making an assessment on the situation is relevant to make a decision to scale-up the organization. Due to problems of information supply, information coming from different disciplines and problems related to human biases, the interpretation of the situation is not always straightforward.
When systems become more intelligent due to technological developments, such as agent technologies, the control and relation with the human changes. Some form of support by automation to cope with the problems in crisis management would apply. Whereas automation would be completely in charge of making an interpretation of the situation, problems might occur. It is better to search for some form of shared interpretation of the situation. In others words, the function of assessing the situation can be allocated between the human and the system. In what way the functions are allocated between human and system depends on the capabilities of both. As a reaction to these problems the following research question was formulated:
In what way can parts of the assessment of the situation be allocated between humans and agents in a human-agent support tool to build situation awareness and appropriate trust?
The structure of the domain of crisis management can be regarded as open, complex and dynamic.
These properties are of influence on how actors operate in the domain. Actors use functions to operate on the domain, thereby influencing the properties of the domain. The interpretation of a situation is done by a process called situation assessment. Situation assessment consists of feature matching, which is an intuitive process and diagnosis, an analytical process. People are said to switch between both information processing strategies. When a diagnosis of a situation is made people build a story to infer missing information. Usually this is an appropriate strategy. In some cases, story building is not sufficient. A meta-recognitial strategy, like critiquing is needed to discover conflicts, uncover unreliable assumptions and test for incorrectness in the initial story.
Function allocation frameworks range from technically oriented (static function allocation) to socially oriented (team-based automation), with in between adaptive automation. As an example of static function allocation, levels of automation offer the possibility to have different configurations of automation for different uses. What configuration (level) of automation is going to be used is determined in the development stage. Static automation is said in some cases to keep the human out of the loop. Furthermore in static function allocation there is no possibility to modify a solution, for instance by critiquing. However with adaptive automation is said to cope with the problems of static function allocation, by keeping the human in the loop. This is done by dynamically configuring the automation depending on the state of the human, system and context. With the introduction of adaptive automation problems such as automation surprises occur. The human does not know what is going on, when the automation is changing the configuration.
Team-based automation might be more applicable as a metaphor to critiquing. Team members in human teams take on roles. When team members become more experienced with each other, the coordination becomes more implicit. In that time a shared mental model is formed. A shared mental model consists of different type of models. The most interesting shared mental model regarding the allocation of roles between team members is team interaction knowledge, which holds information concerning roles, responsibilities of and communication patterns between team members.
The relevance of trust was determine in the description on the problems in crisis management. After a literature study on the concept of trust, it can be said that many type of trust exists. The reason for a literature study on trust was to compare the different frameworks for function allocation in relation with trust. As a most important remark can be made that the difference between trust in static function allocation and adaptive automation is the trust that the system is competent to adapt to the situation.
Faults in static automated system do not always have a negative influence on trust. As these faults become consistent the predictability of the system is guaranteed. In adaptive automation there are more possible configurations in relation to the context to consider. For faults to become predictable is less convenient.
As a combination of the function allocation frameworks, the concept of levels of critiquing was introduced. Levels of critiquing offer the possibility to let the critiquing take place, depending on the context. The levels of critiquing range from human critical thinking, to a shared critiquing, to machine critical thinking. In shared critiquing the human and system can both take a stance (part of the story building) and do the critiquing. It is stated that the levels of critiquing can cope with the brittle character of automation and automation surprises. A solution can be critiqued. When technology further develops future use will be capable of critiquing the allocation of roles.
