Defining artificial intelligence as a set of algorithms, techniques, and computational systems capable of simulating human cognitive processes, such as reasoning, continuous learning, perception, and decision-making [6],[7], is an acceptable concept of AI; indeed, this is what differentiates it from conventional systems: the capacity for adaptive learning from data is what makes AI a powerful tool for solving complex, dynamic problems considered high-dimensional for humans [8],[9].

It is necessary to understand that digital transformation is the process by which organizations and institutions integrate new digital technologies, seeking to implement them in all their areas of operation; this leads to a fundamental change in operations, value delivery, and their relationship with the environment [7],[10]; all of this is thanks to the automation of workflows, the generation of knowledge from big data, and facilitating more natural interaction between humans and technological systems [6],[11].

All this would not be possible without the capacity of PLN, understood as natural language processing; it is a central piece, since it allows computer systems to interpret, generate, and respond to human language, offering an increasingly better accuracy percentage [11],[12]; likewise, machine learning, which enables predictive modeling, results in resource optimization, real-time decision-making, and anticipation of behaviors [2][26][37][61].

All this is not perfect: the social environment and a non-homogeneous economic environment affect the reach of these technologies; the lack of technological infrastructure, the lack of digital skills, and the asymmetries in regulatory frameworks generate a persistent inequality and fuel the debate on digital transformation and AI [12][43][54].

The objective of this study was to present the results of applying a systematic review of the scientific literature related to the role of artificial intelligence in process automation, human-machine interaction, and the social impact of converging technologies. Fig. 1 presents the stages of the method defined for this review.

The research questions formulated are as follows:

RQ1: How does artificial intelligence influence the automation of complex processes in different technology sectors?

RQ2: How do the study of human behavior and the analysis of socio-behavioral profiles influence the design of AI interaction systems?

RQ3: What advances have been developed in natural language processing to improve interaction between humans and intelligent systems?

RQ4: What is the role of predictive modeling in the analysis of future scenarios and in understanding the social impact of converging technologies?

RQ5: What strategies have been proposed to ensure that AI-based technologies reduce technological access gaps?

RQ6: How do AI-based conversational agents contribute to improving communication between users and digital platforms?

Table 1. Definition of general concepts with PIPOH.

Source: own elaboration.

The search strategy consisted of explaining general terms and synonyms in specialized scientific databases, aiming to broaden the coverage of this article. Search terms combined with Boolean operators were determined by applying time filters from 2017 onward, thereby ensuring the relevance and timeliness of the included studies.

Table 2 presents the terms, synonyms, and filters used.

Table 2. Search terms.

Source: own elaboration.

The search process was performed in the databases listed in Table 3, consulting fields such as title, keywords, abstract, and full text.

Table 3. Databases consulted.

Source: own elaboration.

Table 4. Search string.

Source: own elaboration.

The studies identified in the search were analyzed according to the following aspects: title, keywords, abstract, introduction, background, methodology, results, and conclusions.

The inclusion criteria are defined as follows: scientific documents related to the role of artificial intelligence in process automation, human-machine interaction, predictive models, conventional AI agents, and digital equity, all within the context of digital transformation.

The exclusion criteria are defined as follows: studies without a verifiable bibliographic citation, documents or works not available for download, and publications prior to 2020.

Given that this systematic review used 4 databases, along with their respective synonyms and filters, and applied 3 phases in each, it was necessary to conduct the searches at different times, as shown in Table 5.

The selection was carried out in 3 phases:

Phase 1: Duplicate removal

Phase 2: Removal of articles not available for download.

Phase 3: Application of inclusion and exclusion criteria.

Table 5. Dates of consultation and file download.

Source: own elaboration.

To assess the quality of the selected documents, 7 criteria were considered: provenance of the sources, impact of the study, relevance of the content, objective of the research, context of the study, objectivity of the methodological design, and scientific rigor in data analysis. Next, Table 6 presents the results of this evaluation process.

Table 6. Quality evaluation in the search and selection processes.

Source: own elaboration.

Following the selection and quality assessment phases, the 61 documents were read. The data were obtained and organized according to the 6 defined questions, thereby constructing the state of the art on the role of AI in digital transformation from the dimensions of automation and interaction.

Artificial intelligence in the automation of complex processes is applicable to many sectors of knowledge, industry, or production. The reviewed documents show that AI enables the transition from human reasoning to computational thinking that optimizes time, processes, precision, and resources in areas as diverse as the energy sector, health, industry, finance, and the environment [1][2][10][16][26].

In the energy sector, artificial intelligence helps electricity reach homes and businesses more efficiently. This is achieved through specialized algorithms that reduce energy losses in distribution networks, facilitate the prediction of electricity demand, and optimize the use of renewable energy such as solar and wind [1][2][3][4]. Likewise, AI systems can detect and prevent possible cyberattacks on electrical systems [5], which is crucial to ensure the security and continuity of energy supply.

Within the health sector, AI is also being applied to disease diagnosis, the design of new drugs, and hospital management. By analyzing large volumes of information, such as clinical images or patient data, deep learning models can identify patterns that would be impossible to detect manually [10][11]. Additionally, intelligent systems support risk assessment in clinical environments and the personalization of medical treatments [12][13][14][15].

In the industrial domain, AI systems enable real-time anomaly detection and equipment failure prediction, thereby facilitating preventive maintenance and improving operational efficiency [16][17][18][19]. Likewise, in construction projects, AI-assisted automation optimizes planning, time control, and resource distribution [9][20].

