See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/332231225
A Normative Decision Making Model for Cyber Security
Article in Information and Computer Security · April 2019
DOI: 10.1108/ICS-01-2019-0021
CITATIONS
0
READS
425
6 authors, including:
Some of the authors of this publication are also working on these related projects:
Project config.Play View project
XENebula View project
Andrew M’manga
Bournemouth University
7 PUBLICATIONS 5 CITATIONS
SEE PROFILE
Shamal Faily
Bournemouth University
134 PUBLICATIONS 561 CITATIONS
SEE PROFILE
John Mcalaney
Bournemouth University
99 PUBLICATIONS 949 CITATIONS
SEE PROFILE
Chris Williams
Defence Science and Technology Laboratory DSTL
55 PUBLICATIONS 435 CITATIONS
SEE PROFILE
All content following this page was uploaded by Shamal Faily on 05 April 2019.
The user has requested enhancement of the downloaded file.
A Normative Decision Making Model for Cyber Security
Andrew M’manga, Shamal Faily, and John McAlaney
Bournemouth University, Poole, UK
Chris Williams
Defence Science and Technology Laboratory, Porton Down, UK and
Youki Kadobayashi and Daisuke Miyamoto
Nara Institute of Science and Technology, Ikoma, Japan
Abstract
Purpose – The purpose of this paper was to investigate security decision making
during risk and uncertain conditions and to propose a normative model capable of
tracing the decision rationale.
Design/methodology/approach – The proposed risk rationalisation model is
grounded in literature and studies on security analysts’ activities. The model design
was inspired by established awareness models including Situation Awareness and
Observe Orient Decide Act (OODA). Model validated was conducted using
cognitive walkthroughs with security analysts.
Findings – The results indicate that the model may adequately be used to elicit the
rationale or provide traceability for security decision making. The results also
illustrate how the model may be applied to facilitate design for security decision
makers.
Research limitations/implications – The proof of concept is based on a
hypothetical risk scenario. Further studies could investigate the model’s application
in actual scenarios.
Originality/value – The paper proposes a novel approach to tracing the rationale
behind security decision making during risk and uncertain conditions. The research
also illustrates techniques for adapting decision making models to inform system
design.
Keywords Normative, Decision-making, Rationalisation, Awareness, Uncertainty,
Perception, Risk, Security
Paper type Research paper
1. Introduction
Security analysts regularly face the challenge of justifying decisions made under risk
and uncertain conditions. While uncertainty stems from various sources such as
dynamic conditions and information limitations, complications in decision making
also arise because risk stems from multiple factors, rather than a single root cause
(Hoffman et al., 2017). Analysts aim at identifying the best possible option, given the
limited information as few decisions are actually made with absolute certainty
(Huber, 2014).
Information limitations exemplify the difference between optimising in rational
decision making where all options for a decision are known and satisficing in
bounded rationality driven decision making where limited options are known
(Simon, 1972). When decision making under risk and uncertainty is unsuccessful, the
post-incident privilege of hindsight availed to others fails to portray the complexity
of decision making in action. Equally, value is lost when decision making knowledge
gained from those experienced remains tacit and incommunicable.
To facilitate the transparent understanding of security decision making, we present a
conceptual model providing systematic traceability to risk rationalisation. We
validate the model using cognitive walkthroughs and illustrate its application in a
study where analysts’ views on automating aspects of risk rationalisation during
security analysis were elicited. This paper is an extended version of earlier work
presented in M’manga et al., (2018).
2. Related work
Decision making research typically follows the normative or descriptive approach.
Normative approaches model how decisions should be made; descriptive approaches
understand how decisions are actually made. The normative approach’s usefulness
may be seen in its ability in providing theoretical adequacy for rational choice,
whereas the descriptive approach’s usefulness may be seen through empirical
validity by uncovering insight in decision making (Bell et al., 1988). An alternative
approach is to categorise decision making research based on the study environment.
This may be the lab-based approach where studies are conducted in controlled
environments and data collection is determined by predefined tests (MacKenzie
2013), or the naturalistic approach where studies are conducted in real settings and
data collection is based on the observation of actual events (Klein, 2008). The
differences in approaches do not imply that one is better than the other, but that each
is suitable based on research objectives.
Descriptive research on expert decision making during risk and uncertainty focusses
on context-specific decision making. This has been led by Klein’s (1999) research on
naturalistic decision making, where he identified that during uncertainty, experienced
firefighters use situational familiarity to make quick decisions as opposed to
weighing all available alternatives. Similar work by Wong and his colleagues has
examined how criminal intelligence analysts think (Wong, 2014: Wong and
Kodagoda, 2015). They suggest that analysts flow from fluidity to rigour during
decision making which is explained as moving from a loose story to account for
identified data, to a formal rigorous and defensible argument. Hibshi et al., (2016)
explores techniques taken by security experts as they transition through levels of
situation awareness to identify security requirements. They identify that experts seem
to skip some stages of situation awareness and attribute this to situation familiarity
based on experience.
