Bridging the Knowledge Void: A Synthetic Near-Empty Review of Intelligent Evolutionary Games’ Employment in Healthcare

Abstract

Background: The convergence of Evolutionary Game Theory (EGT) and Artificial Intelligence (AI) has established the field of Intelligent Evolutionary Games (IEGs). While IEG applications have flourished in general systems and social sciences, their operationalization within healthcare (IEG Health) remains significantly underdeveloped. This study identifies a “knowledge void” in the literature, where the bottleneck is not a lack of clinical data but a scarcity of frameworks that integrate intelligent strategic modelling into clinical practice. Methods: We employ the Synthetic Near-Empty Review (SNER) framework, utilizing Synthetic Knowledge Synthesis (SKS) and bibliometric triangulation via VOSviewer. Three distinct corpora—IEG Health, EG Health, and IEG All (IEG)—were harvested from Scopus and mapped to identify thematic clusters and translation pathways. Results: The analysis reveals that IEG Health is a nascent domain currently focused on service regulation in elderly care and chronic disease management. We demonstrate a “Translation Framework” to bridge the research void, mapping concepts like Social Trust and Reputation Management from the broader IEG literature into clinical-specific models, such as Doctor-AI Adoption and Adaptive Coordination Games. Conclusions: By shifting from static Replicator Dynamics to Adaptive Learning Strategies (e.g., MARL and Bayesian updating), IEG Health can address critical challenges like algorithm aversion and clinical deskilling. Furthermore, transitioning these models into clinical environments requires the incorporation of structured ethical guidelines, such as ALTAI, to ensure algorithmic accountability. This study provides a structured foundation for future research to transition from theoretical modelling to AI-augmented clinical decision-making.

1. Introduction

The synthesis of Evolutionary Game Theory (EGT) [1] and Artificial Intelligence (AI) [2] has given rise to a potent interdisciplinary framework: Intelligent Evolutionary Games (IEGs) [3,4]. While classical EGT focuses on the dynamics of strategy propagation within populations based on fitness and natural selection, the integration of “intelligence”—ranging from reinforcement learning [5], cognitive modelling [6] and co-evolutionary algorithms [7] to Multi-Agent Reinforcement Learning [6]–introduces a more nuanced dimension to strategic interaction [8].

However, a critical research gap persists within the healthcare sector: the primary bottleneck is not a lack of raw biological or clinical data but a profound scarcity of peer-reviewed publications that operationalize IEG frameworks in real-world healthcare contexts. This creates a “near-empty” research space that hinders the transition from theoretical modelling to clinical practice. To navigate this, we employ the Synthetic Near-Empty Review (SNER) framework. SNER is a methodological approach designed to extract and structure knowledge from emerging or sparse literature landscapes in which traditional systematic reviews are ineffective. The core engine of SNER is Synthetic Knowledge Synthesis (SKS), a process that combines bibliometric mapping with thematic analysis to evaluate the entire corpora of literature, bypassing the sampling bias of small-scale reviews [9]. This approach addresses the problem of reviews being “as empty as possible” by generating a structured, evidence-based foundation where traditional systematic reviews fail. By applying the SNER framework, researchers can identify underlying strategic paradigms from fields characterized by the proven efficacy of IEGs and EGT. This approach serves to bridge the critical void in IEG employment in healthcare through the generation of a systemic, AI-augmented synthesis. In that manner we formulated three research questions:

• RQ1 (Knowledge Void): What is the nature and extent of the “knowledge void” regarding the operationalization of Intelligent Evolutionary Games (IEGs) in real-world clinical practice?
• RQ2 (Thematic Synthesis): What is the taxonomy of foundational AI algorithms, game strategies, and overarching social themes that can be extracted from the broader literature landscapes of general IEGs (IEG All) and standard Evolutionary Games in health (EG Health)?
• RQ3 (Translation Framework): How can these extracted strategic paradigms be successfully translated into clinical-specific models to address modern healthcare challenges such as AI adoption and algorithm aversion?

The main contributions of this study are

○

Identification of a Knowledge Void: We systematically identify the scarcity of frameworks that integrate intelligent strategic modelling into clinical practice, noting that the bottleneck is a lack of peer-reviewed operationalizations [10] rather than clinical data.

○

Methodological Innovation: We apply the Synthetic Near-Empty Review (SNER) framework [9], utilizing bibliometric triangulation to extract structured knowledge from sparse literature landscapes.

○

Translation Framework: We propose a framework that maps broad Intelligent Evolutionary Game (IEG) concepts (like Social Trust) into clinical-specific models (such as Doctor-AI Adoption).

