COMPUTING IN SCIENCE & ENGINEERING, VOL. X, NO. X, MONTH–MONTH 2026
1
A Field Guide to Decision Making
arXiv:2604.20669v1 [cs.CY] 22 Apr 2026
Richard B. Arthur
Abstract—High-consequence decision making demands peak performance from individuals in positions of responsibility. Such executive authority bears the obligation to act despite uncertainty, limited resources, time constraints, and accountability risks. Tools and strategies to motivate confidence and foster risk tolerance must confront informational noise and can provide qualified accountability. Machine intelligence augments human cognition and perception to improve situational awareness, decision framing, flexibility, and coherence through agentic stewardship of contextual metadata. We examine systemic and behavioral factors crucial to address in scenarios encumbered by complexity, uncertainty, and urgency.
A. Present Orientation
Decision-making authority is the defining responsibility for executive roles. That responsibility requires calculated commitment to action within an expected timeframe, budget, and availability of resources to develop sufficient confidence despite uncertainty and risk. Success results from balancing these components wisely, with a pragmatic perspective on current and future implications. Behaviors can be highly influenced by performance metrics, such as efficiency, throughput, yield, cycle time, and margin. Agile methodologies motivate action through rapid launchIndex Terms—Decision making, knowledge management, artificial intelligence, VUCA, decision provenance, agentic systems, and-learn of minimally viable solutions, which subsequently accountability. improve through iteration. However, urgency to act and incentives to select expedient, cheap, or low-effort options ECISIONS drive our actions in the present and shape can result in short-term, makeshift outcomes. Engineers refer the unfolding future. We make those decisions within the to these quasi-solutions as tech debt, a liability on a notional limits of our perception of the present moment and knowledge ledger of deferred resolutions, which have the potential for compounding risks, costs, and complexity in the intervening learned from past experiences. time. In daily life, decision making can seem straightforward beSome fields sensitive to decision-making consistency have cause we typically have sufficient time to consider alternatives, developed systems for disciplined compliance with procedural flexibility to change with minimal consequence, and the option norms. Professionals that routinely perform high-stakes proceto defer to commit or decline altogether. Professionals making dures, such as aviators, surgeons, and contract lawyers, employ consequential decisions bear the burden of accountability to act formalized techniques for due diligence and confirmatory wisely within the confines of windows of opportunity, available checklists. Regulatory authorities may also assert information resources, and a tolerance for uncertainty. Decisions may incorbe recorded for availability to audit. porate tacit experience, psychological and emotional factors, Information technology (IT) capabilities can address operaand other contextual aspects that are difficult to distill from tional encumbrances of urgency, complexity, and uncertainty. the human mind into machine data. Cassie Kozyrkov studies The concept of the digital thread [2] represents a notional these interactions in her materials on decision intelligence [1]. formalization of a connected genealogy of related records and To simplify the distinction of decisions that are consequential, data made accessible and interoperable for cross-functional consider whether the decision is documented into record, collaboration: for example, linking engineering information and rather than requiring thresholds of impact, irreversibility, or data from the design, manufacture, operation, and servicing of dedication of resources. The recording serves to clarify action a product. in the present and archive information for future reference. Knowledge management tools and practices facilitate capTwo example contexts will illustrate use of decision records: ture, collaboration, and learning through annotation meta1) medical patient records (diagnostic tests, therapies) and data, taxonomies for semantic alignment, and ontologies for 2) engineering records (design changes, tradeoff assessments). understanding. Historically, knowledge management suffers from burdensome complexity and intrusive capture of tacit knowledge within otherwise intuitive workflows. Advances in I. D ECISIONS IN C APTIVITY artificial intelligence (AI), such as speech-to-text and large Temporal orientation serves to frame decision scenarios: language models (LLMs), offer potential solutions for lowinvestigating the past, awareness in the present, and planning friction collection and curation of contextual metadata through for the future. This section examines the current state of side channels. practice and highlights opportunities for improved information infrastructure. B. Past Orientation
D
Richard B. Arthur is with GE Aerospace Research, Niskayuna, NY 12309, USA (e-mail: [email protected]). Digital Object Identifier 10.1109/MCSE.2026.3676981 1521-9615 © 2026 IEEE. All rights reserved, including rights for text and data mining, and training of artificial intelligence and similar technologies.
