Evolutionary Game Theory A General Review

2.1. The Effect of LLMs and AI

The escalating intricacy of Multi-Agent Systems (MASs), combined with the rise of Artificial Intelligence (AI) and Large Language Models (LLMs), has brought to light substantial gaps in our comprehension of the conduct and interfaces of diverse entities in dynamic environments. While traditional game theory approaches have been commonly employed in this domain, their effectiveness is impeded by the static and uniform nature of their models. A more dynamic and nuanced theoretical framework is essential because of the transformative impact of AI and LLMs on business and society. This framework transcends game theory by acknowledging and tackling the diverse

Yıldız Social Science Review, Vol. 10, No. 2, pp. 85−98, 2024

Pioneered research on kin selection, offering insights into cooperative behavior.

Homo-moralis is identified as an evolutionarily stable moral preference class, unlike other moral preference categories.

The coexistence of risk-dominant and return-dominant strategies can lead to coordination issues (Straub,1995). Riskdominant strategies may act as stochastically stable equilibria (Staugidl, 2011).

Adaptive learning processes with monotonic error terms lead to Kalai-Smorodinsky bargaining solutions (Young, 1998). Introduced coalitional stochastic stability, showing how groups deviate from conventions to reach Nash equilibria (Newton, 2012).

Provided empirical evidence supporting EGT principles in small-scale cultural contexts (Henrich et al, 2015). Observational learning across groups enhances evolutionary convergence to optimal strategies (Fan & Hui, 2020).

Leaders enhance communication success, and positive feedback among players fosters cooperation, reducing coordination failures (Gou & Deng, 2021).

Dong et al. (2020), Shleifer (2000), Evstigneev et al. (2006)

Stock prices influenced by access to market information and reduced speculation (Dong et al, 2020). Random choice strategies may yield higher profits than rational strategies under certain conditions (Shleifer, 2000). Evolutionarily stable stock markets rely on expected relative dividends for valuation (Evstigneev et al at., 2006).

Dominating strategies improve financial and insurance market stability (Hens & Hoppe, 2005). Reducing uncertainty in yield structures facilitates payoff-dominant equilibria (Mielke & Steudle, 2018).

Cooperation can emerge in heterogeneous populations with diverse cultural preferences (Bilancini et al., 2018).

Stochastic equilibrium choices in coordination games converge to Nash solutions (Naidu et al., 2010). Coordination failures in populations with highly diverse parameters, due to topological issues (Roca et al., 2010).

Incomplete markets may disrupt Pareto optimal allocations, challenging the market selection hypothesis (Blume & Easley, 2006).

EGT-based models simulated COVID-19 dynamics, aiding in evaluating interventions like social distancing (Bukkuri & Brown, 2021).

EGT models explored the strategic interaction between students and educational institutions in AI-driven environments, revealing challenges in supervision and adaptation (Newton, 2018).

Clarified philosophical principles underlying game theory (LaCasse & Ross, 1994). Linked game theory assumptions to microeconomic and macroeconomic principles (Ross & LaCasse, 1995).

interactions (economic transactions, social relationships, information exchange) and the disparities in risk aversion, social preferences, and learning capabilities among entities. The recent progress in evolutionary game theory has emphasized the significant role of large language models (LLMs) in improving agent-based modeling and simulation. Traditional agent-based models rely on simple rules or limited-intelligence neural models to imitate the behaviors of agents within a system. However, integrating LLMs, such as GPT-3.5 and GPT-4, into these models has brought about a new level of complexity and intelligence, making it possible to create more sophisticated and human-like agents (Gao et al., 2023). Large language models (LLMs) have enhanced the simulation of complex systems in various domains, including cyber, physical, social, and hybrid environments. They have been used to simulate economic activities, social interactions, and market behaviors more accurately. Researchers can create more realistic simulation environments by using LLMs and developing agents capable of advanced strategic reasoning and adaptive behavior. Although LLMs generally display rationality to some extent, they often do not reach Nash equilibrium. Among the tested LLMs, GPT-4 has demonstrated superior strategic reasoning abilities and quicker convergence to Nash equilibrium compared to other models such as GPT-3.5. In their 2023 study, Zarza and colleagues emphasized the significant potential of leveraging large language models (LLMs) to facilitate cooperation, enhance social welfare, and promote resilient strategies within multi-agent environments. The proposed Evolutionary Computation (EC) framework provides valuable insights into the intricate interplay between strategic decision making, adaptive learning, and guidance informed by LLMs in complex and ever-changing systems. This research addresses the challenges of modeling Multi-Agent Systems (MASs) and sets the stage for further exploration and advancement in this rapidly evolving field. In their research, Suzuki and Arita (2024) seek to illustrate the capacity of large language models (LLMs) in advancing the study of human behavioral evolution through the application of evolutionary game theory. They propose an evolutionary model using LLMs loaded with high-level psychological and cognitive character descriptions to simulate human behavior choices in game-theoretical scenarios. The initial phase involves introducing an evolutionary model of personality traits associated with cooperative behavior, utilizing a comprehensive language model. In this model, linguistic descriptions of personality traits related to cooperative behavior are treated as genes, and the deterministic strategies derived from the LLMs. The population evolves through selection based on the average payoff and mutation of genes by instructing the LLM to modify the parent gene towards cooperative or selfish traits slightly. Through experiments and analyses, they demonstrate that such a model can exhibit the evolution