Table of contents
INTRODUCTION ________________________________________________________________6
1 PROBLEM FIELD ___________________________________________________________7
1.1 COMPANY FACTS __________________________________________________________7 1.2 DECIS ___________________________________________________________________7 1.3 CRISIS DOMAIN____________________________________________________________8 1.4 ACRISIS SCENARIO:FIRE IN THE HARBOUR _____________________________________8 1.5 CRISIS ORGANIZATION_____________________________________________________11 1.6 DECISION MAKING IN CRISIS MANAGEMENT ____________________________________12 1.7 PROBLEMS IN CRISIS DECISION MAKING________________________________________13 1.8 CRISIS MANAGEMENT SUPPORT______________________________________________15 1.9 TRUST__________________________________________________________________16 1.10 CONCLUSION ____________________________________________________________16
2 RESEARCH OUTLINE ______________________________________________________17
2.1 AUTOMATION AND FUNCTION ALLOCATION ____________________________________17 2.2 TRUST__________________________________________________________________17 2.3 RESEARCH GOAL _________________________________________________________18 2.4 RESEARCH QUESTION______________________________________________________19 2.5 RESEARCH MODEL________________________________________________________19 2.6 SUB QUESTIONS __________________________________________________________20 2.7 RESEARCH METHOD_______________________________________________________20 2.8 DOCUMENT STRUCTURE____________________________________________________20
3 SITUATION AWARENESS: FUNCTION DESCRIPTION_________________________22
3.1 FUNCTIONS______________________________________________________________22 3.2 PROPERTIES OF THE DOMAIN ________________________________________________23 3.3 SITUATION AWARENESS____________________________________________________24 3.4 ANALYTICAL VS.INTUITIVE INFORMATION PROCESSING __________________________25 3.5 DIAGNOSIS______________________________________________________________27 3.6 TYPES OF AUTOMATION____________________________________________________29 3.7 DEVELOPMENT OF CRITICAL THINKING TOOL ___________________________________31
4 FUNCTION ALLOCATION __________________________________________________32
4.1 STATIC FUNCTION ALLOCATION______________________________________________32 4.2 REVIEW_________________________________________________________________37 4.3 CONCLUSION ON STATIC FUNCTION ALLOCATION________________________________39
5 ADAPTIVE AUTOMATION (DYNAMIC FUNCTION ALLOCATION) _____________40
5.1 DEFINITION______________________________________________________________40
5.2 BENEFITS _______________________________________________________________41 5.3 PROBLEMS ______________________________________________________________41 5.4 DESIGN ASPECTS _________________________________________________________42 5.5 CONCLUSION ON ADAPTIVE AUTOMATION_____________________________________45
6 TEAM-BASED AUTOMATION _______________________________________________46
6.1 HUMAN TEAMS___________________________________________________________47 6.2 ROLE ALLOCATION________________________________________________________47 6.3 PROPERTIES OF EFFECTIVE TEAM FUNCTIONING _________________________________47 6.4 CRITICAL THINKING IN TEAMS_______________________________________________52 6.5 CONCLUSION TEAM-BASED AUTOMATION______________________________________53
7 TRUST AND FUNCTION ALLOCATION ______________________________________54
7.1 DEFINITION OF TRUST______________________________________________________54 7.2 FUNCTIONALITY OF TRUST__________________________________________________56 7.3 CONTENT OF TRUST _______________________________________________________57 7.4 FUNCTION ALLOCATION IN RELATION TO TRUST_________________________________61
8 IMPLEMENTATION ________________________________________________________64
8.1 CRITIQUING FRAMEWORK __________________________________________________64 8.2 INVOKING STRATEGY______________________________________________________66 8.3 META-CRITIQUING________________________________________________________68 8.4 CONCLUSION ____________________________________________________________69
9 CONCLUSION______________________________________________________________70
9.2 CRITICAL DISCUSSION AND RECOMMENDATIONS ________________________________74
10 REFERENCES ____________________________________________________________75
APPENDIX I, THE COMPLETE FITTS LIST _______ ERROR! BOOKMARK NOT DEFINED.
APPENDIX II, LEVELS OF AUTOMATION BY BILLINGS_____ERROR! BOOKMARK NOT DEFINED.
APPENDIX III, AGENT TECHNOLOGY ___________ ERROR! BOOKMARK NOT DEFINED.
Introduction Automation in crisis management
6
Introduction
This document is written as a master thesis in Technical Management Science at the University of Groningen. In the beginning of February the assignment was given out by TNO Human Factors. The goal of this research is to come to function allocation in the domain of crisis management.
As an introduction to the problem field TNO Human Factors is described in chapter one. The focus in chapter one is on identifying some problems in the domain of crisis management. To illustrate the reaction of the crisis organisation on an incident and the related problems a scenario is described.
With this scenario and the problems in mind, in chapter two a prologue is taken to a research question.
In chapter two the sub questions are described, as well as a conceptual model.
In the chapters four to six a study on the relevant literature on function allocation is described. In literature is found that the frameworks for the allocation of functions range from a technical oriented view on function allocation to a more socially view.
Chapter seven introduces a theoretical outline on trust. The subject of trust was identified in chapter one as being a relevant subject in this domain. In literature the relation between trust and function allocation is presumed, but not empirically shown. To further outline this relation, in chapter seven the theory on trust is related to the function allocation frameworks.
In chapter eight the assessment of a situation by critiquing is implemented by function allocation.
Levels of critiquing are presented, which implements a form of shared critical thinking.
Chapter nine is a concluding chapter, where recommendations are given on the subjects described in this research. The recommendations are directed to TNO Human Factors and Combined Systems.
7
1 Problem field
This chapter describes the problem field and the partners that co-operate in the problem field. In the first paragraph a short introduction to TNO Human Factors, as a market oriented research institute, is given. This is followed by a description of a cooperation TNO participates in, called Decis. Decis is further introduced by describing the project Combined Systems. In this project crisis management is chosen as a problem domain. The motivation for this choice is explained in the following sections.
Describing some problems in decision making in crisis management gives a direction in proposing solutions by applying automation.