Root cause analysis (RCA) is a well-established approach for responding to emergent, unexpected issues, such as a patient suffering from a rash or a pattern of part failures in a product. Questions posed consider evidence in the present relative to
COMPUTING IN SCIENCE & ENGINEERING, VOL. X, NO. X, MONTH–MONTH 2026
past conditions, events, decisions, and actions—such as: What was the cause? How and why did this happen? and What was the cause of that cause?—iteratively tracking contributing factors to identify root causes, perhaps formalized through Bayesian inference models. The primary intent of RCA is to understand the scope of the problem, discover and validate causes, and then devise and evaluate options for a corrective solution. However, additional questions can offer crucial learning opportunities, such as: What decisions led to this? How and why were those decisions made? (In particular, contextual framing of the decision at the time: alternatives, selection criteria, assumptions, unknowns). Corrective urgency leading to a fix sufficient to move on, but incomplete as a remedy, can introduce tech debt. RCA encumbered with motives to assign blame may result in learned behaviors detrimental to promoting a culture of decisionmaking confidence and accountability.
2
urgency can further constrain time and availability of resources to achieve desired clarity, resulting in satisficing behaviors described by Herbert Simon’s Bounded Rationality [4]. Additionally, organizational boundaries and intervals of elapsed time introduce opportunities for inconsistency between related decisions. The politics of blame and socialized negativity bias expose decisions and actions to selective retrospective judgment in hindsight. Unmitigated, these risks undermine confidence, motivate a culture of mediocrity through blame-avoidant behaviors, and result in wasted time, resources, funds, diminishing operational robustness, agility, and eroding trust required to coordinate. Systemic, behavioral, and exogenous factors can also impede responsible action, particularly under duress and urgency when events place lives, property, and order in jeopardy. A. VUCA Fog of War
C. Future Orientation
The U.S. Army War College characterized the post-Cold War operational landscape with the term VUCA: for volatility, uncertainty, complexity, and ambiguity, elements of the “fog of war” in military doctrine stretching back to Sun Tzu. The VUCA terms may be defined [5]:
Anyone in a position to make consequential decisions will themselves bear repercussions for decisions made (or inaction) by way of being entrusted with that authority. Awareness of this accountability can influence behaviors in reflecting upon • volatility: prevalent forces/catalysts of rapid change the possibility of judgment in hindsight, such as in the findings • uncertainty: factors limiting prediction/confidence of a future RCA. • complexity: many-factor interdependencies Rational strategies to mitigate that potential risk include • ambiguity: unclear, nuanced, and mixed interpretations. consultation and conservatism. While measured implementation of those practices is reasonable and customary, particularly While urgent decision making can employ systematic prein regulated and high-consequence contexts, excessive de- paredness to mitigate complexity and uncertainty, volatility and liberation, such as exhaustive analyses or consultations, can ambiguity pose more elusive challenges. waste time, resources, and miss windows of opportunity to Complexity may be addressed through preparedness, learning be effective. Likewise, compensating for uncertainty while from past experiences, codifying choices, and processes where under duress of urgency or inadequate resourcing promotes best practices as standard operating procedures (SOP) can conservatism. bound and simplify options and context. These proactive For example, consider product design “overengineering,” measures help reduce errors and improve timely, sound, and resulting in extraneous costs, delays, and reduction to perfor- consistent decisions, even under the duress of urgency. SOPs mance specs or operational guidance. In health care, medical are commonplace in both medical practices and product practitioners employ guarded caution with which patient manufacturing to address routine tasks and reinforce complimedical records must be recorded in learned response to ance. Increasingly powerful analytical tools aim to tame everthe proliferation of malpractice litigation and exposure to growing complexity, most notably through the recent advances subpoena. This pragmatic hedge characterizes one of the tactics deriving models from vast data through machine learning (ML), employed in the practice of “defensive medicine” to safeguard including LLM semantic models. medical practitioners, often to the detriment of patients. Adverse Uncertainty and error persist in limiting data-derived model consequences might include diminished quality of care, burden performance. While gathering additional data can reduce of increased tests (incremental logistics, costs, and time delays), epistemic uncertainty, aleatoric uncertainty arising from noisy, and reduced candor in communication between physician and