Abstract
Traditional simulations and wargames evaluate plans, train decision makers, and assess mission outcomes within predefined operational scenarios. Although these environments may incorporate uncertainty, adaptive adversaries, and realistic operational complexity, they generally treat the decision environment as part of the simulation background rather than as the primary object of analysis. Modern influence operations increasingly exploit this limitation by modifying the conditions under which decisions are formed, progressively degrading trust, coordination, legitimacy, and other factors essential to effective decision making.
This paper introduces the Decision Environment Laboratory (DEL), an AI-enabled laboratory that treats the decision environment as an explicit object of representation, experimentation, and analysis. DEL is founded on a dual-use architecture supporting two complementary applications. In Learning Mode, it enables organizations to train leaders and analysts to recognize, diagnose, and manage decision-environment health under conditions of uncertainty and influence. In Decision Support Mode, it enables planners to develop and evaluate alternative courses of action by assessing how environmental changes and influence activities may affect decision-environment health before decisions are implemented.
DEL represents decision environments in terms of variables, their values, and the constraints defined over them, providing a common foundation for experimentation, analysis, learning, and operational decision support. Its adaptive collaboration architecture integrates environment generation, stakeholder simulation, environment monitoring, decision-environment analysis, and user response evaluation within a unified framework that continuously adjusts the allocation of cognitive responsibility between human users and artificial intelligence according to user proficiency, operational objectives, and context. The paper develops the conceptual foundations of decision environments, describes the architecture of DEL, and outlines a research agenda for measuring decision-environment health, evaluating interventions, and developing the scientific principles required to engineer decision environments that remain healthy under conditions of uncertainty, disruption, and sustained influence. By transforming decision environments from implicit features of operational settings into explicit objects of systematic scientific investigation, DEL establishes a foundation for a new class of AI-enabled laboratories for studying and engineering decision-environment health.
Introduction
Simulation has long been one of the principal tools for preparing organizations to operate under uncertainty. Military wargames, emergency management exercises, business simulations, and policy planning environments allow participants to evaluate plans, explore alternative courses of action, and gain experience before confronting real-world events. Advances in artificial intelligence have further expanded these capabilities through adaptive scenarios, autonomous agents, and increasingly realistic operational environments.
Despite these advances, most simulations share a common assumption: they evaluate decisions within a decision environment whose essential characteristics are treated as part of the operational scenario rather than as an object of study in its own right. While these environments may evolve during an exercise, participants are typically assessed by the quality of their decisions and mission outcomes rather than by their ability to recognize, diagnose, and preserve the conditions that make sound decision making possible.
This assumption is becoming increasingly problematic. Modern influence operations seek strategic advantage not simply by changing information or persuading individuals, but by modifying the conditions under which organizations interpret information, coordinate activity, assess risk, establish legitimacy, and maintain trust. Their objective is to reshape the decision environment so that poor decisions become progressively more likely, even when much of the available information remains accurate. Understanding and preserving the health of the decision environment has therefore become an operational requirement in its own right.
In this paper, a decision environment consists of the variables, their values, and the constraints defined over them that together determine the conditions under which decisions are formed. The health of a decision environment reflects an organization’s ability to maintain or restore the conditions under which sound decisions can be formed despite uncertainty, disruption, and external influence. When that adaptive capacity is exceeded, degradation accumulates, reducing the organization’s ability to coordinate effectively and make sound decisions.
The question, then, is how organizations can systematically study, measure, and improve decision-environment health. Existing simulations provide only limited support because they treat the decision environment as the backdrop for decision making rather than as the object of experimentation. To address this limitation, this paper introduces the Decision Environment Laboratory (DEL), an AI-enabled laboratory for experimentation with decision environments. Unlike conventional simulations, DEL treats the health of the decision environment as the primary object of study rather than the evaluation of the decisions made within it.
DEL is founded on a dual-use architecture supporting two complementary applications. In Learning Mode, it enables organizations to train leaders and analysts to recognize, diagnose, and manage the health of decision environments under conditions of uncertainty and influence. In Decision Support Mode, it enables planners to develop and evaluate alternative courses of action by assessing how environmental changes and influence activities may affect decision-environment health before decisions are implemented. Artificial intelligence serves as the enabling technology for both modes by constructing realistic operational settings, simulating stakeholders and institutions, monitoring indicators of decision-environment health, and providing analytical support throughout the experimentation process.
The remainder of the paper first develops the conceptual model underlying decision environments and decision-environment health. It then describes the architecture of DEL, its dual-use operating modes, and the research opportunities created by treating decision-environment health as a measurable and manageable property of complex organizational systems.
Decision Environments as the Object of Study
To support systematic study of decision-environment health, DEL represents the decision environment as a first-class object whose properties can be observed, analyzed, manipulated, and evaluated. This explicit representation provides the foundation for both experimentation and decision support by making it possible to analyze how changes in the environment influence decision-making.
The decision environment is represented by variables, their values, and the constraints defined over them. Variables represent properties of the environment that may change over time, while constraints define the relationships and conditions that govern how those variables influence decision making. Together, they characterize the operational conditions within which organizations interpret information, coordinate activities, allocate authority, and select courses of action.