1.1 Company facts
TNO Human Factors (TNO-HF) is part of TNO Corporate, a market-oriented research institute. At TNO-HF knowledge is developed about human behaviour and human performance in demanding environments. This knowledge is applied to the innovation of processes, products and services for the Dutch Armed Forces, private enterprises and government. Although their primary client is the Ministry of Defence, TNO-HF also focuses on civil markets. These include traffic, public safety and Information and Communication Technology. TNO-HF participates in a wide network of collaborative arrangements within TNO, NATO, EUCLID, the European Union and the academic world. TNO-HF has a multidisciplinary staff of 180 physicists, engineers, psychologists, biologists and medical doctors. TNO-HF also participates in a social-technical project called Decis.
1.2 Decis
Decis is a co-operation between four commercial and non-commercial organisations. The co- operating parties are TNO (Human Factors and “Fysisch Electronisch Laboratorium”), Thales, Technical University of Delft and the University of Amsterdam. The aim of the co-operation is to develop techniques to implement in intelligent systems. One of the research areas is collaborative decision-making. In collaborative decision making multiple decision-makers, human or software agents, make decisions in favour of a common goal. The first project in the field of collaborative decision making is a project called Combined Systems.
1.2.1 Combined Systems
The meaning of Combined Systems is to find techniques and technologies to automate decision chains in open, chaotic and complex domains. To reach this goal a case study is used to develop a system to support the human crisis team in making decisions about a crisis situation. The full domain of the project Combined Systems is pictured in figure 1-1. In figure 1-1, the grey circle is
filled with agents, showing the future intelligent system. The agents in the system gather information
Service Agents SMDS Agents
SNE Agents HCI Agents Actuators (+ sensors) Human organizations
(operational level) Distributed
Perception Networks
Human organizations (tactical / strategic level)
Other platforms Information
Sources
Figure 1-1, Full domain support of crisis management
A
Problem field Automation in crisis management
8 from several external systems and supply information to these systems. The systems can be humans (supported by technical means), in strategic and operational settings. The external systems can also be arranged as artificial systems (sensors, databases). Information is supplied by “service agents”, and is presented by the “HCI (Human Computer Interaction) agents” to the operator(s). The essential factor in relation to human-computer interaction is to model the information in a sufficient way to support human computer symbiosis. As Combined Systems has chosen the domain of crisis management, first this domain is described.
1.3 Crisis domain
It is important in such a scattered research project like Combined Systems to have a common objective. This is especially important when participating parties have their own independent research objectives. A common objective not only improves the communication between the members, it also enables them to integrate their products in a better way. To obtain a common objective all the parties have to focus on a common domain.
To offer a challenge for the development of an intelligent agent system a domain has to have certain properties and problems. Crisis management has some of these properties, for instance a high change of escalation, because of the many influences. In the description on crisis management a scenario is given first. It shows some of the problems the crisis management organisation has to cope with. They have to respond to the changing situations. Crisis response needs an ad hoc organisation, which is also described. As scaling-up is an important activity for the crisis organisation, the scaling-up and related problems are described.
1.4 A Crisis Scenario: Fire in the Harbour
Crisis management has several properties that make it an interesting research subject for the development of systems. To give an indication on the difficulties and thereby the interest for developing (intelligent) systems, a problem scenario is given. A problem scenario tells a story of current practice. It reveals some of the stakeholders and their relations with the problem domain. The scenario is based on a scenario that is used by Combined Systems. It explains a crisis in Rotterdam Harbour and it gives an indication on the escalation of the crisis and the reaction of the crisis organization by scaling-up.
In the Rotterdam Harbour an explosion is heard and fire can be seen.
Some people ashore have seen the accident. Most of them haven’t seen the accident and move closer to see what is going on.
People are taking out their cameras and start shooting photos and films. Others take out their phone and start calling. Some call their friends about what they have just seen; others make a call to 112. Because of the noise, more people are coming out to see what has actually happened. People in the neighbourhood see the smoke from far away and also start to call. The first reaction is the old factory, or maybe a terrorist attack. The old factory has been the object of kleptomaniac attacks, so it is reasonable to think of the old factory burning. There have been many complaints by residents on the slow start of the demolishment of the factory. Although people can not see the fire itself, but only the smoke, they still infer the old factory as a cause.
9 Some people infer that because there is a Eurotop conference in the area, where several presidents and other political officials are present, terrorist attack is also likely to be the cause of the fire.
The explosion is heard over a large area, and is also heard by the officers in the incident rooms in the World Port Center. It contains 32 floors, and the 20th floor contains the Harbour Coordination Center (HCC, 'Haven Coördinatiecentrum') of the Rotterdam Port Authority (RPA, 'Gemeentelijk Havenbedrijf'). The incident rooms for police, fire brigade and first aid (Regional Rescue Services Rotterdam, http://www.rhrr.nl) are located on the 21st floor.