Yıldız Social Science Review, Vol. 10, No. 2, pp. 85−98, 2024

of cooperative behavior based on diverse and higher-order representations of personality traits. The researchers noted multiple occurrences of cooperative and selfish personality traits arising from variations in the manifestation of personality traits. The terms that surfaced in the modified genes mirrored the behavioral inclinations of the linked personalities in semantic terms, impacting individual behavior and evolutionary dynamics. In their 2023 paper, You et al. emphasized the potential benefits of appropriately applying AI in education while highlighting the potential for harm through its misuse. The authors used EGT to explore AI’s potential in the future of education. By studying the behavior of two agents, the school and the students, they aimed to identify strategies to enhance the effectiveness of the education model in the AI era. They developed a stable evolutionary strategy for the school and the students across various scenarios. In their study, the authors performed a detailed numerical analysis to investigate the influence of various essential factors on developing a stable strategy. The analysis findings indicated that it would benefit educational institutions to establish robust supervision mechanisms to effectively manage AI integration in education. Moreover, the results highlighted the importance of students proactively embracing and interacting with AI technology to maximize its potential benefits in the educational context. Han et al. (2020) developed an AI model that simulated different hypothetical scenarios for AI races to determine which AI races should be prioritized for regulatory oversight. The team’s model had to account for the complex choices involved when there was no predetermined path, so it integrated various concepts from biology and mathematics, such as EGT. Researchers conducted multiple simulations, tweaking the variables to observe the long-term outcomes. They used a model with virtual agents as competitors in simulated AI races, each with randomly assigned behaviors reflecting real-world situations. Some agents were cautious while others took more risks, leading to AI products that were inadequately tested or vulnerable to hacking and data leaks. The researchers found that the most crucial factor was the length of the AI race, which determined the time taken to produce a functional AI product. Recent developments in evolutionary game theory have emphasized the important contribution of large language models (LLMs) in improving agent-based modeling and simulation, as demonstrated earlier.

2.2. Treatment of Cancer

A very recent application of EGT is the modeling of cancer treatment. For many years, the conventional approach to treating cancer involved administering the maximum tolerated dose of a medication to target a high number of cancer cells with minimal side effects.However, the cells that make up a tumor are diverse. Some of these cells develop mutations, or changes in their genetic material, by chance, making them resistant to a drug. These resistant cells can

Yıldız Social Science Review, Vol. 10, No. 2, pp. 85−98, 2024

then grow and repopulate the tumor, leading to therapeutic resistance that renders the drug ineffective. In such cases, physicians typically transition to an alternative medication that addresses a distinct characteristic of the cancer cells. This process continues until a treatment can effectively control the cancer or until no more drugs are available. If no more drugs are available, patients are then provided with hospice care to ensure their last days are as comfortable as possible. To improve long-term patient outcomes, cancer researchers seek answers to two fundamental questions: What drives tumor growth, and how do tumors develop resistance? An ecological and evolutionary approach to studying cancer, which examines how the body’s environment influences and is influenced by evolving cancer cells over time, offers valuable insights into these queries (Bukkuri, 2021). Understanding the impact of selective pressures, external factors influencing an organism’s survival. Evolutionary game theory provides valuable insights into the dynamics of interactions between cancer cells and their environment, helping researchers comprehend the effects of these selective pressures. Hence, EGT offers valuable insights for researchers and oncologists in forecasting cancer response to various treatments and potentially influencing the evolutionary path of cancer. This knowledge holds the potential to optimize patient outcomes. Substantial strides have been taken in cancer treatment, yet there remains a substantial challenge in rendering all cancer types treatable conditions. One encouraging strategy for reaching this objective involves harnessing evolutionary mechanisms to consistently exert selective pressure on cancer cells. Some important articles are also in this field. Orlando et al. (2016) discussed cancer treatment as a strategic game, where oncologists select a therapy and tumors adapt. They suggested that oncologists can gain an advantage by choosing treatment strategies that anticipate tumor adaptations. The study examined the potential benefits of exploiting evolutionary tradeoffs in tumor adaptations to therapy. The authors analyzed a mathematical model in which cancer cells faced tradeoffs in allocating resistance to two drugs. The tumor “chooses” its strategy through natural selection, while the oncologist selects her strategy by solving a control problem. The study found that when tumor cells perform best by investing resources to maximize their response to one drug, the optimal therapy is a consistent delivery of both drugs simultaneously. The study also emphasized that drug interactions can significantly alter these results. In conclusion, the authors stressed the importance of understanding both evolutionary tradeoffs and drug interactions in planning optimal chemotherapy schedules for individual patients. Czako et al. (2017) provided a model aiming to reduce medical expenses during cancer treatment by creating personalized administration plans with the help of control