The values of variables may be obtained through direct observation, measurement, or inference. Some variables, such as available resources or communication latency, may be directly measurable. Others, such as organizational trust, perceived legitimacy, or confidence in leadership, are not directly observable but may be inferred from behavioral indicators or other evidence. Regardless of how they are obtained, variable values represent the current state of the decision environment and provide the basis for assessing change over time.
Constraints are parameterized conditions defined over one or more variables. A particular constraint is instantiated by assigning values to its parameters. For example, an organization may require that emergency evacuation decisions be approved by the Emergency Operations Director. During a rapidly evolving crisis, that same constraint may be instantiated differently by delegating approval authority to designated local commanders. The underlying constraint remains unchanged, while its instantiation adapts to changing operational conditions. As the decision environment evolves, constraints may be introduced or removed, existing constraint instantiations may change, and variable values may change, collectively altering the conditions under which decisions are formed.
Treating health as an explicit property of the decision environment makes it possible, in principle, to observe, assess, and compare it across different situations. Although health is not directly measurable as a single quantity, it is reflected in measurable characteristics of the environment, including the evolution of its variables, the satisfaction of its constraints, and the organization’s ability to sustain effective decision-making conditions over time.
Changes within a decision environment rarely remain localized. Because variables and constraints are interconnected through the activities of the people and organizations operating within the environment, local changes may propagate outward as participants adapt their behavior, modify procedures, redistribute authority, or establish new constraints in an effort to maintain effective decision making. The resulting propagation is therefore not an intrinsic property of the representation itself, but a consequence of the continual actions taken by individuals and organizations responding to changing conditions.
Representing decision environments in this way establishes the conceptual foundation for DEL. It provides a common representation for observing environmental change, assessing decision-environment health, evaluating alternative interventions, and comparing their effects over time. The following section describes how DEL employs this representation within an AI-enabled laboratory designed to support both learning and operational decision support.
DEL Architecture
DEL is founded on a dual-use architecture that supports both learning and operational decision support. Beneath these applications lies a common adaptive collaboration architecture in which human and artificial intelligence work together to understand, evaluate, and preserve decision-environment health. Rather than defining separate AI capabilities for the two applications, DEL employs a single architectural framework that adaptively allocates cognitive responsibility between the human and the AI.
Building upon the representation introduced in the previous section, DEL integrates five complementary AI capabilities within a common experimental framework: environment generation, stakeholder simulation, environment monitoring, decision-environment analysis, and user response evaluation. These capabilities operate continuously in both modes. The modes differ not in architecture, but in how their analytical results are disclosed to the user and incorporated into decision-making.
At the core of DEL is the explicit representation of the decision environment. Every AI capability operates on this common representation of variables, their values, and the constraints defined over them. Environment generation modifies the representation, stakeholder simulation acts upon it, environment monitoring observes its evolution, decision-environment analysis interprets its behavior, and user response evaluation compares the user’s responses with DEL’s assessment of the evolving decision environment. Because every capability operates on the same representation, DEL maintains a coherent model of the evolving decision environment.
The first AI capability is environment generation. DEL constructs realistic operational settings tailored to learning objectives or planning problems by generating organizations, resources, communication structures, operational constraints, uncertainties, and influence activities. Unlike traditional scripted simulations, these environments evolve dynamically as conditions change and as participants interact with them, creating continually evolving conditions for experimentation.
The second capability is stakeholder simulation. AI represents the behavior of individuals, organizations, institutions, media, and other actors whose interactions shape the decision environment. These simulated stakeholders exchange information, pursue objectives, allocate authority, establish or modify constraints, and adapt to changing conditions. Their interactions provide the mechanism through which local changes propagate throughout the decision environment, producing broader changes that influence subsequent decisions.
The third capability is environment monitoring. As the decision environment evolves, DEL observes variables, constraint satisfaction, propagation patterns, and other measurable characteristics of the environment. These observations maintain an up-to-date representation of the evolving decision environment and provide the empirical foundation upon which subsequent analysis is performed.
The fourth capability is decision-environment analysis. Using information produced by the monitoring capability, DEL continuously analyzes the evolving decision environment to identify significant changes, determine their causes, assess their implications for decision-environment health, evaluate alternative interventions, and forecast potential consequences. What changes is the degree to which they are disclosed to the user. During learning, analytical results may be withheld, partially revealed, or progressively disclosed as users develop and demonstrate their own analytical abilities. During operational decision support, the same analyses are made fully available to augment planning and decision-making.
The fifth capability is user response evaluation. DEL evaluates the user’s observations, diagnoses, predictions, proposed interventions, and decisions by comparing them with its own assessment of the evolving decision environment. These evaluations are accompanied by explanations that identify supporting evidence, overlooked factors, alternative interpretations, and likely consequences. In Learning Mode, it develops diagnostic and decision-making skills. In Decision Support Mode, it functions as a partner that challenges assumptions, exposes blind spots, and strengthens planning through constructive critique.