In the several incident rooms there is a collective incident room communication system. Although this system is used during regular incidents, the incident rooms are reluctant to use it when a crisis escalates. They say this is not for technical reasons, they just think that they loose the overall picture.
The calls that the 112 emergency workers get from the harbour area are all redirected to the incident rooms on the 21st floor of the World Port Center. Some people in the neighbourhood call 112 to say that they think the old factory is on fire.
Also some messages relate to a fire on two ships that have collided.
The smoke can be seen from the Maasboulevard and some surrounding highways. People reduce their speed (or even stop) to watch the cloud. This causes a traffic jam. Although oncoming traffic is warned for the congestion and is advised to reduce speed, they do not exactly know what is going on. People in the traffic jam start to call 112.
As the exact location of the incident is still unknown the police officer on duty sends two police units, one to the north shore and one to the south shore of the harbour. The crowd on the north shore is now about 200 people large. At that moment the fire brigade arrives at the scene. What people on the shore can not see is that one of the officers starts to report to his commander in the World Port Center, while the other officers start to put out the fire. The exact location and cause of the incident is clear. It is not the old factory on fire, but two ships which have collided in the harbour. A decision has to be made on how many and what sort of rescue workers to employ. In other words, in what degree is scaling-up needed.
At that moment the first 112 call, reporting people with pain in their eyes, people fainting and having difficulty breathing, are coming through.
Other reports about a thick cloud of smoke drifting into north-east direction, due to a south-west wind are coming through.
The people in the incident room and harbour authorities are on stand-by and are waiting for information from the rescue units. They want to know if the fire can be controlled, whether two fire brigade units are enough and if they need to start collecting information from other parties. They keep a close watch on calls that are coming in through the emergency center.
This must come from the ships, as the wind is blowing from the ships on to the shore. The gas has reached the shore but everyone is still unaware that this is a toxic gas spreading.
The hazardous materials officer (HMO) is called in to assist in this situation. He wants to make a first estimate of the situation, collecting information on wind direction, wind speed, and weather type. Simultaneously he starts an inquiry about the cargo of the ships: what type of substances carried, and their amounts. Additionally, he issues an order to send substance measuring units to the incident location in order to assess the area of contamination.
The fire brigade commander is warned and takes control of the disaster control operation. He sends out a protective suit unit of the fire brigade to go to the scene. In the coordination centre a map of the area that contains information about the incident and the smoke area is used. When information from the traffic comes in it is blended into the map. Police officers see traffic around the shore area increase.
Because of unknown gas reported, cars are not allowed to cross the smoke area. People in cars need to turn around and leave the area via the A16. The routing system already blocked the 'Maasbouleveard' but people are still trying to go into that direction to see what is going on. The police is send to the 'Maasboulevard' to re-direct these people.
Problem field Automation in crisis management
10 The HMO confirms the information that a toxic gas is spreading, however he still doesn't know what sort of gas is coming out of the ship.
People that are having health problems (sore throat) request first aid by calling 112. Slowly people are moving away from the danger zone on the shore, and the situation is becoming somewhat chaotic. The crowd on the shore is now moving away from the accident. At this moment people inside buildings (in the influence zone) are warned to close windows and doors, and to shut down the air-conditioning because a possibly toxic gas is spreading in the area.
In the mean time the smoke and invisible gas is spreading over the area. People discover that they get problems when they get to close to these clouds. More people are moving out. People who are already in their car (on the highway) are advised to flee from the area. They are advised to follow the directions of the routing system. The advised speed is 50 kilometers an hour.
The health services are notified and they send ambulances to the scene. At this moment all rescue services want to have information on the gas. The health units want to know what problems to expect and how to handle these, the fire brigade wants to know what sort of gas and how to stop it from flowing out of the ships. This information however is not yet available.
At this moment it is clear to all in the incident room of the World Port Center that there might be a large impact on the harbour environment. It is decided to scale up the organization. At that moment the crisis management team is called and all members are on their way to the World Harbour Port. One of the members of the crisis management team is the mayor. He wants to be on the premises and not in his car when the first meeting is held. The police units want to know if they need to inform people in the area to close the doors and windows and stop the airco or if they have to start evacuating everyone.
An expert is called in to establish the possible cause of this incident.
The protective suit unit confirms this from reading the toxic chemical signs they found on the ship. Two extra measurement units are deployed to confirm this.
All services are notified about the exact area under threat. The first ambulances arrive at the scene and confirm that the gas is a danger for the people in the area. They confirm that more backup is needed, a call that is confirmed from all units in the field.
The expert concludes that this is probably not a terrorist attack. Evacuation of the 'Influence' zone is enough for now. The biggest building that is on the 'route' of the gas cloud is the Erasmus Building, and this is still a problem for the rescue units, they are not informed yet. The crisis team makes a list of all teams and equipment needed and informs the officers in charge.