engineerin. Kaznatcheev and colleagues (2017) introduced the double goods game to explore the dynamics of certain traits using evolutionary game theory. They used glycolytic acid production to represent a public good for all tumor cells, and vascular endothelial growth factor production for oxygen from vascularization as a club good benefiting non-glycolytic tumor cells. This resulted in three viable phenotypic strategies: glycolytic, angiogenic, and aerobic non-angiogenic. Authors have categorized the dynamics into three different types: (1) fully glycolytic, (2) fully angiogenic, and (3) polyclonal in all three cell types. The third type allows for dynamic heterogeneity even with linear goods, which was not possible in prior public good models that only considered glycolysis or growth factor production in isolation. In 2018, West et al. constructed a mathematical model centered on the replicator dynamics of three interacting cell populations: cooperative (healthy), defecting (sensitive), and resistant (defecting) cells.This model mirrored prostate-specific antigen measurement data from three clinical trials for metastatic castration-resistant prostate cancer patients treated with different medications. The study found that continuous maximum tolerated dose schedules initially reduced the sensitive cell population, decreasing tumor burden. However, this allowed the resistant cells to thrive and regrow the tumor in a resistant form. The researchers used this evolutionary model to evaluate responses to conventional therapeutic strategies and design adaptive strategies, which were found to be robust to small timing changes and extended the simulated time to relapse from continuous therapy administration. Stankova et al. (2019) studied cancer treatment as a game involving the physician’s therapy and the cancer cells’ resistance strategies. This game has two important differences: (1) Only the physician can make rational decisions. Cancer cells can only adapt to current conditions; they cannot predict or evolve adaptations for treatments that the physician has not yet applied. (2) It has a unique leader-follower (or “Stackelberg”) dynamic; the oncologist takes the lead by making the first move, and the cancer cells then respond and adapt to the therapy. Current treatment protocols for metastatic cancer usually do not take advantage of these differences. Through the repetitive administration of the same drug(s) until disease progression, the physician employs a static approach while the cancer cells continuously evolve to develop effective adaptive responses. Additionally, by altering treatment only upon tumor progression, the physician effectively allows the cancer cells to dictate the course of treatment, leading to almost certain treatment failure. Without fundamental changes in strategy, conventional cancer therapy typically leads to “Nash solutions,” where no individual change in treatment can enhance the outcome. Bayer et al. (2021) utilized game theory tools to create a Markov chain model for cancer treatment. The model represents cancer therapy as a patient’s Markov Decision

Problem, aiming to maximize the patient’s discounted expected quality of life years. Patients can choose the number of treatment rounds based on the progression of the disease and their own preferences. The authors developed an analytical decision tool that allows patients to select their preferred treatment strategy. In another model, patients can also decide on the timing of treatment rounds. By delaying therapy, patients can postpone the side effects while forgoing the benefits for a period. The authors obtained an analytical tool that enables numerical approximations of the optimal delay times. Wölf et al. (2022) described cancer as an evolutionary game, consisting of cells of different types undergoing frequency-dependent selection. In the article, it was provided an overview of previous studies that used evolutionary game theory to model cancer and improve its treatment. Some of these game-theoretic models suggest potential strategies to predict and guide cancer’s eco-evolutionary dynamics towards more favorable states for the patient through evolutionary therapies. These therapies promise to improve patient survival and reduce drug toxicity, as evidenced by recent studies and clinical trials. During the discussion, researchers explored the practical implications of existing game-theoretic models in understanding and treating cancer. They also highlighted the need for advancements in cancer biology to fully leverage the potential of these models. Ultimately, the application of evolutionary game theory in understanding tumor behavior was shown to have important medical implications, bridging the gap between real-world observations and mathematical modeling. The article concluded by examining evolutionary treatment approaches and their potential to extend life expectancy beyond that of traditional therapy. It suggested that evolutionary therapy could be more effective when dealing with pre-existing or therapy-induced resistant types. The main hurdle in integrating EGT treatment strategies in clinical settings is the challenge of precisely estimating tumor composition. Currently, this is only viable for specific cancer types. This treatment methodology is particularly effective for cancers where such information is obtainable.

