Artificial intelligence technologies of the symbolic paradigm and their interdisciplinary implications for Information ScienceABSTRACT
Introduction: Artificial intelligence has been transforming science, the economy, the job market, and the social context. These technologies began to gain popularity in 2022, with the emergence of generative artificial intelligence models such as ChatGPT and DeepSeek. Artificial intelligence encompasses technological paradigms with distinct applications, such as the symbolic, which uses symbols and inductive logic to develop autonomous systems.
Objective: To analyze the production of articles in this paradigm, in the Web of Science database, in relation to popularity over time, by field, and by country, highlighting connections with Information Science theories.
Methodology: The quantitative data from the articles were analyzed through graphs and charts, calculating absolute totals, partial totals, means, and percentages. Furthermore, DeepSeek was used to relate the technologies to Information Science theories and authors.
Results: China's leadership in research was observed, with Brazil's limited participation, and, specifically, Information Science's modest participation. Regarding interdisciplinarity, seventeen Information Science theories/technologies were identified, highlighting knowledge representation, ontologies, and information retrieval.
Conclusion: Some artificial intelligence technologies are on the rise, while others are stabilizing or declining. Information Science needs to invest in research in this niche, and Brazil needs to strengthen public-private partnerships to leverage its Artificial Intelligence ecosystems, following China's example.
KEYWORDS:
Artificial intelligence; Information science; Symbolic paradigm; Scientometrics
RESUMO
Introdução: A inteligência artificial está transformando a Ciência, a economia, o mercado laboral e o contexto social. A popularização dessas tecnologias se deu a partir de 2022, quando surgiram modelos de inteligência artificial generativa como Chat GPT e DeepSeek. Inteligência artificial é um termo que contempla paradigmas tecnológicos com distintas aplicações, tal qual o simbólico, que utiliza símbolos e lógica indutiva para desenvolver sistemas autônomos.
Objetivo: Objetiva-se analisar a produção de artigos nesse paradigma, na base Web of Science, em relação à popularidade ao longo do tempo, por área e por país, evidenciando pontes com teorias da Ciência da Informação.
Metodologia: Os dados quantitativos dos artigos foram analisados em forma de gráficos e tabelas, calculando-se totais absolutos, parciais, médias e porcentagens. Ademais, utilizou-se o DeepSeek para relacionar as tecnologias com teorias/autores da Ciência da Informação.
Resultados: Como resultados, notou-se a liderança da China nas pesquisas, uma participação tímida do Brasil, e, especificamente, tímida da Ciência da Informação. Concernente à interdisciplinaridade, foram apontadas dezessete teorias/tecnologias da Ciência da Informação, destacando a representação do conhecimento, ontologias e recuperação da informação.
Conclusão: Verificou-se que algumas tecnologias de inteligência artificial estão em ascensão, enquanto outras em estabilização ou declínio. É preciso que a Ciência da Informação invista mais em pesquisas neste nicho e que o Brasil acentue parcerias público-privadas para alavancar seus ecossistemas de Inteligência artificial, a exemplo da China.
PALAVRAS-CHAVE:
Inteligência artificial; Ciência da informação; Paradigma simbólico; Cientometria
1 INTRODUCTION
The topic of Artificial Intelligence (AI) has sparked significant interest in recent years, among laypeople, as well as educators, other professionals, the scientific community, and governments around the world. AI can be understood as an umbrella term that comprehends a variety of paradigms, techniques, and applications, encompassing computing, mathematics, and other related fields. The difference between an AI system and traditional programming is that these systems are trained using a large volume of data and 'learn' patterns to perform tasks and make decisions autonomously, without the programmer pre-determining each step.
“AI system” means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions [...] (European Union, online , 2025).
This topic has clearly entered the public debate with much greater projection from 2022 onwards, with the launch of Chat GPT, which popularized, or even vulgarized, the use of systems trained from Large Language Models (LLMs).