○

Strategic Evolution: We outline the necessary shift from static Replicator Dynamics to Adaptive Learning Strategies to address challenges in healthcare.

2. Methodology

The SNER framework is based on Synthetic Knowledge Synthesis (SKS) [11,12]. SKS exhibits several advantages over traditional review frameworks. It reduces the time and resources required to synthesize articles, thus enabling the analysis of the “whole corpora” of publications (all publications from fields of interest) rather than being limited to small, manually selected samples. In this manner sampling bias is avoided. Additionally, the use of bibliometric landscapes makes it easier to visualize, identify and analyse associations between different author keywords, topics or other units of interest. Furthermore, triangulation” (combining bibliometrics and content analysis of publications metadata) supports a holistic understanding of the research presented in the corpora being synthesized.

The methodological framework for this study is illustrated in Figure 1. Initially, SKS is employed to map the research landscapes of IEG Health (publications presenting the use of IEGs in healthcare), EG Health (publications presenting the use of EGT in healthcare), and IEG All (publications presenting the use of IEGs in general), facilitating the identification of prominent themes within each respective field. Subsequently, a comparative analysis of IEG Health themes against those of IEG All and EG Health is conducted to distinguish specific knowledge areas and themes suitable for translation into the IEG Health void. Each identified theme is then systematically decomposed into its constituent health domains and AI technologies, thereby delineating potential implementations of IEGs within the healthcare sector.

The three corpora presented above were harvested from the Scopus bibliographic data base on 16th of February 2026, using the search strings presented below. No further inclusion or exclusion criteria were used.

• IEG Health: TITLE-ABS-KEY(“evolutionary game*” and (“intelligen*” OR “machine learning” OR “deep learning” OR “intelligent system” OR “support vector machine” OR (“decision tree” AND (induction OR heuristic)) OR “random forest” OR “Markov decision process” OR “hidden Markov model” OR “fuzzy logic” OR “k-nearest neighbours” OR KNN OR “support vector machine*” or SVM OR “naive Bayes” OR “Bayesian learning” OR “artificial neural network” OR “convolutional neural network” OR “recurrent neural network” OR “generative adversarial network” OR “deep belief network” OR “perceptron” OR {natural language processing} OR {natural language understanding} OR {general language model})) AND (LIMIT-TO (SUBJAREA,“MEDI”) OR LIMIT-TO (SUBJAREA,“HEAL”)).
• EG Health: TITLE-ABS-KEY(“evolutionary game*”) AND (LIMIT-TO (SUBJAREA,“MEDI”) OR LIMIT-TO (SUBJAREA,“HEAL”)).
• IEG All: TITLE-ABS-KEY(“evolutionary game*” and (“intelligen*” OR “machine learning” OR “deep learning” OR “intelligent system” OR “support vector machine” OR (“decision tree” AND (induction OR heuristic)) OR “random forest” OR “Markov decision process” OR “hidden Markov model” OR “fuzzy logic” OR “k-nearest neighbours” OR KNN OR “support vector machine*” or SVM OR “naive Bayes” OR “Bayesian learning” OR “artificial neural network” OR “convolutional neural network” OR “recurrent neural network” OR “generative adversarial network” OR “deep belief network” OR “perceptron” OR {natural language processing} OR {natural language understanding} OR {general language model})).

Using the aforementioned query strings, the search yielded 16 publications for IEG Health, 335 publications for EG Health, and 1124 publications for IEG All.

Bibliometric mapping was conducted using VOSviewer (version 1.6.20) [13]. To optimize the clustering algorithm, the association strength method was applied. The number of occurrences of a keyword was calculated using Zipf law [12] (IEG = 40 keywords; EG Health = 33 keywords and IEG Health = 19 keywords). All other VOSViewer parameters were set to default values. Node sizes in the resulting maps were interpreted based on keyword frequency, while the thickness of the links represents the co-occurrence strength between terms. Colours were automatically assigned by the software to denote distinct thematic clusters representing distinct subfields of research.

The resultant research landscapes, which illustrate the conceptual clusters and evolutionary trajectories of the field, are presented in Figure 2, Figure 3 and Figure 4. Complementing this visualization, the findings derived from the systematic content analysis of these landscapes are synthesized in

Table 1. The results of the thematic analysis.