dimensionally sparse, and difficult (or expensive) to measure patient. data persist in confounding predictive capability. Furthermore, Awareness that high-consequence decisions may later be the principle of “garbage in, garbage out” can instead become judged retrospectively can encourage behavioral flaws toward “garbage amplified” with naïve acceptance of, or reliance upon, irresponsible caution [3]. insufficiently informed (poorly trained) models. Volatility can rapidly erode even well-established situational II. D ECISIONS IN THE W ILD clarity, as critical factors become obsolete and consequently The executive decision makers entrusted with responsibility require persistent adaptation to changes, guided by appropriate for high-consequence decisions confront systemic and behav- guardrails [6]. ioral factors that undermine confidence. The gathering and Although SOPs, standard work, best practices, and dataassessment of supporting information must contend with volatil- derived models (ML) can provide preparedness strategies ahead ity, uncertainty, complexity, and ambiguity (VUCA). Decision of or in swift response to scenarios driven by urgency, model
COMPUTING IN SCIENCE & ENGINEERING, VOL. X, NO. X, MONTH–MONTH 2026
applicability will remain confined to the time period and scope of their framing reference data. Some information and data remain unknowable until their emergence in future events, confounding even extensive precautions. Only persistent vigilance over the passing of time may afford desired confidence, including resolution of the shifting interpretations that confer ambiguity. Fluid interpretations of ambiguity form one source of perceived inconsistency; additional factors can also challenge coherence of thought across decisions. B. Inconsistency Decisions are rarely isolated or independent. Interrelated decisions may occur sequenced over time and across stakeholders with significantly different experience, perspectives, and incentives. Affected parties and outside observers correspondingly form interpretations through their own distinct lenses. Perceived inconsistency among these participants undermines confidence and trust. IT tools and practices can compensate for the unreliability of the human mind to reliably perform precise and accurate recall from memory or to apply persistent focused attention over extended spans of time. Traditional approaches, such as shared database archives and procedural checklists, facilitate consistency over time and between collaborators. For example, at an annual physical checkup, a physician will consult medical records to refresh their memory of that patient’s health context. The discussion should retain coherence of thought and strategy for care despite the physician having seen numerous other patients during the intervening year. The records capture and summarize contextually relevant information and changes outside the examination, including family medical history, visits to other specialists, and perhaps insurance or regulatory policies for approval of diagnostics or therapies. As medical institutions impose patient throughput targets on practitioners, timely and accurate records become crucial for physicians to maintain the urgent pace of visits while minimizing poor outcomes due to error or omission. Therefore, patient record systems must be sufficiently robust and intuitive to use to maintain consistency event to event and among practice specializations. Discontinuities due to organizational boundaries, roles with differing incentives, mismatched access to resources, or from agents with disparate sociopolitical norms can introduce inconsistencies in values and assumptions influencing consequential decision making. The impact of such factors can range from naïve misalignment to fomenting competition that motivates behaviors counter to desired institutional results. For example, engineering design inherently navigates between tradeoffs allocating costs and resources versus targets, such as performance, efficiency, reliability, and time to market. Consider installation of a freeway ramp with bounds on project cost and urgency to complete. Savings on upfront capital costs, time, and resources for the project may lead to higher total costs and effort for ongoing maintenance. Prioritization criteria dependencies are sensitive to budget cycles, access to resources, and situational variables across municipal, contractor,
3
and regulatory decision makers. Misaligned or contradictory assumptions shift the organizational burden needed to address consequent tech debt. Inconsistency undermines trust and clarity, which can result in avoidance of accountability when responsibility becomes instead perceived as liability. C. Self-Interest Decisions of consequence routinely prompt unsolicited observations, commentary, and actions by others. This can be praise or admiration, lending credit, but results short of flawless success can provoke motivated scrutiny, critique, and judgment, subjecting the decision maker to blame. Oxford University findings published by Christopher Hood find blame carries four times the cultural weight as comparable credit [7]. Social amplification of blame, termed negativity bias, reinforces this effect, with