Artificial intelligence, specifically deep learning, numerical optimization, and scalable computing, can play a key and innovative role in many areas of knowledge, including epidemiology, especially in infectious diseases. This is due to its potential to enable the accurate prediction of transmission patterns and the most effective measures to control outbreaks, ultimately promoting collective well-being. The objective is to highlight the advantages of using artificial intelligence to predict the epidemiological behavior of infectious diseases during epidemic outbreaks. This is a narrative review of high-quality scientific articles. Based on the analysis of the data found, the relevant aspects were categorized into the following sections: artificial intelligence in epidemic infectious diseases; disease propagation; modeling of mechanisms; detection of emerging and re-emerging pathogens; data collection and structuring; and information management for the community during epidemics, with the aim of facilitating reading and comprehension. It is concluded that artificial intelligence is a valuable ally in the fight against infectious diseases.

The intelligent behavior of machines and computers based on logic and deep learning (the ability to learn patterns and relationships from data without programming for each condition), known as artificial intelligence (AI), includes methods of machine learning, probability, numerical optimization, and scalable computing (in short, AI involves a machine's ability to learn from experience and adapt to new information in tasks similar to human capabilities). AI is undoubtedly capable of transforming many aspects of all sciences, including epidemiology, a fundamental discipline of public health. Epidemiology, which focuses on population health, combines quantitative and qualitative designs and techniques to determine the true and probable causes of diseases [1-5]. 

The usefulness of AI in epidemiology encompasses aspects such as the study of the determinants of health, the prediction of the risk of infection and disease occurrence, the diagnosis of pathologies, and support for making sound, specific, and equitable health decisions aimed at the population as a whole, based on representative data. However, during disease outbreaks, such information is scarce, imprecise, and geographically biased, hindering comprehensive epidemiological research [5-7].

AI allows us to overcome the Big Data paradox, as it is not solely about collecting large volumes of data (scarce in epidemics, but AI can be pre-trained with related data from similar infections to identify emerging or re-emerging threats), but also about the quality and representativeness of that data. These are key to improving the cost-effectiveness of surveillance, mitigating the unequal representation of data, and ensuring optimal health decision-making. AI improves the ability to infer fundamental epidemiological parameters for accurately predicting the epidemiological behavior of emerging and re-emerging etiological agents [4, 5, 7, 8].

The objective of this paper is to precisely and analytically demonstrate the aspects that make AI useful in the field of epidemiological knowledge, specifically during outbreaks of emerging and re-emerging infectious diseases. AI contributes to identifying intervention targets to improve population health. In this context, the following topics or aspects were considered: artificial intelligence in epidemic infectious diseases; Disease propagation. Modeling of mechanisms; detection of emerging and re-emerging pathogens; data collection and structuring; and information management for the community during epidemics.

The objective of this paper was achieved through a literature review on the aspects that link AI with infectious diseases in the context of epidemic outbreaks. To this end, specialized health science databases such as PubMed and Medline were reviewed, using keywords specifically related to the research topic. This narrative review included full-text articles published up to 2025 in English, Spanish, and Portuguese.

Artificial Intelligence In Epidemic Infectious Diseases

The foundation of public health policies and programs at regional and global scale in the case of infectious diseases studied by epidemiology in relation to their transmission and strategies to prevent, control and limit their spread, are currently focused on mathematical, computational and statistical models within the framework of AI, despite the difficulty in obtaining large-scale, standardized and representative data, since new AI approaches work excellently with scarce or limited data, among which the fine-tuning or transfer learning of large pre-trained networks and self-supervised learning stand out [5, 9-11].

In this sense, AI improves the collection and integration of heterogeneous data sources, enabling their essential inclusion in decision-making for the socio-health benefit of the population. This is because it allows for a better understanding of the severity and epidemic potential of infectious agents, going beyond observational data (case-control studies, cohorts, or household surveys). This involves estimating: case fatality rates, serial and generational intervals, transmissibility, epidemic growth, identification of infection networks, and heterogeneity of transmission [5, 12, 13]. Therefore, AI overcomes the limitations of imperfect or unrepresentative observation of the epidemic, focusing on the actual chain of transmission, the locations where it occurs, the incubation period, and the intensity of transmission. Furthermore, underreporting, censorship, and unequal data reporting can cease to be problems with AI using gradient-based deep learning techniques in approximate Bayesian inference, similar to variational inference with normalizing flows, particularly useful when speed and accuracy are required [5, 14, 15]. 