(Source: based on “Combined scenario by Josine van de Ven”)
The scenario describes a possible crisis situation. It describes the first hour of the crisis, also called the golden hour, because of the relevance of the decisions. Decisions taken in the first hour have great influence on further escalation or de-escalation of the situation. In this scenario two boats collide, which results in a potentially dangerous situation. A quick response from several parties is needed.
As shown a large amount of time passes before a clear statement can be made on the cause of the incident. After the cause is known some time passes before there is knowledge on the type of gas that is spreading. Is the gas harmful? The “HMO” integrates several variables to assess the situation.
11 When variables are missing, for instance the properties of the cargo, additional information can be looked for. A possibility to recover the properties of the cargo is to subtract the history of the ship out of an information system.
As more emergency units become involved, more actions have to be coordinated. It becomes more difficult to control the situation. Escalation of the situation may occur when there is a possibility that toxic materials are involved. More people need help in a situation like this. In this scenario an emergency center is coordinating all activities of the emergency units. The activities of this incident coordination center do not only involve coordination around the fire fighting activities, it also involves coordination of medical assistance, reduction of casualties, organizing transport and registration of casualties, etc.
The scenario makes it clear that when a crisis occurs an organization is formed to challenge the crisis.
To give an indication on how the organisation builds up during the crisis, the following section describes a part of the organisation that is build in response to the crisis.
1.5 Crisis organization
It is clear that the crisis organization contains different professionals from different fields. During a crisis, tasks and responsibilities have to be allocated between the participants of the crisis organization. In case of the Rotterdam Harbour scenario firefighters are involved as well as police officers, ambulance personnel and port authority workers. But the organizational structure contains more than only these officers operating in the field. In the regular crisis organization structure four levels can be distinguished. The levels are described in figure 1-3 and are defined as policy level, tactical level, operational and field level.
When disaster strikes the emergency calls come in at the report centers. The report centers reside in the operational level. In this level all direct actions are coordinated. The coordination is accomplished by the “Coordination incident place”, which is located near the disaster area. Based on the incoming information from various data sources, the activities are planned and resources are allocated. The
Figure 1-2, Crisis management organisation
Problem field Automation in crisis management
12 action center is responsible for the coordination of actions outside of the disaster area, such as the movement of evacuees and the registration of the people involved.
The speed of acting on the environment ranges from non real-time on the strategic level to real-time on the operational and field level. The policy level consists of a policy team and a management team, responsible for setting the strategic direction of the crisis management operation. This consists of making a general policy and anticipating on future scenarios (van Rijk, 2002). The policy made by the policy level is prepared and achieved by the tactical level. As the tactical level is the link between the policy makers and the field level, this level determines if additional resources are needed, this means they determine if the organization needs to scale-up. The process of scaling-up is described in the following section.
1.5.1 Scaling-up the organisation
The most important strategic activity in crisis management is the principle of scaling-up. Scaling-up means that the crisis organisation is scaled to the proportions of the crisis. Scaling-up the organisation occurs when there is a necessity for employing additional resources. It involves the employment of additional resources and support. Although formal structure and procedures are used in the scaling-up process, situations occur where scaling-up is not applied efficiently. This makes the situation excessively costly, ineffective and poor to control. Because of the ambiguity, which exists in building a picture of the situation, before scaling-up the crisis organisation, research on decision making is focussed on the scaling-up process. In determining the appropriate scaling-up a classical decision loop is followed. The decision to scale-up is executed in different phases, which are described in the following section.
1.6 Decision making in crisis management
The decision making process in crisis management contains different functions. Four functions are often placed in a decision cycle.
TA DM
secondary
SA DC
primary Data
sources
Deployment
Knowledge of own assets Knowledge of
domain
IInnffoorrmmaattiioonnttrraannssffeerr
Goal
Figure 1-3, Decision-making in crisis management
13 A decision cycle that is used to define the decision-making process in crisis management is the command and control model of decision making. The description of the decision cycle is presented in figure 1-3. The full cycle consists of situation awareness (SA in figure 1-3), threat assessment (TA in figure 1-3), planning and decision-making (DM in figure 1-3) and direction and control (DC in figure 1-3). As an incident occurs, messages come in at the incident rooms. The messages that are coming through have to be interpreted to get an actual picture of the situation. This happens in the process of situation awareness. After the situation is assessed an assessment of the threat is made. The assessment of the threat is based on the knowledge of the domain. To make a decision on the level of scaling-up, one needs to know the magnitude of the crisis (how many resources one needs) and the internal structure (how many resources one has). The decision is always related to the goal. In direction and control the internal structure plays an important role, as the resources are directed to the crisis scene. The full cycle is an ongoing process, as new messages about the crisis situation come in frequently (due to the dynamic character of the domain).