successes taken for granted while failures are reliably and asymmetrically spotlighted. Anticipation of exposure to liability and other retrospective judgment in hindsight motivates devoting greater effort to avoiding blame than to earning credit. Hood terms diminished potential resulting from blame-avoidant behaviors mediocrity bias. Examples of blame-avoidance strategies Hood cites are: • Deflect: Scapegoat, delegate, encode as policy. • Distract: “Spin,” bury in news/procedural cycle. • Decline: Shun risk of toxic exposure despite merit. • Diffuse: Mutual deniability (credit optional). Self-interest behaviors and survival instincts conflate challenges already prevalent in the dynamic, inscrutable, interconnected, and nuanced VUCA environment. Passive forms of accountability evasion include reluctance to document decisions and actions: For example, to curtail defensively the candor, transparency, and completeness of information supplied to systems of record (e.g., cautious detail in patient medical records, prioritizing mitigation of risk from a malpractice subpoena over patient care utility). Less passive tactics may cross lines established on principles of integrity, accountability, and even legality. The present political landscape has highlighted intentional disinformation as one blatant method to evade and reassign accountability. Its merited discussion lies outside of the scope of this article, however, beyond acknowledging the practice as a factor to consider. Game theory offers a framework for characterizing and analyzing interaction behaviors among interdependent decision makers. Urgent and high-consequence scenarios amplify the stakes in calculations for compromise, prioritization, and bounds for inclusion. Measures should be taken to promote trust and confidence, and to recognize and compensate for gullibility, insincerity, and ineptitude. Example archetypes characterizing extremes of behaviors and motivations are: • Wise: Acknowledges uncertainty and their own ignorance, proactively and thoroughly records accurate data with transparent intent to qualify, quantify, and learn.
COMPUTING IN SCIENCE & ENGINEERING, VOL. X, NO. X, MONTH–MONTH 2026
4
Benignly naïve: Lacks awareness, experience, and/or to revisit a decision in the future, formally recording decisionresources, and is open to learn if afforded the opportunity. time caveats, concerns, assumptions, limitations, and unknowns • Stubbornly naïve: Lacks awareness and interest to inquire as searchable decision metadata. or improve, firmly clutch unfounded certainty, and is These metadata constitute decision provenance, providing subdued by cognitive biases. transparency into the framing context at the time the decision • Defensive: Reflexively avoids accountability by employing was made. The term adopts the notion of provenance as blame avoidance strategies by default, seeking survival as metadata as used by scientists to improve understanding and best-case outcome after the storm passes. reproducibility of experimental data and analyses [8]. Decision • Self-interested: Prioritizes personal objectives over orga- provenance metadata provide similar pedigree for the origin nizational goals, willfully projects unmerited certainty, and lineage of decisions of significance. exploits logical fallacies, and counters inquiry aggresDecision provenance (metadata) examples are: sively. • criteria to select and evaluate Decisions “in the wild” cannot presume the luxury of peers • alternatives considered exclusively from the wise. On the other hand, draconian • assumptions distrust and verification anticipating purely selfish coactors • constraints (budget, time, expertise, . . . ) may waste time, resources, and foment distrust, prompting • (known) unknowns defensive postures among coactors. • supporting references . . . Confidence lies at the foundation of high-consequence Documenting present constraints and tacit factors seen as decision making. Insufficient confidence and coordination relevant to making the decision can mitigate concern that a confronting VUCA and coherent connectedness over time future retrospective will miss present nuance or limitations, and across stakeholders can prompt flawed decision-making which may no longer hold in that future. An example is behaviors, from unintentional inconsistency to avoidance of awareness of additional diagnostic test options for a patient, but accountability. Technological advances in knowledge systems election to hold off based on logistical and financial constraints. offer tactics to improve collaborations across organizations and Numerous prior efforts to capture implicit, tacit knowledge over time, improve consistency of thought, and operationalize have failed due to challenges, ranging from management of adaptation. semantic consistency and complexity to onerous, nuanced, and resistant human interaction. Advances in AI technologies have III. D ECISIONS E VOLVED shifted the state of the possible for both the computational Enlisting the rapidly advancing capabilities of digital in- obstacles (through data-derived models, such as ML and LLMs) frastructure and machine intelligence, we can envision a and user acceptance of collaboration with intelligent