The movement from ambiguity to certainty is the common point in the literature,
however, much is still required to understand the rationale behind decision making.
Normative models are particularity useful for this as they act as blueprints upon
which sensemaking may be traced and communicated. Early work by Rasmussen
(1974) on the Decision ladder template has played a key role in identifying the
generic categories of activity in decision making, similarly, Boyd (1996) and
Endsley (1995) played key roles in formalising the awareness steps leading to
awareness. Regrettably, normative approaches are usually too high-level and
generalised, rendering them incapable of providing low-level context-specific
guidance.
3. Model Design
Figure 1: Risk Rationalisation Flow
The model builds on our findings in M’manga et al., (2017), where we identified
factors influencing risk interpretation and conflicts to security decisions, and lessons
learned from the literature on awareness and analysts’ decision making activities
(Werlinger et al. 2010; D’Amico et al. 2005). The model was formalised
systematically using OODA (Boyd, 1996).
The normative model consists of eight steps to risk rationalisation and contains two
complementary elements; the Flow and Actions collectively referred to as the risk
rationalisation process (RRP). The first element is a risk rationalisation flow (RRF)
highlighting cognitive sequences and iterations during risk rationalisation. Illustrated
in Figure 1, RRF indicates two alternative starting points; Reactive risk analysis
beginning with Situation assessment and continues to Goal formation, while
Proactive risk analysis has an inverse flow of beginning with Goal formation and
continuing to Situation assessment. The difference is based on the understanding that
incidents precede response strategy in reactive analysis. Therefore, situation
assessment begins before goals are formed, while the inverse is true in proactive
analysis where goals are set beforehand. The second phase of RRF consists of the
three information related steps, these are; Information needs assessment, Information
exploration, and Information limitations analysis. Their adjacent positioning in
Figure 1 indicates that the steps may overlap and occur in varying order. The final
three steps occurring in sequence relate to options. These are; Options generation
and analysis, Option validation, and Option selection. Risk rationalisation is an
iterative process; this is illustrated in RRF by the reverse arrows at each point of
possible iteration.
The second element of RRP consists of the risk rationalisation actions illustrated in
Figure 2. The actions address the lack of low-level detail in normative models by
providing context-related meta-cognitive questions at each rationalisation step.
Metacognition is defined as awareness or analysis of one’s own thinking processes,
this may be extended as the knowledge of knowledge (what one knows about their
thinking), and the regulation of knowledge (how one uses that knowledge to regulate
thinking) (Schraw and Moshman, 1995). For instance, to understand the rationale
behind the characterisation of a situation, the question “how may a situation be
understood?” is posed. The questions are posed retrospectively, hence metacognition. The risk rationalisation actions also present procedures for clarifying the
questions. In the case of “how may a situation be understood?” the procedure could
be through data correlation, which is the putting together of disparate data sets to
derive meaning. By using the rationalisation steps, meta-cognitive questions, and
procedures, RRP aims at understanding the rationale behind decision making
irrespective of the decision maker’s expertise. We detail the eight RRP steps below.
3.1. Situation assessment
Situation assessment corresponds to OODA’s Observe. During this step, the aim is to
understand how the decision maker identifies factors aiding in situation
understanding and not the actual analysis of the situation. The meta-cognitive
question “how may the situational be understood?” is presented and expanded into
four possible procedures;
• Knowledge of a situation: recognition through situation familiarity and the
knowledge of normal.
• Knowledge of evidence: recognising information affordances in an
environment to achieve greater awareness.
• Situational time-line: recognising whether a situation is static or evolving,
current or elapsed.
• Data correlation: recognising data correlation needs to achieve greater
awareness.
Figure 2: Risk Rationalisation Actions
3.2. Goal formation
Goal formation is a step also corresponding to OODA’s Observe. The objective is to
understand the strategies used to establish decision goals, identify tensions that may
restrict goals from coming to fruition, and the determination of the relevance scope
within which a decision is made. The relevance scope acts as a minimum level for
the continued pursuit of a goal. For example, analysts interviewed in M’manga et al.,
(2017) expressed that the inner workings of some of the proprietary security products
they used were unknown to them. However, based on the product’s benefit, they
found uncovering the potential risk unnecessary. In other words, the relevance scope
for pursuing the products risk was below minimum.