	IEG All	EG Health	IEG Health
Overreaching research themes	Reputation management	Public participation in governmental supervision	Service regulation in community elderly care
	Public good management	Collaborative governance in public health emergencies	Public medical services for chronic disease diagnosing and treatment
	Solving social dilemmas	Green technology innovation	Privacy protection
	Trust evaluation	Government regulation and privacy protection in health crises	Doctors’ adoption of AI-based medicine
	Decision-making	Regulation strategies in elderly care services	Adaptive security of healthcare in smart homes
	Task allocation	Carbon emission reduction
	Smart supply chains	Drug resistance and cancer evolution
	Resource allocation
	Multi-objective optimization
	Value co-creation
	Intelligent construction
	Population dynamics
	Auction
	Social networking
Specific healthcare domains		Privacy protection	Smart home
		Green technologies	AI-based medicine
		Cancer evolution	Chronic diseases
		Drug resistance Food and product safety (MAH—Marketing Authorization Holder mechanism)
		Immunotherapy
		Disease transmission
		Angiogenesis
Evolutionary game strategies	All from EG Health	Cellular automata
	Collective intelligence	Prisoners’ dilemma
	Replicator dynamics	Darwinian dynamics
	Nash equilibrium	Cumulative prospect theory
	Stackelberg game	Hawk–dove game
	Cooperation dynamics	Reciprocal altruism
	Snowdrift game	Agent-based technologies
Integrated AI algorithms and domains	All from IEG Health		Deep learning
	Swarm intelligence		Large language models
	Neural network		Genetic algorithms
	Federated learning		Adversarial search
	Edge intelligence
	Deep reinforcement learning
	Multi-agent reinforcement learning
	Q-learning

. This table delineates the corpus across four critical dimensions: (i) overarching research themes, (ii) specific healthcare domains, (iii) evolutionary game strategies, and (iv) integrated AI algorithms and paradigms.

3. Results

3.1. Knowledge Void

The search resulted in 16 publications presenting the use of IEGs in health, 335 publications presenting the employment of EG in health and 1124 presenting the use of IEG in general. The research landscapes generated by VOSViewer are shown in Figure 2, Figure 3 and Figure 4, and the result of thematic analysis are presented in Table 1. The themes across the three corpora of publications reflect a transition from broad social management to specialized medical regulation.

• IEG All: Focuses on high-level systemic issues such as reputation management [14], public good management [15], and solving social dilemmas [16]. It also touches on infrastructure and logistics like smart supply chains [17] and intelligent construction [17].
• EG Health: Shifts the focus toward governance and public safety, emphasizing collaborative governance in emergencies [18], green technology innovation [19], and governmental supervision [20].
• IEG Health: Narrowly targets the intersection of technology and medicine, specifically service regulation in elderly care [21], AI adoption by doctors, and smart homes [22].

Table 1 highlights where these game-theoretic models are applied within the healthcare field, as shown below:

• EG Health Applications: Covers biological and systemic health issues including cancer evolution, drug resistance, immunotherapy, and disease transmission [23,24,25,26]. It also addresses industrial health factors like food and product safety [27].
• IEG Health Applications: Focuses on technology-driven healthcare delivery, such as smart homes, cancer treatment, and chronic disease management [28]. An example of real-world IEG use in clinical settings is adaptive oncology. Namely, high-dose chemotherapy frequently fails by eradicating drug-sensitive cells, inadvertently selecting for chemo-resistant phenotypes. IEG models address this by integrating pharmacokinetics to calculate dynamic, adaptive dosages. By maintaining competitive suppression between sensitive and resistant cellular subpopulations, these algorithms mitigate rapid mutation and significantly prolong patient survival [29].
• General IEG/EG Strategies: Utilize classic models like the Prisoner’s Dilemma, Snowdrift game, and Hawk–dove game [30,31,32].
• Dynamics and Theory: Both domains rely on Replicator Dynamics, Nash Equilibrium, and Stackelberg Games to understand how cooperation or competition evolves over time [33,34,35].
• Advanced Modelling: Includes Cellular automata, Cumulative prospect theory, and Agent-based technologies to simulate more-complex, non-linear human or other health agents’ behaviours [35,36].
• Broad Intelligence (IEG All): Leverages a wide array of AI techniques, including Swarm Intelligence, Federated Learning, and Edge Intelligence. It also uses Multi-Agent Reinforcement Learning and Q-learning for decision-making [37,38,39,40,41].
• Healthcare Intelligence (IEG Health): Specifically highlights the use of Deep learning, Large Language Models (LLMs), Genetic algorithms, and Adversarial search to solve health-specific problems [41,42].