systems collaborative knowledge system to bolster confidence and (normalized through home automation and chatbots). While legacy processes may already capture some of this accountability with improved contextual clarity of the past, information informally in annotations of archived reference awareness in the present, and anticipated preparedness for documents (e.g., slides, spreadsheets, or e-mail), the explicit future needs. Implementing such a framework requires cultural mapping of contextual metadata into a searchable knowledge commitment from leadership and an integrated knowledge representation forms a foundational function of the envisioned infrastructure to navigate decision making within the VUCA knowledge infrastructure. environment. We set the following goals to improve decision making: • confidence to act despite uncertainty B. To Act Despite Uncertainty • decision coherence and consistency This archival of decisions, their supporting data, and prove• preparedness through adaptation by design. nance metadata into an enterprise system for knowledge stewardship establishes a recognized authority that can be queried to A. Decision Provenance clarify sensitivities and limitations current to decision making. The effectiveness of a technical knowledge infrastructure Cultural leadership to recognize and employ an authoritative to achieve these goals despite VUCA factors will require reference offers decision makers qualified accountability as a leadership and cultural commitment to safeguard learning as safeguard against capricious blame or liability. Assurance of qualified accountability can lend confidence to a strategic operational foundation. Learning shifts perspective from decisions as potential liabilities to viewing them as key commit to act despite uncertainty and reduce deliberation and conservatism, even when constrained by urgency: for example, control points for adaptation and improvement. First, recognize decisions as the most consequential result of present shortage of the otherwise most desirable raw material the myriad activities gathering, processing, qualifying, quanti- for manufacturing a part. fying, verifying, validating, analyzing, synthesizing, modeling, In the design of a system for capturing decision context (with simulating, evaluating, and framing data and information. candor and clarity), the provenance metadata merit information Second, notionally transform decisions from a result to be sensitivity assessment akin to the underlying data and decisions archived to become a primary index for underlying data, models, themselves. Exposing the thought processes and state of mind and analyses supporting said decision. Finally, anticipate cause framing consequential decisions, the assumptions, unknowns, •
COMPUTING IN SCIENCE & ENGINEERING, VOL. X, NO. X, MONTH–MONTH 2026
evaluation criteria, and alternatives taken together merit careful assessment for confidentiality and protection. Additionally, the adaptive and dynamic nature of a system for knowledge stewardship creates opportunities to undermine integrity if unethical actors can rewrite history, motivated by stealing credit or avoiding blame. Without reliable and robust version controls and immutable audit records, the system will struggle to establish itself as an authoritative system record and cornerstone for providing qualified accountability. Blockchain techniques may offer the needed trust, transparency, and auditability. C. Coherence and Consistency The Minding Organization describes problem-solving strategies to envision the (desirable) future and bring it into the present [9]. This facility to shift perspective through time and across organizations with a coherent frame of reference develops a continuum mindset. This perspective pursues coherent (consistently logical) thought spanning intervals of time and crossing organizational discontinuities, to deliver precocious enterprise situational awareness and to robustly adapt to emerging knowledge. Records of decision provenance support a coherent frame of reference from which to understand, learn, and adapt over time, offering decision makers in the present the opportunity to act with the advantage of future insights. This capability can uncannily mediate the effects of VUCA that foster inconsistency, wasteful deliberation, and mediocrity resulting from hedging against accountability. Implemented with a robust ontology or elegant language model, the process of submitting decision provenance can detect inconsistencies between previously archived assumptions, criteria, and identified unknowns, flagging contradiction to the attention of both the present and prior decision makers. This promotes improved coherence between stakeholders and over time: For instance, when prescribing a newly available drug with milder side effects, identifying prior patients under similar treatment who may now benefit from switching. Future reassessment of decisions may then be carried out with the benefit of leveraging relevant prior efforts that considered figures of merit, candidate alternatives, processes to reproduce supporting analyses and previous results, mappings between dependent decisions, and even further cross-linking via like assumptions and unknowns. An example is prior effort selecting part suppliers and performance metrics to apply in contract management. D. Adaptation by Design Digital systems customarily perform searches against archived data retrospectively. But affordable abundance of processing in modern digital technology offers opportunity to build far more sophisticated information infrastructure. Entwining decisions with contextual provenance enables language-savvy software agents to perform directed semantic searches on unstructured data feeds to find information valuable for clarification and consistency. An example is intelligent agents that persistently search emerging data, then react in a