AI substantially improves the realism of complex epidemic scenarios and the speed of analysis (from weeks to hours) because it facilitates, among other things, establishing links between the heterogeneity of individual transmission and its population projection, and the complete generation of the transmission process. This substantially surpasses analyses based on routinely collected aggregate data (case counts, hospitalizations, or deaths) from the mechanistic paradigm or model of disease transmission, which are marked by uncertainty and high computational costs due to the complexity of the numerical methods required and the high dimensionality in constructing counterfactual scenarios [5, 16-20]. Based on the number of current cases and the forecast of future cases, decisions regarding health interventions are made using statistical time series techniques for quantitative trends (demographic, clinical, geospatial, and genomic information). Currently, with AI, supported by machine learning models (which mimic human learning with algorithms that use data to learn and improve their performance) and deep learning (a subset of machine learning consisting of neural networks for human-like data processing), it is possible to determine infectious individuals at a given time and the future behavior of cases. By eliminating the biases that characterize the reporting, testing, and sampling of traditional epidemiological surveillance, these are powerful tools for understanding time series epidemiological surveillance data [5, 21, 22].

Spread of diseases. Modeling mechanisms

Epidemiological modeling of infectious diseases draws on standardized information and theoretical postulates to describe disease transmission in close relation to population dynamics. It is based on methodologies that employ computer simulations, statistical analyses, and mathematical procedures to determine the effects on public health. However, it requires advanced mathematical knowledge and the assumption of ideal circumstances to guarantee the production of reliable and widely applicable theoretical information. This is why AI is particularly useful in the field of epidemiology, as it provides advanced methods for analyzing observed and recorded data [4, 23].

 Infectious diseases are traditionally modeled as simplified, definable processes based on mechanistic models (mathematical equations: susceptible-infected-recovered or models based on individuals/agents) capable of generating valuable information on transmission and the interventions to be used for control. However, the lack of analytical components hinders efficient inference, a circumstance that AI improves through innovative approaches that use deep neural networks for the analysis of demographic information, human and animal movements, climate change, and environmental elements [4, 5, 24-31].

With new AI neural network architectures, variational Bayesian inference is well-suited to models with highly phenotypically variable and dynamic agents, as it allows for rapid implementation. AI thus becomes a valid and necessary alternative to the computational intractability generated by models with these infectious agents, because it provides the necessary details to relate individual behavior to population dynamics. Likewise, modeling plausible epidemiological scenarios (revealing the transmission dynamics of the infectious agent and population behavior over time and its effects on public health) with AI models has the potential to achieve high predictive accuracy and to learn more complex hidden mechanisms [4, 5, 15, 32, 33]. 

Similarly, in the detailed prediction of the transmission dynamics of communicable diseases, graphical neural networks operating with discrete structured AI data play a significant role, because they include: contacts in individual and population spread, the phylogenetic tree of pathogens, social and information networks in the realm of health-related behavior, and the social influence on the spread of information and misinformation. Graphical neural networks allow for the differentiation of all model components to facilitate joint inference. Finally, graph-based models enable the transfer of knowledge from data-rich environments to those with data scarcity [5, 15, 34].

Detection Of Emerging And Re-Emerging Pathogens

Emerging and re-emerging etiological agents pose a continuous threat to public health and a challenge to public health due to alterations in the patterns of zoonotic diseases that influence their transmission patterns, even linking them to humans. This underscores the importance of epidemiological surveillance, but surveillance that goes beyond clinical or sentinel surveillance; that is, surveillance that includes laboratory confirmation and allows for predictions and early warnings of epidemic outbreaks of infectious diseases (new or nearly controlled), especially useful when the diagnostic method is complicated or unavailable [35].

Traditional epidemiological surveillance of infection outbreaks relies on experimental genomic research to investigate circulating genetic diversity and determine whether pathogens are new, re-emerging, or simply variants. However, determining the epidemiological phenotype through this method requires significant time and financial resources. This is why AI, which uses genetic sequences as input to generate information about pathogen proteins, is relevant. AI allows for highly accurate prediction of the structural and non-structural proteins of pathogens [5, 36, 37]. In this regard, cross-prediction approaches have great potential for inferring the protein structure of epidemic-causing pathogens, as well as for phylogenetic inference. This facilitates the characterization of the infectious process in aspects such as the location, timing, and cause of the etiological agent's emergence, as well as the transmission process and host specificity. AI is particularly useful in determining escape variants by inferring the evolutionary trajectory of infectious agents [5, 38-40]. AI prediction of phenotypic variations in circulating pathogens based on genomic sequences allows for the development of attack strategies such as drugs, monoclonal antibodies, and vaccines. AI plays a key role in determining the phylogeny of etiological agents based on Bayesian variational inference supported by probabilistic programming; however, the analysis must not overlook intervening environmental and human factors such as human behavior, cross-immunity, and climatic conditions [5, 41, 42].