The synthesis indicates that, while IEG All provides the foundational AI tools and broad social theories, EG Health provides the biological and regulatory context. IEG Health represents the synthesis of these two, applying advanced AI (like LLMs and Deep Learning) to specific healthcare challenges such as elderly care regulation and chronic disease diagnosis.

3.2. Thematic Synthesis

3.2.1. The Taxonomy of the Used AI Algorithms

Based on the analysis of Table 1, the taxonomy is provided below.

Machine Learning and Deep Learning Models Encompass Algorithms That Teach Data to Make Predictions, Generate Content, or Recognize Patterns

• Neural Networks [43]: The foundational architecture for deep learning, mimicking human brain nodes.
• Deep Learning [44]: Advanced neural networks with multiple layers used for complex pattern recognition.
• Large Language Models (LLMs) [45]: Advanced deep learning models specifically trained on vast amounts of text to understand and generate human language.

Reinforcement Learning

This subfield focuses on training agents to make sequences of decisions by rewarding desired behaviours.

• Q-learning [46]: A foundational, value-based reinforcement learning algorithm used to find the best action to take given a current state.
• Deep Reinforcement Learning (DRL) [47]: The integration of deep learning neural networks with reinforcement learning principles to solve highly complex, high-dimensional problems.
• Multi-Agent Reinforcement Learning (MARL) [48]: An extension of RL in which multiple AI agents interact, compete, or cooperate within the same environment to achieve their goals.

Distributed and Decentralized AI

These are paradigms and algorithmic frameworks designed to train or deploy AI models across multiple devices or nodes without centralizing the raw data.

• Federated Learning [49]: A privacy-preserving machine learning technique in which a model is trained across multiple decentralized edge devices holding local data samples, without exchanging them.
• Edge Intelligence [50]: The deployment of AI algorithms directly on edge devices (like sensors or smartphones) to process data locally and reduce latency.

Bio-Inspired and Evolutionary Optimization

These algorithms are inspired by biological evolution and the collective behaviour of decentralized, self-organized systems.

• Genetic Algorithms [51]: Search heuristics inspired by the theory of natural evolution (mutation, crossover, selection) used to solve optimization problems.
• Swarm Intelligence [51]: Algorithms based on the collective behaviour of decentralized, self-organized systems (e.g., ant colonies or bird flocking).
• Collective Intelligence [51]: A broader concept in which shared or group intelligence emerges from the collaboration and competition of many individuals or agents.

Search and Multi-Agent/Game-Theoretic AI

Algorithms and models used for decision-making in competitive or cooperative environments.

• Adversarial Search [52]: Algorithms used in competitive multi-agent environments (like board games) to anticipate and counter an opponent’s moves.
• Agent-based Technologies [4]: Computational models for simulating the actions and interactions of autonomous agents to assess their effects on the system as a whole.
• Cellular Automata [53]: Discrete mathematical models consisting of a grid of cells that evolve through discrete time steps according to a set of rules based on the states of neighbouring cells (highly utilized in modelling complex systems and spatial games).

3.2.2. The Taxonomy of the Used Evolutionary Game Strategies

In Evolutionary Game Theory, the focus shifts from hyper-rational individuals making isolated decisions to populations of individuals interacting over time, where successful strategies reproduce and spread. The taxonomy is provided below.

Classic Game Models

These are standard mathematical scenarios used to study conflict, cooperation, and social dilemmas among interacting agents.

• Prisoner’s Dilemma [54]: The foundational game illustrating why two completely rational individuals might not cooperate, even if it appears that it is in their best interests to do so.
• Snowdrift Game [55]: Also known as the game of “Chicken.” A situation in which both players want the other to yield, but a failure to yield by both results in the worst possible outcome for both. It is often used to model the evolution of cooperation when mutual defection is disastrous.
• Hawk–Dove Game [56]: A model used to analyse the evolution of animal behaviour and conflict over resources, in which “Hawks” are aggressive, and “Doves” are peaceful, yielding to aggression.

Solution Concepts and Strategic Frameworks

These concepts define what constitutes a “stable” state or a winning strategy within a game.

• Nash Equilibrium [57]: A state in a game in which no player has an incentive to deviate from their chosen strategy after considering an opponent’s choice. In evolutionary terms, it often relates to an Evolutionarily Stable Strategy (ESS).
• Stackelberg Game [36]: A strategic game in economics and game theory in which the leader firm (or agent) moves first, and then the follower firms move sequentially. It models hierarchical decision-making.

Evolutionary Dynamics

These mathematical frameworks describe how the frequency of different strategies in a population changes over time.