3.3. Information needs assessment
Information needs assessment is one of three steps corresponding to OODA’s Orient.
The objective is to understand how the decision maker identifies information
relevant for decision making and the filtering of excess information. The decision
maker’s assessment is based on information credibility determined by factors
identified during Situation assessment and the relevance scope identified during Goal
formation. An example would be procedures taken in identifying false positives
during an incident.
3.4. Information exploration
During Information exploration, it is recognised that decisions are determined by
information availability and when information is unavailable, possible alternatives
are explored. The focus is therefore placed on understanding the strategies for
identifying the alternative sources of information. To the decision maker, the
exploration of additional information sources is subject to available time.
Information sources may be subject matter experts within an analysts’ environment
e.g. legal officer, public relations manager, or external expertise such as CERT
advisory (Computer Emergency Response Teams).
3.5. Information limitations analysis
Information limitations analysis is driven by the question, what remains unknown?
This is presented with the aim of understanding how the decision maker identifies
critical information gaps and the conclusion drawn from the knowledge. Information
gaps refer to the known-unknowns critical for informed decision making. For
example, it would greatly aid an analyst to acknowledge that an attack vector has
been identified although the motive and capabilities remain unknown. Knowledge of
the motive could hint at the possibility of follow-up attacks leading to better
preparedness (Rashid et al., 2016).
3.6. Options generation and analysis
Options generation and analysis is the first of two steps corresponding to OODA’s
Decide. Based on the cumulative understanding from the previous steps, the decision
maker identifies possible options for decision formulation and their implications. For
example, an analyst’s response to a data breach could be to refrain from disclosing
the breach, even though data protection regulations advise otherwise. The aim of the
step is to identify and understand the reasoning behind options considered by the
decision maker. At this point, limited understanding may inadvertently lead to the
first form of meta-risk, which is risk resulting from risk response e.g. increased threat
exposure.
3.7. Options validation
Options validation focusses on uncertainty by verifying if there were elements of
uncertainty hindering the decision making process and how it was managed.
Evaluating one’s own decision is no easy task; the failure to validate an option
introduces a second form of meta-risk, which is the risk of risk understanding.
To provide a comprehensive understanding of uncertainty, we categorise it to the
following four groups:
• Environmental factors: dynamic environments, inconsistent or incomplete
information from the environment.
• Contextual factors: time limitations, situation complexity or magnitude.
• Personal factors: experience, training and cognitive limitations.
• Information factors: accurate, current, relevant specific, understandable,
comprehensive, unbiased and comparable (Wang et al., 2005).
3.8. Option selection
Option Selection corresponds to OODA’s Act. As a final step, the most informed and
objective option is put forward as the basis for a decision. The option should not
come as a surprise where the rationale is traceable.
4. Model flow validation
The model was validated using cognitive walkthroughs (Rieman et al., 1995) with
three security analysts from the UK (P1 -3). The three validated the model’s logic
flow and not the tracing of risk rationalisation. P1 and P2 worked as part of a cyber
security team monitoring events within their organisation and possessed 1-3 years’
professional experience in security. P3 worked for a counter-terrorism and
intelligence unit and possessed over 24 years of relevant experience.
Each participant was provided with a copy of RRP and given a brief tutorial on its
use. Participants were then presented with a scenario about a hypothetical data
breach that incorporated tensions related to possible decisions and uncertainty due to
insufficient information. The scenario required the analysts to decide whether to
make a breach on a university’s network known to affected parties in advance, after
remediation or not at all. In addition, they had to take into account that some of the
breached data was already on the dark web.
The participants were asked to compare the model with decisions they would make
in the scenario. Additionally, P3 ran a second validation scenario, based on his
experience in counter-terrorism. Each walkthrough took approximately 40 minutes,
and the participants presented their critiques of the model’s logic. Opinions were
divided on whether Option validation was an independent step or a part of Option
generation and analysis. We concluded that it remains an independent step to cater
for understanding inexperienced decision makers lacking the ability to generate and
validate decision alternatives consecutively.
5. Model Application
To demonstrate the application of RRP, a study was conducted with information
systems (IS) personnel from a variety of organisations in Japan with the aim of
eliciting their views on automating aspects of risk rationalisation during security
analysis. Under real settings, the understanding gleaned would form the basis for
nuanced security automation requirements that are grounded in user understanding as
opposed to default automation for all.
The findings do not aim to present statistical accuracy as we did not work with a
representative sample size. Rather, the aim is to illustrate how RRP may be applied
in facilitating understanding and designing for security decision makers (proof of
concept). Demographic factors such as experience, security roles and industry could
be used in representative and longitudinal studies to uncover decision rationalisation
norms and deviations thereof. Findings could hint at hidden biases and suggest
training and design requirements.