Beyond generative AI tools, which offer rapid and, more or less, accurate answers to countless imaginable questions, AI can be used in fields as diverse and crucial as the environment, health, business, and others. This is why the Brazilian government has created a set of policies for AI development aimed at serving various spheres of strategic interest to society, with investments totaling R$23 billion until 2028 (CNN Brasil, 2024). This movement aligns with the recent rush by major powers, such as China and the United States, to use AI as an instrument of national sovereignty and technological self-sufficiency (Chang; Arcesati; Hmaidi, 2025).
Keeping up with AI developments is, therefore, essential for survival in the contemporary world. Therefore, those who fail to keep up with this evolution will be doomed to obsolescence, to being dispensable and unnecessary. This same logic also applies to areas of knowledge and research, such as Information Science (IS), the focus of this article. Furthermore, it is worth highlighting the existence of a 'natural' intersection between IS and computing through its primary object of study, which is, precisely, information.
Ferneda (2006) discusses the use of AI neural networks in information retrieval systems, strongly advocating for a deeper relationship between information science and computing. However, the author cautioned that IS is responsible for critically contextualizing this relationship, as its knowledge is not constructed in the 'cultural void' of algorithms. The greatest scientific contribution of this work, therefore, is to highlight the importance of this symbiotic relationship between the two areas, where computing provides IS with technical tools and IS provides computing with the pragmatic and critical perspective to consider the future of this technology, ensuring that it meets real human needs. It is also worth highlighting a gap in international literature regarding the contribution of IS to the discussion of more technical AI topics, specifically symbolic paradigm technologies, which, as discussed and demonstrated throughout this work, are of particular interest to the semantic web, knowledge representation, organization, and management, among other subareas. It is paradoxical that the computing-IS relationship around the themes is limited in prestigious databases such as the Web of Science.
In this context, the present research aims to analyze the characteristics of scientific production regarding the symbolic paradigm of AI, the interest of the scientific community in the topic, the popularity of each of its approaches, and possible points of convergence with CI.
2 THEORETICAL FRAMEWORK
As explained in the previous section, AI essentially aims to create computer systems that can learn through training with pre-existing data, aiming for autonomy in performing complex tasks such as pattern recognition, classification, clustering, and others. Developing this 'intelligence' requires intensive use of computing resources, specialized hardware, processors, memory, complex software frameworks, and electrical power. These systems, therefore, consist of true layered 'ecosystems': Graphic Processing Unit (GPU) chips interconnected in clusters; High Bandwidth Memory (HBM) associated with GPU chips; and software frameworks for programming AI models (Chang; Arcesati; Hmaidi, 2025).
Figure 1, below, highlights the current AI 'war' between the United States and China, concerning their most popular natural language processing models, anchored in chips by Nvidia and Huawei.
It is highlighted that AI encompasses a wide variety of approaches and techniques for application in diverse problem domains. To better understand this panorama, Corea (2018) mapped the AI paradigms, composed of three macro-approaches: symbolic, which represents intelligence through symbol manipulation; subsymbolic, where no specific representation of knowledge is given in advance; and statistical, which solves problems through mathematics. According to this author (2018), within the symbolic paradigm, the knowledge-based tools (ontologies, databases, information, and rules) are: Robotic Process Automation (RPA) and Expert Systems (ES); as well as logic-based tools (used for knowledge representation and problem-solving): Inductive Logic Programming (ILP).
This symbolic paradigm, therefore, represents knowledge through symbols related by deductive logical rules, combining principles of Cognitive Science and Mathematics, approaching the functioning of human language.
A symbolic AI system works by carrying out a series of logic-like reasoning steps over language-like representations. The representations are typically propositional in character, and assert that certain relations hold between certain objects, while each reasoning step computes a further set of relations that follow from those already established, according to a formally specified set of inference rules (Garnelo ; Shanahan , 2019, p. 17).
This paradigm has a relatively long history, based on a framework of previously defined knowledge/symbols, which differs from more recent techniques. Symbolic AI was the dominant paradigm until the late 1980s, unlike the more recent data-driven AI, which uses statistical techniques. This approach represents objects as symbols and the relationships between them, such as the "has" relationship between the object "person" and the object "car" (Kalota, 2024). This type of AI requires complete knowledge of the context of its application to function, which results in transparency regarding the method used to obtain results. It is widely used in expert systems, aiming to solve complex problems that require a high level of human competence (Kalota, 2024). Humans, therefore, become essential as holders of specialized knowledge who help the machine interpret and make sense of the relationships between data in a specific context. Furthermore, the techniques and tools built under this paradigm have the advantage of explainability and clarity, as they are based on explicit rules, which differ from other types of AI, which function as black boxes.