5
prescribed manner to changes (volatility), developments and discoveries (uncertainty), and clarifications (ambiguity). This machinery offers the capability to perform adaptation by design. For example, the system for enterprise knowledge stewardship may employ an agentic framework to periodically or continually monitor sensitive criteria, assumptions, or unknowns against diverse information feeds. When a contradiction, confirmation, or discovery is detected, an agent can alert relevant stakeholders, indicating specific decision dependencies. Looking from the vantage of the decision in the past, this agent performs a search into the future to allow correction of a flawed assumption and revision of related decisions. This capability empowers consequential decisions to be made with greater confidence and reduced time, effort, and conservatism imposed by the prevailing uncertainties of the past: for instance, commitment to approve a product design on the condition that the product is never installed above a certain altitude. In effect, the system acts as a safety net to hedge upon a decision conditionally, aware that a priori recorded concerns of consequence (known unknowns or asserted assumptions) will be persistently monitored and trigger prescribed review upon confirmation (or contradiction). The integration of high-consequence decisions and contextual provenance into a semantics-capable agentic system offers a tremendous opportunity toward an awakened enterprise [10]. For example, such a system would be capable of executing other novel forms of risk mitigation, such as monitoring a catastrophic “black swan” scenario by submitting a contingent antidecision, encoding into the provenance failure modes, or similar factors meriting vigilance in rare but high-risk scenarios meriting proactive lead time for adaptive response. IV. C ONCLUSION Systematized data gathering and knowledge management underlie the “awakened enterprise” strategy to collect, manage, and monitor decision provenance to tame VUCA complexity and pursue agility over volatility, understanding over uncertainty, and judgment over ambiguity. The visibility and consistency of decision context over spans of time and across organizational boundaries affords decision makers’ qualified accountability and ability to act despite uncertainty and opportunistically adapt to emergent changes in the future. Timely, proactive intervention reduces overmanagement and mitigates fear of retrospective liability. This approach promotes the candor and transparency for effective organizational operations and learning. R EFERENCES [1] C. Kozyrkov, Introduction to Decision Intelligence. (Oct. 8, 2019). Accessed: May 1, 2026. [Online Video]. Available: https://decision. substack.com/p/introduction-to-decision-intelligence-569 [2] R. Arthur et al., “Digital twin: Definition & value,” Amer. Inst. of Aeronaut. and Astronaut., Reston, VA, USA, Dec. 2020. [Online]. Available: https://doi.org/10.2514/9.wpdeic2020dtdv [3] R. Arthur, “Irresponsible caution,” LinkedIn, 2023. Accessed: May 1, 2026. [Online]. Available: https://www.linkedin.com/pulse/ irresponsible-caution-rick-arthur/ [4] H. Simon, Models of Man: Social and Rational. New York, NY, USA: Wiley, 1957.
COMPUTING IN SCIENCE & ENGINEERING, VOL. X, NO. X, MONTH–MONTH 2026
[5] N. Bennett and G. J. Lemoine, “What VUCA really means for you,” Harvard Business Review, Jan. 2014. [Online]. Available: https://hbr.org/ 2014/01/what-vuca-really-means-for-you [6] U. Gasser and V. Mayer-Schönberger, Guardrails: Guiding Human Decisions in the Age of AI. Princeton, NJ, USA: Princeton Univ. Press, 2024. [7] C. Hood, The Blame Game: Spin, Bureaucracy, and Self-Preservation in Government. Princeton, NJ, USA: Princeton Univ. Press, 2010, doi: 10.1515/9781400836819. [8] “PROV overview,” W3C Working Group, Wakefield, MA, USA, 2013. [Online]. Available: https://www.w3.org/TR/prov-overview/ [9] M. Rubinstein and I. Firstenberg, The Minding Organization: Bring the Future to the Present and Turn Creative Ideas into Business Solutions. New York, NY, USA: Wiley, 1999. [10] R. Arthur, “Awakened enterprise: Adaptation by design to mitigate uncertainty and incomplete knowledge,” LinkedIn, 2024. Accessed: May 1, 2026. [Online]. Available: https://www.linkedin.com/pulse/ awakened-enterprise-rick-arthur-kwitc/
RICHARD B. ARTHUR serves as a senior principal engineer at GE Aerospace, Niskayuna, NY, 12309, USA. His research interests include computational methods, knowledge systems, and model-based engineering. Contact him at [email protected].
6