• Replicator Dynamics [8]: A fundamental mathematical model in Evolutionary Game Theory that dictates that strategies with a higher payoff than the average population payoff will reproduce and increase in frequency.
• Darwinian Dynamics [58]: A broader term for the mechanisms of natural selection, mutation, and inheritance that drive the evolution of strategies in a population.
• Cooperation Dynamics [59]: The specific study of how cooperative strategies emerge, stabilize, or collapse within a population of self-interested agents over time.

Behavioural Theories and Mechanisms

These concepts explain the psychological or biological mechanisms that influence how agents make choices or why specific behaviours persist.

• Reciprocal Altruism [60]: A behavioural mechanism in which an organism acts in a manner that temporarily reduces its fitness while increasing another organism’s fitness, with the expectation that the other organism will act in a similar manner at a later time (often modelled using repeated Prisoner’s Dilemma).
• Cumulative Prospect Theory [61]: A behavioural economic theory describing the way people choose between probabilistic alternatives that involve risk, weighting the probability of extreme events heavily. It modifies traditional game theory by accounting for real-world human psychological biases.

3.3. The Translation from EG Health and IEG All to IEG Health (Translation Framework)

The integration of Intelligent Evolutionary Games (IEGs) into the healthcare sector necessitates a multidimensional transposition of existing theoretical frameworks. This process involves the systematic recalibration of four core pillars: the themes of social governance, the contextual shift of health domains, the underlying logic of strategic interaction, and the computational engines of Artificial Intelligence. More precisely, to operationalize this translation, researchers should follow a four-step iterative process: (1) Identify the Social Governance Theme: Extract a broad mechanism from IEG All (e.g., Trust Evaluation). (2) Define the Health Domain Context: Map this mechanism to a specific clinical need (e.g., Doctor-AI Adoption). (3) Adapt the Strategic Logic: Recalibrate payoff matrices to account for clinical risk asymmetry and accountability. (4) Select the Computational Engine: Implement the appropriate adaptive learning AI (e.g., Federated Learning) to execute the model dynamically.

The following analysis explores the transition from static evolutionary dynamics to Adaptive Learning Strategies, ensuring that AI-driven healthcare systems remain resilient, secure, and patient-centric.

3.3.1. The Translation of Social Governance Themes

Foundational IEG constructs, such as Reputation Management and Trust Evaluation, are operationalized within the medical context as “Clinician AI-Adoption” and “Patient Digital-Health Trust.” Within the IEG Health framework, the “intelligent” component enables the simulation of trust trajectories as they evolve in response to AI-generated clinical interventions. Crucially, this translation expands “Reputation Management” beyond simple linear Bayesian updating by acknowledging human cognitive heuristics. It introduces “Uncertainty Communication,” positing that an AI agent accrues reputation and trust not solely by providing correct diagnoses but also by accurately communicating its confidence intervals and explicitly admitting its operational limits. Furthermore, traditional themes of Public Participation and Governmental Supervision are translated into the “dynamic regulation of community-based elderly care.” While traditional models address policy at a macro-level, IEG Health introduces an algorithmic layer, utilizing data-driven heuristics to regulate these services in real time.

3.3.2. The Contextual Shift of Health Domains

The transition across health domains involves an ontological shift from passive biological processes to autonomous agent-based simulations. This transition in the literature from broad EG Health domains to specialized IEG Health applications represents a profound ontological shift, driven by distinct clinical imperatives. While EG Health operates predominantly at a macro-level—addressing collaborative governance in health emergencies and public participation in governmental supervision—this high-level approach lacks the granular responsiveness required for direct patient care. The clinical driver prompting the narrowing focus onto community elderly care, smart homes, and chronic disease management is the urgent need for real-time, dynamic regulation. Traditional governance models approach health supervision as a static policy concern. However, the intricacies of modern healthcare demand a transition from static regulatory compliance to the implementation of self-evolving protocols. By utilizing data-driven heuristics, IEG Health introduces a crucial algorithmic layer capable of regulating community-based services dynamically. Traditional biological conflicts are thus re-contextualized by replacing passive entities with intelligent agents:

• From Drug Resistance to AI-Augmented Chronic Disease Management: Rather than modelling stochastic bacterial evolution, IEG Health models how a patient’s “intelligent” wearable interface adaptively modifies treatment schedules to pre-emptively mitigate resistance.
• From Privacy Protection to Adaptive Security in Smart Environments: This entails a shift from static regulatory compliance to the implementation of self-evolving, AI-driven security protocols designed to safeguard sensitive health telemetry within smart-home ecosystems.