2.1 Expert systems
Having briefly explained the paradigm, its three approaches need to be discussed, starting with expert systems. Janjanam, Bharathi and Manjunatha (2021) specifically mention Knowledge-Based Expert Systems (KBES), which incorporate human experience, assisting people in solving poorly structured problems that require decision-making, anchored in synthesis, evaluation, simulation, judgment, and heuristic solutions based on experience. For these authors (2021), such systems consist of: a robust knowledge base, composed of facts from the application domain, as well as heuristics for its application; a control mechanism, inference engine, or rule interpreter; and a user interface.
This type of system acts as a human expert such as a doctor, a chemist, a geologist, drawing logical inferences from the data it receives, interpreting them using its own knowledge base, guided by the rules specific to the application context. Its architecture is composed of three main blocks, as shown in Figure 2: the knowledge base; the inference engine; and the user interface.
The user interface allows ordinary users, even those without programming or technological knowledge, to interact with the system, as it should be simple, understandable, and useful, simulating a person's interaction with an expert (Ahmadi; Abadi, 2020). The interface hides the system's true complexity from the layperson, serving as a data entry point and presenting the output of the processing performed by the inference engine.
For Zwass (2024), the inference engine is a layer that interprets and evaluates the facts/data present in the knowledge base, aiming to provide the response requested by the user. The inference engine basically takes the input data provided in the interface and submits it for investigation to the knowledge base, considering internal and external factors (Ahmadi; Abadi, 2020).
An expert system for medicine, for example, could collect vital signs from a patient and, based on its knowledge of symptoms, judge whether they are positive or negative for disease 'X'.
For Janjanam, Bharathi and Manjunatha (2021), inference engines determine which heuristic search techniques are employed on the rules present in the knowledge base; this layer determines which rules are followed in each case, implements these rules, and decides how to find an optimal solution. Possible solutions are found through "if-then" statements: if a certain condition is true, then a certain inference or course of action can be taken (Zwass, 2024). Hypothetically, if the patient's saturation is such, their body temperature is above such, and their lung X-rays show such signs, with x percentage of certainty, it could be stated that they are infected with COVID. However, if any of the conditions are not met, the probability of a negative diagnosis is greater.
The knowledge base itself is composed of facts and a potentially myriad of rules, collected through interviews and observation with experts in the field for which the system is developed (Zwass, 2024). The conclusions, or recommendations obtained by the system are, therefore, tied to probability, given the impossibility of complete certainty in the results, according to Zwass. To conclude, the group of knowledge-based AI techniques, RPA shall be discussed.
2.2 Robotic Process Automation
Another relevant AI approach within the symbolic paradigm is Robotic Process Automation (RPA), which aims to increase the efficiency of repetitive services in organizations in general. According to Jovanović, Đurić and Šibalija (2018), contrary to popular belief, the term "robot" here does not represent any physical electronic or mechanical device, but rather a software solution programmed to perform highly structured and repetitive tasks and processes normally performed by humans. According to the authors (2018), this technique allows for the automation of tasks considered strenuous and tedious, efficiently and at low cost, in which technology "learns" to reproduce the work "tirelessly." RPA does not modify the functioning of existing systems, but acts at the user interface layer, reproducing tasks that would otherwise be performed by the system, such as filling out spreadsheets, replying to emails, interacting with customers via chat, and so on. It is in this sense that the trend towards replacing human workers with 'robots' in repetitive and 'mechanical' tasks becomes clear.
Even the analysis of unstructured information is possible through RPA, presenting users with useful trends for decision-making, such as analyzing transaction patterns that can help prevent fraud/money laundering (Madakam; Holmukhe; Jaiswal, 2019). The following is a more formal and complete definition of the term:
This term amalgamates robotics, referring to software agents acting as humans beings in system interactions, and process automation [...]. In essence, RPA is a relatively new technology comprising software agents called ‘bots’ that mimic the manual path taken by a human through a range of computer applications when performing certain tasks in a business process. The tasks that bots perform are typically rule-based, well-structured, and repetitive. Examples of tasks that bots perform include data transfer between applications through screen scraping, automated email query processing, and collation of payroll data from different sources (Syed, 2020, p. 1).