1.7 Problems in crisis decision making
As was explained before, the most important decision in crisis response is the decision to scale-up the organization. This decision is in the initial stage taken by incident rooms, by supplying the first emergency workers. As the tactical level becomes involved the scaling-up is taken over by the coordination incident place. The components of the scale-up decision were explained as a decision cycle. As was said in the introduction crisis management decision making involves some difficulties.
As a crisis team in the tactical level tries to scale-up the organization, problems occur. The problems in decision making in crisis management are divided into two types, which are related to each other.
Firstly there are problems related to the supply of information about the coordination and communication. It is difficult to supply an actual and interdisciplinary picture of the incident, and the effects of an action. Due to these problems, which are related to the organizational structure, it is difficult to get a clear view on the situation. The second problem is related to the human biases in interpreting the information.
1.7.1 Problems in the supply of information.
As the organization build itself by scaling-up, different disciplines are made part of the organization.
For instance, when there are casualties at the incident, ambulance workers are included in the organization. Because of the high amount of different disciplines that the organization is build off, different information is coming through. The following problems are relevant.
Processing of the messages coming in - To select and recognize the appropriate management scenario and necessity to scale-up to the appropriate management level starts when the messages come in. Due to the different parties the organization consists of, limited or maybe conflicting information is coming through.
Problem field Automation in crisis management
14 Changing organization - The structure of the organization changes due to the scaling-up. When the organization changes it becomes more difficult to know who are part of the organization. Awareness on the organization structure is not always apparent. This was the case in the Volendam fire. Due to an incomplete picture of the situation inefficient scaling-up occurred. The other way around excessive scaling-up of resources can lead to high costs. During the fire disaster in the city of Enschede two
“command and control” centers were established at the same time. The two centers were not aware of each other’s existence. Coordination is difficult in a situation where the organization changes. Also, obtaining information is difficult when the information sources, in the form of organizational units changes.
Difficulty of practicing the decision making - Because every crisis shows a different setting, it is difficult to practice the decision to scale-up the organization. There is a large amount of time between the decision and the actual result of the implemented action.
The problems can be related to the components of the decision-making cycle in figure 1-3.
SA Conflicting messages coming through
TA Different models of the situation and the organization DM different organizational units make different plans DC Different parties execute different plans
The consequence of these problems is a lack of information on the situation and a lack of overview on the overall picture. As a consequence of the lack on overview the crisis response can not anticipate on the situation.
1.7.2 Problems related to human biases
As the interpretation of the situation is influenced by the supply of information on communication and coordination, there are some human biases that are of influence. Decision-makers in ambiguous situations infer missing information by constructing stories. Generally this is an effective strategy.
Occasionally people (even experts) tend to stick with their initial interpretation of the situation, even when contradicting information is presenting itself (framing bias) (Oomes, van Dongen & te Brake,2003). Also people tend to search for information that is confirming their initial hypotheses (conformation bias). For instance, in the scenario described above, the initial cause of the fire was to believe the old factory. People then tend to search for information that confirms the cause as a factory fire.
15
1.8 Crisis management support
One way to cope with the problems related to crisis management is changing the organisational structure. Frequently research is done on the organisational processes, mostly after events have occurred (analysis of cases). Although research on organisation changes in crisis management is effective, it is hypnotised that the problems in crisis management may be supported in another way.
With the development of intelligent systems, called agent systems, crisis management decision making can be supported by technical means. In this research the focus is not on changing the organisational structure, but on supporting decision making in crisis management by technical means.
It is stated that support of decision making by an intelligent system would reduce the problems around the interpretation of the situation. A fairly simple support configured as an information system or database can show information. One can imagine that when demands are high, for instance when many messages are coming through, the interpretation is difficult and biases are more likely to occur.
In such a situation more intelligent support is wishful. An example is the visualisation of the information in a certain way to support building an appropriate picture of the situation. In crisis management the demand on the person(s) which makes the decision is fairly high. Intelligent support during the process of scaling-up the organisation offers possibilities.
A drastic form of intelligent support is to completely eliminate the human decision-maker by an autonomous system, called the prosthetic approach. Is it wise to eliminate the human completely out of the decision making picture? In developing an intelligent system for supporting decision making, one may think the human is out of the decision making picture. But as Dekker and Woods appropriately observe, even a complete automated system almost always has a human operator somewhere at some level (Dekker & Woods, 2002).
Probably it is a better solution for building a shared awareness on the situation to let the human decision-maker do some tasks, as well as the software system. This is the concept of function allocation.