2.3. Post Pandemic Educational Models

COVID-19 represents the much-anticipated global pandemic with the potential to severely impact both societies and economies, leading to widespread disruption. It’s crucial to understand how COVID-19 unfolded as a health and development crisis, think about potential transformations after the pandemic, and reconsider development more broadly. Based on over a decade of research on epidemics, it was argued that analyzing the origins, progression, and effects of the COVID-19 pandemic requires addressing both structural political-economic conditions and less orderly, more unpredictable processes reflecting complexity, uncertainty, contingency, and context-specificity (Leach et al., 2021).

Yıldız Social Science Review, Vol. 10, No. 2, pp. 85−98, 2024

This duality in the conditions and processes of pandemic emergence, progression, and impact provides a way to understand three key challenge areas. The first challenge is related to how scientific advice and evidence are used in policy when conditions are firmly controlled by established power relations yet uncertain. The second involves how economies operate, as the COVID-19 crisis has revealed the limitations of conventional economic growth models. The third challenge concerns how new forms of politics can shape citizen-state relations in dealing with a pandemic, such as those centered around mutual solidarity and care and education. Due to the pandemic crisis, educational institutions have been compelled to adopt new pedagogical approaches. This adaptation is imperative to mitigate the impact of the epidemic on the educational sector. While certain students may demonstrate proficiency in acclimating to the novel learning environment, the level of adjustment will undoubtedly vary among students, given that the majority are encountering this educational format for the first time (Heikkinen, 2023). Hence a new educational system has to be organized and EGT can help to researchers. There is an important paper conducted by Accinelli et al. (2021). In this paper, researchers used tools from evolutionary game theory to analyze the evolution of the behaviors of professors, students, and schools in relation to the use of information and communication technologies (ICTs) in the teaching-learning process in the post-pandemic era. For this purpose, replicator dynamics were used, and it was shown that the solutions to this dynamical system could indicate the evolution that the teaching-learning processes will follow over time in the post-pandemic world. In a recent study, Li and Wang (2022) utilized evolutionary game theory to investigate the management of primary and secondary school students’ online learning during the COVID-19 pandemic. The study also aimed to assess the impact of stakeholders’ behavioral choices on managing students’ online learning. The researchers created two game models, “schools-students” and “schools-students-parents,” and analyzed the influence of behavioral interactions on game equilibrium in these scenarios. Furthermore, researchers employed MATLAB 2018 software to conduct numerical simulations for their analysis. The study revealed significant differences in strategy choices of students who play games and their management of online learning. It found that the main factor influencing students’ strategic choices is the benefits they receive from learning. While involving parents can positively influence students’ game strategy selection towards cooperation, there is a limit to their involvement. The study also found that the use of punishment or reward by schools does not significantly affect students’ management of online learning. However, punishments and rewards from parents have a more significant impact on promoting students’ strategic choices towards cooperation compared to those implemented by schools.