In finance, machine learning models make it possible to assess credit risks, detect fraudulent behavior, and make market predictions with greater precision [26][27]. Finally, in environmental applications, AI contributes to tasks such as solid waste classification, tsunami detection, and earthquake response [31][32][33].

The study of human behavior and the analysis of socio-behavioral profiles are key foundations in the design of AI interaction systems. The reviewed documents show that the incorporation of behavioral models into intelligent systems enables greater personalization, adaptability, and efficiency in user-system interaction [39][40][41][42].

A prominent approach in the literature is Human-Centered AI (HCAI), which proposes developing intelligent systems by placing human well-being, autonomy, and values at the center. This approach argues that AI systems should complement, rather than replace, human capabilities, and that their design should integrate ethical principles from the initial stages of development [43].

In digital marketing, AI algorithms analyze browsing behaviors, purchase preferences, and interaction patterns to build highly detailed psychological and consumer profiles [44][45][46]. This intelligent segmentation enables personalized experiences and advertising messages tailored to the user's individual profile [49].

However, the use of socio-behavioral profiles raises significant ethical concerns regarding privacy, informed consent, and the risk of manipulation. Micro-targeting, although powerful as a personalization tool, can affect user autonomy in the absence of regulatory frameworks that guarantee its responsible use [43][51].

Natural language processing (NLP) has advanced significantly, transforming the way intelligent systems understand and generate human language. The reviewed studies highlight that the Transformer architecture and its derivatives, such as BERT and GPT, represent a turning point in models' ability to capture semantic context in depth and generate coherent text with high accuracy [22][23][24].

These technologies are applied in a wide variety of contexts, from the production of academic and journalistic content to the automation of tasks in business environments. In scientific research, NLP tools facilitate the search, classification, and synthesis of large volumes of specialized literature [22][23][24].

Despite these advances, studies identify a critical phenomenon known as "hallucinations," in which language models generate information that appears plausible but is factually incorrect. To address this problem, strategies such as Retrieval-Augmented Generation (RAG) systems have been developed, which combine text generation with querying verified knowledge bases [23][50].

Furthermore, it is acknowledged that NLP models can reflect biases present in training data, potentially perpetuating stereotypes or generating discriminatory responses. This reinforces the need to implement auditing and continuous evaluation processes for natural language-based systems [51][54].

AI-based predictive modeling represents one of the most powerful tools for anticipating behaviors and trends in social and economic contexts. The studies reviewed show that machine learning and deep learning models allow predictions to be made with an unprecedented level of granularity, ranging from individual behavior to large-scale collective patterns [26][27][28].

In the financial domain, predictive models have demonstrated a superior ability to assess credit risk compared with traditional statistical methods. The application of techniques such as artificial neural networks and decision trees enables the identification of hidden patterns in large volumes of transactional data [26][27][37].

In the military and security sector, predictive AI is used to anticipate threats, optimize logistics, and support strategic decision-making. However, its use in this context also gives rise to debates concerning the ethics of autonomous combat and the limits of human intervention in critical decisions [28][29].

In the field of social sciences and communication, predictive models make it possible to analyze trends in digital media, detect patterns of disinformation, and anticipate behaviors in online communities [30][52]. However, the quality and representativeness of the training data are decisive for the reliability of these models, with algorithmic biases being one of the main challenges to overcome [43][54].

Equity in access to AI-based technologies is one of the most urgent challenges in the context of digital transformation. The studies reviewed show that the digital divide is not only an infrastructure problem, but also includes multiple dimensions: access to devices, connectivity, digital skills, language, and regulatory frameworks [12][43][54].

One of the most prominent strategies in the literature is the development of explainable AI (XAI) systems, which seeks to make intelligent models understandable to non-specialized users. Algorithmic transparency not only contributes to trust in AI systems, but also facilitates their adoption in contexts where technological distrust is high [43][51].

Another relevant line of action is the design of inclusive and accessible interfaces that consider the needs of populations with low digital literacy, linguistic diversity, or physical limitations. Studies such as that of Hidalgo Suarez et al. propose the use of computer-supported collaborative learning (CSCL) to bridge the gap in digital competencies necessary to take advantage of AI-based technologies [40].

AI-based conversational agents represent one of the most visible interfaces of digital transformation in the daily interaction between users and digital platforms. Based on NLP and deep learning techniques, these systems have evolved from simple command execution to contextual understanding, intent inference, and the generation of personalized responses [25][47].

In education, chatbots and virtual assistants integrated into online learning platforms can provide immediate feedback, guide students in their learning process, and adapt content to their learning profile [40][41]. In digital health, conversational agents facilitate initial patient triage, appointment management, and treatment follow-up, reducing the burden on clinical staff [12][47].

In marketing and customer service, AI-based chatbots transform information search by integrating advertising in a conversational and highly relevant manner, progressively displacing traditional methods [41]. Affective computing, which enables agents to interpret and respond to user emotions through the analysis of language and tone, opens new possibilities for a more empathetic and closer interaction [47].

However, the implementation of conversational agents poses challenges related to data privacy, the veracity of the information generated, and the risk of creating technological dependence among users [54][55]. Transparency regarding the artificial nature of the agent and informed consent mechanisms emerge as fundamental ethical requirements [43][51].