In respect with these remarks, a few questions come to mind when discussing the relation between automation and humans. One of them is on what level is the human actively involved and on what degree does automating a function negatively influences the awareness of the human? As partner in Combined Systems, TNO Human Factors and Thales together participate in a challenge called the
“human challenge”. As the name of the challenge implies, the focus is on human interaction with the agent decision support systems. The human challenge emphasis is on the symbiosis between human and machine. This research is a part of the human challenge. By supporting the human operator in rethinking a situation when ambiguity is present, awareness on the situation is created. The first research question of the project is related to building situation awareness: “How can a system of agents support people in establishing situation awareness and how is this system a support in communicating their understanding of the world to others?”
Problem field Automation in crisis management
16
1.9 Trust
When people use or work with systems this is influenced by the reliability of the system. In crisis management problems with information systems confirm this statement. Especially when demands are high, the decision to use a system becomes more relevant. As the recently introduced communication system C2000 has shown a system occasionally lacks to do what it is supposed to do, it is not used at all. Another example in crisis management is the use of the “Generall Report center Communication System (GMS)”. Usually this system is used by all report centers to communicate on an incident. In a crisis this system is not used, because the structure, processes and procedures and also the knowledge of the employees are not calibrated for crisis escalation. There is no trust in the system to supply support in a crisis situation.
1.10 Conclusion
As this chapter gave an introduction to the field of practice. This chapter also described some of the problems the field of crisis management has to deal with. It was stated that in crisis management making an assessment on the situation is relevant in making a decision to scale-up the organization.
Due to problems of information supply, information coming from different disciplines and problems related to human biases the interpretation of the situation is not always straightforward. Some form of support by automation would apply. When automation would be completely in charge of making an interpretation of the situation, problems might occur. Better it is to search for some form of shared interpretation of the situation. The next chapter is an outline to come to a research question related to developing an allocation of functions to make a shared assessment of the situation.
17
2 Research outline
In the previous chapter an outline on the project “Combined Systems” was given. A scenario was described, presenting some problems in a crisis situation. The crisis response organisation has to cope with ambiguity in information, which results in a difficulty in making an assessment on the situation.
Also human biases influence the assessment of the situation. It was stated that using automation in supporting the human in making an assessment of the situation reduces these problems. By developing a shared awareness human and machine are able to cope with the ambiguity of the information. But how is shared awareness build, what functions does it consists of and how are these functions allocated? This chapter firstly explains what automation is in relation to function allocation.
Trust in automation is further introduced. The reason to introduce these two subjects is to come to a research question. After the research question is formed, it is divided into sub questions.
2.1 Automation and function allocation
Automation has been defined in various sorts, ranging from more mechanical definitions to a focus on the information component. Parasuraman and Riley give a definition from a mechanical insight. They define automation as “the execution of a function by a mechanical machine that was usually executed by a human” (Parasuraman and Riley, 1997). Usually, when a task involves a high level of potential risk for the person that accomplishes the task automating the task is a possibility. Also when the costs of the human accomplishing the task is higher than the costs of the task execution by an automated system, automation can be efficient. In some cases automation is applied just because the techniques are available, which is not always the best solution to a problem. In a general term, automation is applied to relief the operator from physical or cognitive workload. When functions are allocated to an automated system, the burden on the operator is relieved. Some functions may apply for automation.
Although some functions may apply for automation, there are others that are more suited for human execution. This is a question of function allocation between human and machine. Many function allocation concepts have been introduced in the past. It started with allocating function to human and machine in a static way, developing to a more dynamic allocation, In dynamic allocation the allocation of the functions depends on the environmental changes and the cognitive status of the human user. The human is supported when it is most needed, when the cognitive burden becomes too high. Because machines are no longer seen as tools, but are becoming more autonomous, its is interesting to see if the recent function allocation concepts (both static and dynamic) are still applicable.
2.2 Trust
Trust has a significant influence in innovation and the implementation of a new technology (Numan, 1998). In the introduction of a new technology trust is relevant at the first contact with the technology.
When confronted with a lack of trust the introduction of a potential technology can fail. On the
Research outline Automation in crisis management
18 contrary, a high level of trust can make a potential user overtrust the new technology. As a result the user may fail to interrupt in a potentially dangerous situation.
Microsoft Word is often used as an example of trust (or no trust) in automation and the related use or no use of the automation. The relation with the software is often put at the test as unexpected actions or errors occur. Features are eventually put aside because the have been proven unreliable. For instance the automatic spelling corrector. One might put trust in automatically checking and correcting the words, whole sentences are often checked in manual way.
As was explained in chapter one, research in the area of trust is relevant in the Combined System project. Especially, from an Artificial Intelligence view, an intelligent system may show effective functioning in a prototyping world, but in human reality it might be rejected. In automation trust can lead to use or disuse of the automated system. Different levels of automation may have different influence on trust. When functions are taken over in a dynamic way by an automated system there is a possibility that the trust in the system is influenced. As was said appropriate trust is relevant for suitable use. When do you trust someone or thing enough to work with him, her or it? What is needed to trust someone to allocate a function to it? To answer this question in relation with function allocation an exploration is made in the subject of trust when allocating functions between the human and the agent system.