3.3.3. The Logic of Strategic Interaction

The mathematical foundations of IEGs—including the Prisoner’s Dilemma, Nash Equilibrium, and Stackelberg Games—are adapted to model the complex interactions between humans and AI agents. In this setup, models must explicitly address accountability and risk asymmetry. If an AI fails, the human physician bears the legal, ethical, and emotional burden, creating a stark imbalance in payoffs. Therefore, future IEG Health payoff matrices must be recalibrated to penalize “blind compliance” while heavily rewarding strategies centred on “Cognitive Augmentation,” ensuring that the human player’s disproportionate risk is accurately accounted for in the evolutionary game. A highly profound theoretical advancement within the IEG Health framework is the transformation of the underlying logic governing strategic interactions. In conventional health-based evolutionary game models, mechanisms such as Replicator Dynamics are constrained by fixed differential equations, which limit their utility in unpredictable clinical environments. IEG Health fundamentally transposes these static equations into Adaptive Learning Strategies by embedding engines like Multi-Agent Reinforcement Learning (MARL) and Bayesian learning directly into the game architecture. This transposition from static mathematics to adaptive learning is essential for solving two of the most critical socio-technical challenges in modern medicine: mitigating algorithm aversion and preventing clinical deskilling. For example, a “Hawk-Dove” game applied to healthcare resource allocation is reconfigured as a Deep Reinforcement Learning (DRL) problem, wherein agents (such as hospitals or autonomous systems) converge toward equilibrium through iterative trial and error and environmental feedback.

3.3.4. Intelligent Computational Engines

The most critical architectural translation involves the deployment of specialized AI paradigms to drive the IEG Health engine:

• Federated Learning: This facilitates the training of robust diagnostic models across heterogeneous hospital networks while maintaining data siloization, thereby addressing the core theme of privacy.
• Multi-Agent Reinforcement Learning: This is utilized to model the high-dimensional interactions between diverse stakeholders within chronic disease or smart-home ecosystems.
• Swarm Intelligence: This paradigm is applied to surgical robotics and decentralized diagnostic networks in which coordinated units collaborate heuristically to execute complex medical tasks.

In addition to the above, we must note that transitioning IEG models into clinical care introduces profound ethical risks. If the evolutionary algorithms are trained on historically biased data, the resulting “winning strategies” may inadvertently penalize marginalized patient groups, allocating fewer resources to them in triage or chronic care simulations. To ensure patient safety, IEG Health systems must implement strict algorithmic bounding—hard-coded clinical guardrails that prevent the AI from suggesting treatments outside established medical guidelines. Continuous fairness auditing, adhering strictly to frameworks like ALTAI, is mandatory to prevent these algorithms from optimizing for efficiency at the expense of equitable care.

4. Discussion

The SKS reveals a distinct narrowing of focus:

• IEG All publications deal with high-level systemic issues like public good management, broad social dilemmas, and reputation management.
• EG Health publications transition toward governance, public safety, and passive biological modelling, such as disease transmission and cancer evolution.
• IEG Health publications present the highly specialized convergence of technology and clinical medicine, targeting specific service regulations in elderly care, chronic disease management, and adaptive security in smart homes.

This specialization is driven by the urgent need to bridge a documented “knowledge void” in the healthcare sector. The bottleneck in modern medicine is no longer a lack of raw clinical data; instead, there is a profound scarcity of frameworks capable of translating intelligent strategic modelling into real-world clinical practice. Consequently, researchers are undergoing an “ontological shift”—moving away from modelling passive biological processes (like stochastic bacterial evolution) to creating autonomous, agent-based simulations in which human practitioners and AI entities actively co-evolve.

To accomplish this shift, foundational AI tools and broad social theories are being synthesized with medical regulatory contexts. Advanced AI algorithms, specifically Deep Learning and Large Language Models (LLMs), are serving as the new “intelligent computational engines” required to solve health-specific problems.

• Handling Complexity: Traditional evolutionary models rely on static differential equations (like Replicator Dynamics) to understand how competition evolves. Deep Learning and LLMs replace these static equations with Adaptive Learning Strategies, allowing systems to process vast, complex data streams for precision diagnosing and real-time medical service regulation.
• Transforming Healthcare Delivery: By acting as the “intelligent” component, these models enable real-time adaptation. For example, concepts like reputation management are translated into algorithm transparency (“Explainable AI”), which is a strict prerequisite for achieving equilibrium in Human–AI games. For a physician to cooperate with an AI agent without succumbing to “algorithm aversion”, they must understand not just the output but the underlying process and data sources. This transparency prevents the “black box” effect, helping doctors assess AI diagnostic accuracy to safely adopt these tools into their workflows.