5.1. Participants
As illustrated in Figure 3, the nine participants were part of a group undertaking a
training programme in cyber security. Two of the nine worked in security-specific
roles; while the rest had security as a part of their other IS responsibilities. The
participants had work experience ranging from one to ten years, four had a maximum
of two years (referred to as novices) and the remaining five had a minimum of six
years (referred to as experienced).
5.2. Data collection
Like the validation exercise (Section 4), participants were trained on RRP and
provided with a cybersecurity decision making scenario containing elements of risk
and uncertainty. Participants were then asked to indicate whether or not they would
use automation during the different step of risk rationalisation during security
analysis. We sought to elicit the participants’ general opinion, not one driven by
existing tool capabilities. The steps considered were from situation assessment to
options validation. Option selection was intentionally omitted as the objective was to
automate with the human as the final decision maker (human-in-the-loop).
5.2.1. Scenario
You are a security analyst at a shipping company based in Tokyo. You have been
monitoring the network traffic at the Osaka regional office and have noticed
suspicious Twitter and Internet Relay Chat (IRC) traffic. There is the possibility that
this could be an incident in progress. Following the RRP model steps, which parts of
your security analysis would you automate to facilitate understanding?
Figure 3: Study participants
5.3. Findings
5.3.1. All respondents
Illustrate in Figure 4, the first part of the findings represents all respondents (novices
and experienced) as one group. Situation assessment was the most recommended for
automation with seven of the nine participants selecting to automate. Participants’
comments for the choice included:
“Systems have a better understanding of network traffic and monitoring for
signs of risk.”
“It would be possible to issue alerts at the point of anomaly detection.”
Information needs assessment had the lowest outcome with none of the participants
recommending it, seconded by Option validation with one recommendation.
Participants’ comments for not selecting Information needs assessment included:
“When determining the necessary or unnecessary information, one’s
experience or wider perspective would be required.”
“It is ambiguous for a system to determine unnecessary information.”
Figure 4: Automating security analysis (all respondents)
5.3.2. Novice versus Novices
Illustrate in Figure 5, the second part of analysing the findings aimed at identifying if
there were visible differences in the novice versus experienced view.
As an overall, the results indicate that the experienced where lesser inclined to
automate. We could speculate that this reflects a lack of self-belief and tool
dependency by the novices. Significant differences were observed under Information
limitations analysis and Options validation where none of the experienced chose to
automate, in contrast to the larger number of novices who chose to automate.
Reasons for choosing not to automate the two steps included:
“New information might emerge continuously; thus, automation would be
impossible.”
“It would be possible to establish frameworks, but it cannot be automated.”
Figure 5: Automating security analysis (novice vs experienced)
As explained above, we do not aim to draw too much from the findings due to the
limited sample size, however, the clear difference in the novice versus the
experienced choices indicates the soundness in the application of RRP as a nuanced
view to risk rationalisation.
6. Conclusion
This work presented a normative model for rationalising security analysts’ decision
making. The purpose of the model is not to propose a new approach to decision
making, but rather to propose a systematic approach capable of communicating and
providing traceability to the rationale behind security decision making during risk
and uncertainty. To address this, we considered the shortfalls presented in descriptive
approaches which usually provide no explanation for expert judgement and the
shortfalls in normative approaches which are usually too high level to derive
contextual meaning.
We believe there are three benefits from the use of this model. First, the model is
designed as a series of steps, cognitive questions and procedures that may be used as
a blueprint by stakeholders unfamiliar with risk analysis procedures in security e.g.
proactive or reactive analysis. Second, the model may be used in training analysts be
identifying gaps in their reasoning when compared to model steps. Finally, the
model may be used as a basis for eliciting design requirements that would facilitate
decision making about risk through the identification of crucial areas of risk
rationalisation in security.
The model places emphasis on validation and the consideration of uncertainty by
highlighting the iterative nature of decision flows, presenting an options validation
step, and the consideration of meta-risk in two forms (risk of understanding and risk
of response). We believe that the model complements existing decision making and
awareness approaches lacking a focus on risk and uncertainty.
The case study presented acts as a proof of concept on the application of the model.
It illustrated how the rationalisation steps in security analysis could be considered for
automation based on user requirements and understanding. We believe this is a
sound approach for leveraging human and system capabilities for decision making
about security risk.
7. Acknowledgement
The research was funded by Bournemouth University studentship
DSTLX1000104780R_BOURNEMOUTH_PhD_RBDM, with the initial
collaborative meeting between UK/Japan researchers facilitated by support from the
Great Britain Sasakawa Foundation. We are also grateful to DSTL for their
sponsorship of this work.