Yıldız Social Science Review, Vol. 10, No. 2, pp. 85−98, 2024

Muniz et al. (2023) utilized evolutionary game theory tools to analyze the evolution of the behaviors of professors, students, and schools in embracing the use of information and communication technologies (ICTs) for the teaching-learning process in the post-pandemic era. The replicator dynamics employed to illustrate potential solutions to this dynamic system, shedding light on the potential evolution of teaching-learning processes in the post-pandemic world over time. They have implemented a teaching and learning model where professors, students, and authorities have the option to choose between traditional education, similar to the pre-pandemic era, or educational methods that integrate the use of Information and Communication Technologies (ICTs), capitalizing on the skills acquired during the COVID-19 pandemic. In game theoretical terms, this choice is seen as a three-player game, with each participant having two different strategies: supporting or opposing the use of ICTs in the teaching and learning process. The authors demonstrated that the advancement of pedagogical methods is contingent upon the preferences of all stakeholders. The interplay among educators, students, and educational administrators shapes the progression of knowledge transmission and assimilation techniques. It is evident that embracing digital methodologies will necessitate commitment from educational authorities, teachers, and students. In addition, this work has observed that the future direction of education in the post-pandemic era may be influenced by the decisions made by educational authorities regarding the use of ICTs. The choices made within schools will largely depend on the availability of economic resources and technological infrastructure. If large investments in ICTs are not deemed to be cost-effective, educational authorities may choose to revert to traditional offline education models. On the other hand, moderate investments in ICTs could be sufficient for professors to decide to undergo training and then facilitate the appropriate use of these available technologies by students. While the decisions of educational authorities greatly influence the direction of education, they are not the sole determining factors for the different scenarios to unfold, as all decisions are intertwined with other participants’ preferences in the educational system. Finally, researchers showed that the emergence of teaching cycles is possible. Incentives for the development of digital techniques must be maintained for a while because if they are eliminated, even when the use of these techniques is predominant, the process can be reversed and return to the initial situation. In a significant study, Liu et al. (2023) developed a tripartite evolutionary game model involving schools, enterprises, and government to analyze the strategies of each party and the factors influencing them. The results revealed that the rewards for positive cooperation from sources other than the government mainly influence the stability of

schools’ and enterprises’ strategies. On the other hand, the stability of the government’s strategy is mainly influenced by the benefits of a positive cooperation strategy in the scenario where schools cooperate with enterprises. The study concludes with suggestions to enhance sustainable cooperation among schools, enterprises, and government. The aforementioned research highlights the widespread use of evolutionary game theory methodologies in evaluating the efficacy of recently developed educational frameworks, especially in light of the global pandemic. The significant input of evolutionary game theory enables the modeling of both educational systems and the diverse range of individuals within these systems.

2.4. Network Theory

Network theory in evolutionary game theory (EGT) emphasizes the structure of interactions among agents within a network, where the arrangement of connections impacts the development of strategies. Networked systems frequently display intricate patterns, as the behavior of one agent can influence others, and these interactions evolve over time. In social networks, individuals, or nodes, are interconnected, and their interactions determine the dynamics of information or resource flow. Evolutionary Game Theory (EGT) models these interactions by examining the strategies individuals adopt in response to the behaviors of their peers. EGT has been utilized to analyze competition and cooperation within economic systems, including trade and market dynamics. By representing businesses or countries as players in a network, evolutionary models can elucidate how competitive or cooperative strategies spread. For example, the evolution of market behaviors—such as price-setting or competitive bidding—can be explored within a network of economic agents interacting in a competitive landscape (Vega-Redondo, 2007). In the view of infectious diseases, the integration of network theory with evolutionary game theory (EGT) models provides a framework to analyze the dynamics of disease transmission within populations. The interactions among individuals, conceptualized as agents on a network, critically influence the epidemiological spread of infections. Under specific conditions, cooperation mechanisms, such as vaccination, can emerge as stable strategies, promoting overall population health (Bauch & Bhattacharyya, 2012).

2.5. Cooperative Dynamics in Non-Biological Systems

Cooperation in non-biological systems, analyzed via Evolutionary Game Theory (EGT), explores the mechanisms through which agents engage in collaborative behaviors for collective gain. Within the framework of EGT, cooperative strategies emerge when the advantages of mutual collaboration surpass the associated costs. This dynamic is frequently influenced by the topology of the underlying network, which governs the interactions and payoffs between agents.

One significant application of evolutionary game theory (EGT) in cooperative dynamics is found in the view of global environmental agreements. By modeling the evolutionary dynamics of cooperation among nations facing climate change, we can elucidate how self-interested entities (or countries) might engage in cooperative behavior to address shared challenges, such as the reduction of carbon emissions. The core issue is ensuring that all parties commit to cooperation rather than resorting to defection, which can undermine collective efforts. Stable cooperative outcomes can be achieved through mechanisms that promote repeated interactions and robust enforcement strategies, as explored in Ostrom’s work (1990). In public goods games, participants are tasked with contributing to a shared resource that serves the collective. Yet, each individual is tempted to engage in “free riding,” benefiting from the contributions of others without contributing themselves. Evolutionary game theory (EGT) models, particularly those employing the framework of evolutionary stable strategies (ESS), provide valuable insights into the dynamics through which cooperation can manifest under such conditions. The study of agents in networks interacting within public goods games is particularly compelling, as the characteristics of local interactions and the underlying network topology critically shape the dynamics of cooperative behavior (Bertschinger et al., 2011).