Because the underlying collaboration between the human and the AI is continuously adjustable, Learning Mode and Decision Support Mode represent different operating regions of the same adaptive collaboration architecture rather than distinct system configurations. Adaptive disclosure provides the mechanism through which cognitive responsibility shifts between the human and the AI, allowing analytical support to increase or decrease according to user proficiency, operational objectives, and context. DEL therefore functions as a continuously adjustable human-AI collaboration architecture rather than a fixed decision-support system.
Research Agenda
DEL establishes a laboratory for the systematic study of decision environments. Its value extends beyond supporting training and operational planning by enabling controlled experimentation on questions that have previously been difficult to investigate empirically. Because decision environments are represented explicitly, evolve dynamically, and can be repeatedly manipulated under controlled conditions, DEL creates new opportunities to develop theories, measurement methods, analytical techniques, and intervention strategies for preserving decision-environment health.
Several research challenges naturally emerge.
4.1 Measuring Decision-Environment Health
The first challenge is to develop practical methods for observing and measuring decision-environment health.
Although health is not directly observable as a single quantity, it may be inferred from measurable characteristics of the evolving decision environment, including the evolution of variables, constraint satisfaction, propagation dynamics, organizational adaptation, and other indicators associated with effective decision making.
Research questions include:
- Which indicators most reliably reflect decision-environment health?
- Which combinations of indicators provide early warning of degradation?
- How should uncertainty in health estimates be represented?
- Can health measures be generalized across organizations and domains?
Developing reliable methods for estimating decision-environment health represents one of the central scientific challenges enabled by DEL.
4.2 Understanding Propagation
Because DEL explicitly represents relationships among variables, constraints, organizations, and stakeholders, it provides a unique opportunity to study how changes propagate through decision environments.
Research questions include:
- Which environmental changes remain localized?
- Which changes cascade throughout the environment?
- What organizational characteristics amplify propagation?
- What characteristics dampen propagation?
Understanding propagation mechanisms may ultimately prove as important as understanding the original influence activity itself.
4.3 Evaluating Intervention Strategies
Traditional simulations often evaluate decisions.
DEL makes it possible to evaluate interventions directed at the decision environment itself.
Examples include
- modifying organizational structure,
- redistributing authority,
- improving communication,
- introducing redundancy,
- strengthening trust,
- modifying operational procedures.
Rather than asking whether an intervention improves mission performance directly, DEL asks whether it preserves or restores decision-environment health under changing conditions.
4.4 Adaptive Human-AI Collaboration
DEL also enables investigation of adaptive collaboration between humans and artificial intelligence.
Because the architecture continuously adjusts the allocation of cognitive responsibility through adaptive disclosure, it becomes possible to study questions such as:
- How should analytical support change as expertise develops?
- When should AI provide guidance?
- When should it deliberately withhold guidance?
- How does adaptive disclosure affect learning, trust, and decision quality?
- What forms of collaboration produce the greatest long-term human expertise?
Although developed here in the context of decision environments, these questions may have broader implications for future human-AI systems.
4.5 Towards Engineering Healthy Decision Environments
Perhaps the most important long-term question concerns design.
Current organizations are generally designed for operational effectiveness.
Future organizations may also need to be designed for decision-environment resilience.
DEL provides a laboratory for investigating questions such as:
- Which organizational structures naturally resist degradation?
- Which decision processes remain robust under sustained influence?
- Which communication architectures preserve coordination most effectively?
- How should authority structures adapt under uncertainty?
These questions shift attention from responding to disruption toward designing decision environments that remain healthy despite continual influence.
Conclusion
Decision-making has traditionally been studied by examining the quality of decisions, the performance of decision makers, and the outcomes that result from their choices. The decision environment itself has largely remained an implicit part of the operational setting within which those decisions are made. As influence operations increasingly seek to modify the conditions under which organizations interpret information, coordinate activity, establish trust, and exercise authority, this distinction becomes increasingly important. Understanding the health of the decision environment is no longer simply a matter of context; it has become an operational requirement.
The Decision Environment Laboratory (DEL) addresses this challenge by treating the decision environment as an explicit object of representation, experimentation, and analysis. By representing decision environments in terms of variables, their values, and the constraints defined over them, DEL provides a common foundation for studying how decision environments evolve, how degradation propagates, how interventions alter those dynamics, and how decision-environment health can be observed and assessed over time. Its adaptive collaboration architecture enables these investigations to support both learning and operational decision support within a single continuously adjustable human-AI collaboration framework.
More fundamentally, DEL transforms decision environments from implicit features of operational settings into explicit objects of scientific investigation. Once represented explicitly, decision environments can be observed, manipulated, compared, and evaluated under controlled and repeatable conditions. This creates opportunities not only to improve training and operational planning, but also to develop a cumulative science of decision-environment health grounded in experimentation rather than observation alone.
The long-term objective is not merely to understand decision environments, but to develop the scientific principles required to engineer decision environments that remain healthy under conditions of uncertainty, disruption, and sustained influence.
Author Biography
Dr. Rand Waltzman has had a 43-year career working in Artificial Intelligence. He served two tours as a program manager at DARPA, where he created and managed the Social Media in Strategic Communications program. He coined the term Cognitive Security and is a founder of the Information Professionals Association.