One of the most critical factors driving the transition to specialized medical regulation is the deployment of MARL to manage incredibly complex, multi-stakeholder environments.

• Managing High-Dimensional Interactions: MARL is explicitly utilized to model the high-dimensional interactions between diverse stakeholders—such as doctors, patients, diagnostic agents, and smart-home sensors—within chronic disease ecosystems.
• Pre-emptive Healthcare: Rather than passively tracking how a disease evolves, MARL allows for proactive interventions. For instance, in chronic disease management, an “intelligent” wearable interface can adaptively modify treatment schedules to pre-emptively mitigate drug resistance based on real-time feedback.
• Human–AI Synergy: MARL is also used to resolve behavioural challenges in clinical settings, such as algorithm aversion (in which doctors constantly override AI) and clinical deskilling (in which doctors blindly trust AI). By continuously learning a doctor’s risk tolerance alongside the AI’s diagnostic strengths, MARL helps the system reach a dynamic Nash Equilibrium, fundamentally driven by the strategic choice of “Active Human Oversight.” Evolutionary dynamics, such as Replicator Dynamics, can be structured so that the highest fitness and payoff are awarded to a “co-evolving partnership” rather than to physicians who passively delegate tasks to the AI, resulting in a highly cooperative Human–AI healthcare team.

4.1. Conceptual Use Case

To demonstrate how Social Trust from IEG All might translate into Doctor Adoption of AI in IEG Health. For demonstration purposes we employed a simple but more realistic adopt/override dynamic in an intelligent evolutionary game in which the “players” (the doctor and the AI) must reach a stable equilibrium of cooperation. In the broader IEG All context, trust is a general metric for system stability. In IEG Health, this becomes a high-stakes “Evolutionary Game” with clinical consequences (

Table 2. The translation: from theory to clinical reality.

Concept in IEG All (Social Trust)	Translation to IEG Health (Doctor Adoption)	The Intelligent Logic (Mechanism)
Reputation management	Algorithm transparency	Doctors assess the AI’s “reputation” based on past diagnostic accuracy and explainability (XAI).
Cooperation dynamics	Human–AI collaboration	The game moves from “Human vs. AI” to a “Human + AI” team, where the payoff is the patient’s recovery.
Trust evaluation	Reliability assessment	Using Bayesian learning, the doctor updates their trust in the AI after every successful or failed intervention.

).

In this IEG Health scenario, we model the interaction as an Adaptive Coordination Game (Figure 5):

• The Players: The Clinical Specialist (Expertise-driven) and the AI Diagnostic Agent (Data-driven).
• The Strategy: * Doctor: Choose to Adopt (Follow AI advice) or Override (Use intuition).

  ○

  AI: Provide High-Confidence or Low-Confidence output.

• The Intelligent Twist: Unlike a static game, this uses MARL. The AI “learns” the doctor’s preferences and risk tolerance, while the doctor “learns” the AI’s strengths.

Why This Translation Matters?

• Preventing “Deskilling”: If trust is too high (blind adoption), doctors may lose their critical skills.
• Reducing “Algorithm Aversion”: If trust is too low (constant override), the benefits of AI in reducing human error are lost.
• Finding the Nash Equilibrium: The goal is to find the point at which the doctor uses the AI only when the AI’s “Intelligence” truly exceeds human pattern recognition.

Figure 5. Translating social trust to Doctor-AI adoption.

Figure 5. Translating social trust to Doctor-AI adoption.

4.2. Study Implications

The study offers several significant implications for the intersection of Artificial Intelligence (AI), Evolutionary Game Theory (EGT), and clinical healthcare:

• Methodological Innovation: The research demonstrates the effectiveness of the Synthetic Near-Empty Review (SNER) framework combined with Synthetic Knowledge Synthesis (SKS). This approach allows researchers to extract structured, evidence-based knowledge from “near-empty” or sparse literature landscapes in which traditional systematic reviews would fail.
• A New Translation Framework: The study provides a structured pathway to bridge the “knowledge void” by translating broad social governance themes (e.g., Social Trust, Reputation Management) into specific clinical models (e.g., Doctor-AI Adoption, Adaptive Coordination Games). This shifts the focus from theoretical systems to specialized medical applications.
• Advancement in Strategic Logic: The paper implies a necessary paradigm shift from static evolutionary models (like fixed Replicator Dynamics) to Adaptive Learning Strategies, such as Multi-Agent Reinforcement Learning (MARL) and Bayesian updating. This shift is crucial for modelling unpredictable clinical environments and addressing human–AI interaction challenges like “algorithm aversion” and “clinical deskilling”.
• Ethical and Regulatory Governance: The study highlights that translating theoretical AI-EGT models into actual clinical decision-making requires the integration of structured ethical guidelines. It advocates for adopting frameworks like the Assessment List for Trustworthy Artificial Intelligence (ALTAI) to ensure algorithmic accountability, transparency, and societal well-being.