2.3 Research goal
As was described in the other sections, in crisis management the decision to scale-up the organisation has to cope with some problems. The decision to scale-up the organisation needs an awareness of the situation. Without exactly knowing what is going on, no appropriate decision can be made. Two problems were identified in building a picture of the situation. Due to the different parties that make up the organisation, different information comes through. Also the interpretation may be influenced by human biases. It was stated that the interpretation of the information could also be done by a agent system. It was also stated that when the agent system does the interpretation of the situation alone, the human decision-maker does not know what is going on. In this way the decision to scale-up might be based on a inappropriate facts. The interpretation has to be done by both human and agent system.
It is more reasonable to allocate the interpretation of the situation between the human and the agent system. The goal of the present research is to come to a method of allocating parts of the assessment of the situation between human and agent system. The allocation method has to incorporate appropriate trust. Next, the research question is described together with a conceptual model. In the following paragraph the sub questions are explained.
19
2.4 Research Question
To realise the research goal the following question needs to be answered.
In what way can parts of the assessment of the situation be allocated between human and agent in a human-agent support tool to build situation awareness and appropriate trust?
2.5 Research model
To answer the research question the following model gives an insight on the subjects in the research.
In the introducing chapter an explanation of the combined system and the human challenge was given.
As was described in the sections above, function allocation determines in what way functions or tasks are allocated between human and machine. There are tasks that can be allocated in a static way and tasks that are allocated dynamically depending on the context.
Figure 2-1 introduces a conceptual model that is used to obtain the sub questions. Normally the human makes an assessment of the situation him or her self. As described in the research question this research attempts to find a method for the human and system to make an assessment on the situation.
As is shown in the figure both human and system apply for the situation assessment. Which actor is the part of the situation assessment allocated to, depends on the type of function allocation. As was said the types of function allocation range from static to human based. These allocation frameworks are described in the research. The input of the function allocation are the properties of the human and system, as well as the context. It is stated that the three function allocation frameworks have different influence the trust. This is graphically shown as the concept trust between the human and the function allocation. The dotted line is the assumed relation between trust and the function allocation frameworks.
Figure 2-1, Conceptual model Function allocation
Situation Assessment
Human
System
Situation awareness Static Function
Allocation
DynamicFunction Allocation
Human based Dyn. Function Allocation
Trust Human
Properties
System Properties
Research outline Automation in crisis management
20
2.6 Sub questions
To come to an understanding of the research question described in first sections of this chapter, the conceptual part are divided into sub questions. In the paper the following research sub questions are described:
1. What is a function or task and what is its relation with the domain?
2. What are the main issues of cognition for situation awareness?
3. What is function allocation and what concepts for function allocation are available?
4. What is trust, its function and content?
5. How do the function allocation concepts influence trust?
6. In what way can the allocation of functions be implemented for a system in which people and machines collaboratively assess a situation?
2.7 Research method
The sub questions are answered in the chapters described in the conceptual model. In the research process extensive study of literature on function allocation concepts, trust and human decision making (especially in relation to situation awareness) is conducted. This results in the description of function allocation frameworks and concepts. Based on informal interviews or conversations the research subjects were chosen as relevant for both TNO-HF and Combined Systems. Participants of both organisations confirmed the relevance. An additional advantage is that a large project on adaptive automation is recently started at TNO-HF.
A framework for trust is used to extract relevant properties for trust. To explain the relation between trust and the allocation concepts, the concepts are evaluated in relation with the properties of trust.
With trust as an independent variable the impact of the different allocation concepts on the trust relation is explained.
2.8 Document structure
As a prologue to function allocation the next chapter, chapter three, explains what a function is. It also describes the relation with the domain, which is important in function allocation. In relation to building a picture on the situation also a review is taken on the issues of cognition for situation awareness. As the relevant abstract properties of the domain are explained, this chapter describes how these properties are of influence on situation awareness. The result is a description of a function that determines the building of situation awareness.
In chapter four static function allocation is described. It contains some frameworks to allocated functions in the development stage of the system. The function allocations can be based only on
21 human and machine properties or can depend on events that may occur. The last are levels of automation.
In chapter five the allocation of functions is done in a dynamic way, by adaptive automation. In chapter six this allocation framework is extended with a human view on function allocation, better known as team based automation. Directed by some leading literature, adaptivity in human teams is looked at.
Chapter seven describes theory on trust, guided by a framework on trust in electronic environments.
In chapter seven the construct of trust is related to the function allocation concepts
In chapter eight the properties of the function allocation frameworks come together in an implementation, defined as levels of critiquing. This chapter ends with a few remarks on the levels of critiquing in relation to trust. Chapter nine is a concluding chapter, which also contains some recommendations.