Data Collection And Structuring

In the epidemiology of infectious diseases, the federated AI learning approach is useful, allowing for the collaborative training of models on decentralized servers without necessarily sharing local data (thus guaranteeing data privacy and security), harmonizing disparate sources or formats into a standard and organized model [4, 43-45].

In AI, data collection and structuring in epidemics adopts interoperability standards as a complementary layer over existing data sources within the framework of joint research, also for the distribution of data to various operational (direct healthcare centers) and administrative (designers of social and healthcare programs) sources [4]. Furthermore, the impact of confounding variables can be reduced with AI in multivariate epidemiological models for the analysis of observational data, achieving an efficiency approaching that of controlled experimental designs used in clinical trials. This involves the virtualization or reconstruction of experiments, a process still in the theoretical stage but highly feasible with AI [4].

 Similarly, AI with machine learning models is well-suited for studying the social determinants of health, which involve countless covariates with complex dimensions, in what is known as social epidemiology. It is also useful in life course epidemiology, which conducts longitudinal analyses of dimensions that affect health throughout life [15].  In the case of social epidemiology, the focus is on studying the determinants of health, which often involve a wide range of high-dimensional covariates related to socioeconomic status, such as education, employment, and neighborhood characteristics. In life course epidemiology, researchers analyze high-dimensional longitudinal data to understand how various exposures and factors affect health outcomes throughout life. In these contexts, machine learning methods that do not require specifying a functional form of the relationship between variables could reach their full potential, reducing bias arising from model misspecification [31].

Managing Community Information During Epidemics

During epidemics, misinformation is common, deceiving and confusing the population and, of course, hindering the implementation of control measures. Therefore, it is important to combat misinformation with tools other than the statistical approaches regularly used, such as the analysis and synthesis of accurate information on infectious diseases originating in the digital realm through AI [5]. 

AI provides real-time information on the transmission chain of etiological agents and on the prevention and control measures for the diseases they cause, as well as on the population's attitude during the epidemic and on the health measures implemented by the relevant authorities, both through feedback mechanisms from the public. This information will allow for the redesign, if necessary (due to non-adoption or non-compliance), of the public health policies that have been ordered [5, 46]. The use of AI to transmit validated and real-time information is promising, provided that data, algorithmic (non-introduction of variables) and human biases are reduced, generally resulting from the exclusion of population sectors such as women and ethnic groups. Therefore, the important thing is to avoid the so-called “AI hallucinations”, that is, the propensity to generate false information derived from AI tools, through an intersectional approach (configuration of experiences of those who are at different intersections of identities or social positions) that guarantees understanding and avoids the negative impact of analyses done with AI [5, 47-50].

In current epidemiology, the small scale of observational methods and the deficient characteristics of data collection (number of cases, hospitalizations, and deaths) result in imperfect or unrepresentative observation of the actual behavior of infectious disease outbreaks. These outbreaks form the basis for essential public health decisions for transmission control, which unfortunately often lead to ineffective and delayed responses. However, these problems can be successfully avoided with current immediate prediction approaches that restrict fitted functions to plausible forms, allowing for the precise quantification of uncertainty and a robust estimation of the epidemiological situation. Deep neural networks allow for the most accurate simulation of transmission (based on variational inference) of all its events because they facilitate the correct adjustment to complex data used in generating specific interventions for the complicated control of infectious diseases. AI approaches (Bayerian inference) then simplify inference in models based on very large agents and allow the transfer of knowledge from scenarios with abundant data to those with scarce data.

AI based on variational Bayesian inference, driven by probabilistic programming from genomic sequences for the prediction of new and complex phenotypes, allows for anticipating escape variants and the development of specific treatments, while reducing the amount of experimental work. However, the complexity of the etiological agent/host/environment interaction must be considered to avoid erroneous predictions derived from AI alone. The federated learning approach is useful in the collection and structuring of epidemiological data. Disinformation in public health during epidemics can be combated with AI tools, provided that biases originating from the exclusion of minority population groups are avoided; the important thing is to avoid misunderstandings, that is, AI hallucinations.

Conflict Of Interests

The authors have no conflict of interest to declare. The authors declared that this study has received no financial support.