4.3. Highlighting IEG Health’s Unique Contributions

Unlike standard AI-based Clinical Decision Support Systems (CDSSs) that rely on static supervised learning or rigid rule-based engines, IEG Health embraces the dynamic nature of clinical care. Traditional CDSSs provide one-way recommendations, often leading to algorithm aversion if the physician disagrees. In contrast, IEG Health might model the physician and the AI as interacting agents. By anticipating user behaviour, an IEG system can adapt its strategy over time—such as communicating its confidence intervals more transparently when it detects physician scepticism—thereby fostering collaborative decision-making.

4.4. Study Limitations

The study has several limitations:

• Sparse Literature Base: The core premise of the study is that the specific field of Intelligent Evolutionary Games in Healthcare (IEG Health) is “near-empty,” with the search yielding only 16 relevant publications. Consequently, the findings heavily rely on translating concepts from broader fields (IEG All) rather than on analysing a robust body of existing clinical evidence.
• Lack of Empirical Clinical Operationalization: The study identifies that the primary bottleneck in the field is a profound scarcity of peer-reviewed publications that successfully operationalize these frameworks in real-world healthcare contexts. The proposed models (such as the adopt/override dynamic for Doctor–AI interaction) remain theoretical or demonstrative and lack empirical validation in actual clinical settings.
• Database and Search Constraints: The meta-data used for the SKS and SNER framework was harvested exclusively from the Scopus bibliographic database. This single-database approach may inadvertently exclude the relevant literature indexed in other major databases (e.g., Web of Science). However, it is worth noting that Scopus includes the PubMed bibliographic database.
• Model Simplification: To demonstrate the translation of concepts like “Social Trust,” the study employs a simplified adopt/override dynamic. While useful for demonstration, this may oversimplify the complex, multifaceted nature of clinical decision-making and legal accountability in reality.
• While MARL and Bayesian updating offer dynamic advantages, they present significant limitations in clinical settings. MARL suffers from the “curse of dimensionality”; as the number of clinical variables increases, the model requires massive amounts of exploratory data to converge, which is unsafe in patient care in which trial and error can be fatal. Additionally, MARL models often function as black boxes, making it difficult for clinicians to interpret the rationale behind a changing strategy. Similarly, Bayesian methods rely heavily on the accuracy of prior distributions; if the initial subjective probabilities are biased or flawed, the updated clinical recommendations will inherently skew, potentially compromising patient outcomes.

5. Conclusions

This research has systematically mapped the “near-empty” landscape of Intelligent Evolutionary Games in healthcare, revealing a profound disparity between theoretical AI-EGT models and their clinical application. By implementing the SNER framework, we have demonstrated that the “knowledge void” in IEG Health can be bridged through the strategic translation of established paradigms from social governance and general intelligence.

Our findings underscore that the transition to IEG Health necessitates an ontological shift: moving from passive biological modelling—such as stochastic drug resistance—to autonomous, agent-based simulations in which AI entities and human practitioners co-evolve. The case study on Doctor-AI Adoption illustrates that trust is not a static prerequisite but a dynamic Nash Equilibrium achieved through Multi-Agent Reinforcement Learning and iterative Bayesian updates.

Ultimately, the integration of IEGs into healthcare offers a robust mechanism for regulating community-based care, optimizing chronic disease management, and fostering human–AI synergy. Future research should focus on the “Adaptive Oncology” path and the deployment of Federated Learning to resolve the tension between data privacy and the need for high-dimensional strategic modelling. By treating the clinical environment as an adaptive coordination game, the healthcare sector can leverage IEGs to ensure that AI adoption is both resilient and patient-centric.

Finally, transitioning from theoretical modelling to clinical decision-making requires the integration of established ethical frameworks. Future iterations of the IEG Health engine should adopt comprehensive guidelines such as ALTAI (Assessment List for Trustworthy Artificial Intelligence) [62]. Integrating ALTAI provides the structured, qualitative metrics—including societal well-being, algorithmic accountability, and transparency—needed to ethically govern the quantitative parameters of the IEG models, ensuring social reliability alongside mathematical equilibrium.