8. References
Bell, D. E., Raiffa, H., and Tversky, A. (1988), Decision making: Descriptive, normative, and
prescriptive interactions, Cambridge University Press.
Boyd, J. R. (1996), “The essence of winning and losing”. Unpublished lecture notes, Vol. 12,
No. 23, pp. 123–125.
D’Amico, A., Whitley, K., Tesone, D., O’Brien, B., and Roth, E. (2005). “Achieving cyber
defense situational awareness: A cognitive task analysis of information assurance analysts”.
In: Proceedings of the human factors and ergonomics society annual meeting. SAGE
Publications Sage CA: Los Angeles, CA, pp. 229–233.
Endsley, M. R. (1995), “Toward a theory of situation awareness in dynamic systems”. Human
Factors: The Journal of the Human Factors and Ergonomics Society, Vol. 37, No. 1, pp. 32–
64.
Hibshi, H., Breaux, T. D., Riaz, M., and Williams, L. (2016). “A grounded analysis of experts’
decision-making during security assessments”. Journal of Cybersecurity, Vol. 2 No. 2, pp.
147–163.
Hoffman, R. R., Mueller, S. T., and Klein, G. (2017), “Explaining Explanation, Part 2:
Empirical Foundations”. IEEE Intelligent Systems, Vol. 32, No. 4, pp.78–86.
Huber, O. (2014), “Complex problem solving as multistage decision making”. In: P. A.
Frensch, & J. Funke (Eds.), Complex problem solving: The European perspective, Lawrence
Erlbaum Associates, Hillsdale NJ, pp. 151–173.
Klein, G. (1999), Sources of power: How people make decisions. MIT press.
Klein, G. (2008), “Naturalistic decision making”, Human Factors: The Journal of the Human
Factors and Ergonomics Society, Vol. 50 No. 3, pp. 456–460.
MacKenzie, I. S. (2013), Human-computer interaction: an empirical research perspective.
Morgan Kaufmann, Amsterdam.
M’manga, A., Faily, S., McAlaney, J., and Williams, C. (2017), “Folk Risk Analysis: Factors
Influencing Security Analysts’ Interpretation of Risk”, In Proceedings of the 13th Symposium
on Usable Privacy and Security (SOUPS), Santa Clara, USA,12-14 July 2017, Usenix
Association.
M’manga, A., Faily, S., McAlaney, J., and Williams, C. (2018), Rationalising Decision
Making about Risk: A Normative Approach, In Proceedings of the twelfth International
Symposium on Human Aspects of Information Security & Assurance (HAISA), Dundee UK,
29-31 August 2018, University of Plymouth, pp. 263-271.
Rashid, A., Naqvi, S. A. A., Ramdhany, R., Edwards, M., Chitchyan, R., and Babar, M. A.
(2016), “Discovering ‘unknown known’ security requirements”, In 38th International
Conference on Software Engineering, ACM Press, pp. 866–876.
Rasmussen, J. (1974), The human data processor as a system component. Bits and pieces of a
model. Roskilde, Denmark: Danish Atomic Energy Commission. No. Risø-M-1722.
Rieman, J., Franzke, M., and Redmiles, D. (1995), “Usability evaluation with the cognitive
walkthrough”. In Conference companion on Human factors in computing systems, ACM, pp.
387–388.
Schraw, G. and Moshman, D. (1995), “Metacognitive theories”. Educational Psychology
Review, Vol. 7 No. 4, pp. 351–371.
Simon, H. A. (1972), “Theories of bounded rationality”, Decision and organization, Vol.1 No.
1, pp. 161–176.
Wang, Y. R., Pierce, E. M., Madnik, S. E., Fisher, C. W., and Zwass, V. (2005), Information
quality. Armonk, N.Y. ; London, England, M.E. Sharpe.
Werlinger, R., Muldner, K., Hawkey, K., and Beznosov, K. (2010). “Preparation, detection,
and analysis: the diagnostic work of IT security incident response”, Information Management
& Computer Security, Vol. 18 No. 1, pp 26–42.
Wong, B. L. W. (2014), “How Analysts Think (?): Early Observations”, In Joint Intelligence
and Security Informatics Conference, IEEE, pp. 296–299.
Wong, B. L. W. and Kodagoda, N. (2015), “How Analysts Think: Inference Making
Strategies”, Proceedings of the Human Factors and Ergonomics Society Annual Meeting, Vol.
59 No. 1, pp. 269–273.
View publication stats
The post Decision Making Model for Cyber Security appeared first on My Assignment Online.