2.6. Challenges

The domain of EGT is rife with complexities and debates as scholars aim to construct precise models of realworld situations. One of the main difficulties lies in the intricate and varied behavioral patterns, preferences, and decision-making mechanisms exhibited by real individuals. Conventional rule-based techniques in agent-based modeling have constraints in encompassing the full spectrum of this diversity, requiring a more holistic method. Ensuring a balance between model simplicity and accurate agent representation is critical, as oversimplification may overlook important aspects of agent heterogeneity. Therefore, there is a demand for more advanced methodologies that can effectively capture the diverse behaviors and decision-making processes of real-world individuals (Gao et al.,2023). The current techniques for simulation integration necessitate the incorporation of multiple dimensions of simulation. These dimensions involve the description of systems under known conditions, the explanation of emergent patterns, the prediction of future states, and the exploration of hypothetical scenarios. Rule-based methods are effective in solving descriptive problems, while symbolic or stochastic methods elucidate underlying mechanisms. Machine learning models are adept at predictive tasks, as they can identify concealed patterns in data; however, they are associated with lower interpretability. This disparity poses a significant challenge in the development of models

Yıldız Social Science Review, Vol. 10, No. 2, pp. 85−98, 2024

that harmoniously combine behavioral accuracy, interpretable mechanisms, adaptability, and reliability in evolving environments. One another important consideration is the limitations of current models in supporting integrative simulations for real-world problems. For example, agents based on Large Language Models (LLMs) can capture complex internal characteristics and specialized traits using techniques such as prompting, in-context learning, or fine-tuning. However, these methods still have limitations when simulating realistic processes or systems, which highlights the need for further development in this rapidly evolving field. Integrating artificial intelligence (AI) with evolutionary game theory (EGT) also faces several challenges. Biological and social systems in the real world are inherently dynamic and stochastic, which presents significant challenges for AI algorithms that utilize evolutionary game theory (EGT) for predictive modeling. While simplifications are often necessary for computational efficiency, such reductions can constrain the applicability and fidelity of the models (Bukkuri & Brown, 2021; Blume & Easley, 2006). One another significant challenge is the limitations and representation of data. AI systems necessitate comprehensive, high-quality datasets to inform EGT (Evidence Generation and Translation) models effectively. Insufficiently populated or biased datasets can severely undermine the accuracy of predictions, especially in tailored applications such as cancer therapeutics (Dong et al., 2020). Interpretability and clinical integration present significant challenges in the field of artificial intelligence. To align AI-generated electrocardiogram (EGT) predictions with clinical protocols, it is essential for these predictions to be interpretable and usable for decision-makers. Complex models can produce outputs that are difficult to understand and act upon (Newton, 2012; Gou & Deng, 2021). Moreover, ethical concerns are paramount. The use of patient-specific data in AI-EGT applications raises issues related to privacy, informed consent, and bias, particularly in healthcare settings. One last significant challenge associated with AI systems integrated with Evolutionary Game Theory (EGT) is the substantial computational expense, especially when scaling these systems to accommodate the adaptive and interactive characteristics of agents (Fan & Hui, 2020). The current EGT models have also critical limitations when it comes to supporting integrative simulations for realworld problems due to their inability to effectively address the complexities and multifaceted nature of these problems. While LLM agents are capable of capturing complex internal characteristics and specialized traits through techniques like prompting, in-context learning, or fine-tuning, they still struggle when it comes to simulating realistic processes or systems. This indicates a clear need for further advancement in this rapidly evolving domain. The complexity of these challenges highlights the crucial importance of the audience’s work in this field. Evolutionary explanations of social norms use models that explicitly demonstrate value

Yıldız Social Science Review, Vol. 10, No. 2, pp. 85−98, 2024

commitments and derive other normative commitments from the actions of self-interested agents with limited rationality. This approach offers clarity and reveals the logistical assumptions necessary for manageability, which can affect the model’s results if not handled carefully. Assumptions such as random mating, an infinitely large population size, and nonoverlapping generations are common in many population genetic models and are used to simplify the complexity without fundamentally altering the outcomes. One last controversial point of contention involves the utilization of simplified and abstract models to directly compare them with the EGT problem, instead of accurately representing real-life scenarios. The use of utility-based proportional reproduction methods instead of more realistic reproduction scenarios is a clear example of this approach, emphasizing the delicate balance between the simplicity of the model and its real-world accuracy.