However, not only simple tasks of entering values and clicking can be automated. This technology has become increasingly sophisticated. According to Madakam, Holmukhe and Jaiswal (2019), RPA is capable of producing highly elaborate answers to complex questions due to its ability to process and relate large amounts of information, a capacity far beyond what a human could normally search/assimilate. It can, for example, determine whether Chinese market stocks are on an upward trend after analyzing a range of factors. These authors (2019) also assert that RPA can handle rules-based tasks, as mentioned above, repetitive, freeing employees to serve their customers with higher-value-added services, even though this can result in job losses.
Expert systems and RPA techniques are located at the intersection of the 'knowledge-based' and 'logic-based' quadrants. The next section discusses ILP technology, which falls solely into the 'logic-based' quadrant. However, all three technologies mentioned fall within the symbolic paradigm of AI, grounded in a framework of data and knowledge.
2.3 Inductive Logic Programming
Classical logic consists of a formal system built primarily on four components: logical axioms, formal language, rules of deduction, and possible non-logical axioms. These principles develop notions of theorems, proofs, consistency, and so on. Classical logic derives from the principles of Aristotelian logic. These principles are: the principle of excluded middle, the principle of identity, and the principle of non-contradiction. According to the principle of excluded middle, by given any sentence, either that sentence is true or its negation is, there is no third alternative. The principle of identity reflects that every object is identical to itself. Finally, the principle of non-contradiction states that a proposition and its respective negation cannot be considered true simultaneously and in the same aspect (Mello, 2024).
ILP is a technique based on the process of induction, which consists of obtaining generalizations and logical conclusions from the examination of particular cases, guided by prior knowledge of the context of application, and strongly based on mathematics. ILP, according to Cropper et al. (2022), is a form of machine learning (ML) whose functionality can be summarized as introducing a hypothesis for generalizing training cases. For these authors (2022), while other types of ML represent the data they feed on through vectors and tensors, ILP uses sets of logical rules, highlighting relationships between these data. A more formal and detailed conceptualization of this technology is presented below.
Inductive Logic Programming (ILP) is a research area formed at the intersection of Machine Learning and Logic Programming. ILP systems develop predicate descriptions from examples and background knowledge. The examples, background knowledge and final descriptions are all described as logic programs (Imperial College London, 2024, online).
Therefore, the underlying logic in the relationship between data is learned through training with prior knowledge, which serves to generalize relationships across similar data sets. According to Sen (2022), ILP is ideal for inferring logical rules from labeled data sets. These rules are explicitly symbolic, which makes such models both inspectable and interpretable, providing an efficient means of storing knowledge. When referring to labels in AI, it means that the data set has an expected predictive value attached to it. For example: if the proportion of the petal and sepal dimensions is such, the flower is an iris of the setosa species; if the measurements are different, it is a versicolor species. Thus, when the user enters new values, the AI knows how to classify them based on this learning. Labels are based on so-called background knowledge.
The ability of ILP systems to accommodate background knowledge is also fundamental. Some relationships learned in particular applications have been considered as discoveries within those domains. Application areas include: learning drug structure-activity rules [...] learning rules from chess databases [...] (Imperial College London, 2024, online).
ILP can also serve as a basis for complementing other more complex models, such as MML, which deal with human language, acting on the logic of grammatical functions. Furthermore, Cropper et al. (2022) highlight some advantages of ILP over other ML models: its efficiency in learning from relatively small data corpuses with good accuracy; the representation of knowledge as logical programs allows learning from complex relational information; due to its symbolic nature of knowledge representation, ILP can 'reason' about hypotheses, allowing it to create programs with optimized solutions; and due to the similarity of logic to human language, ILP programs can be more easily understood by humans, which is essential for the explainability of AI functioning.
A distinct but complementary perspective to ILP is that of more complex LLM models, such as DeepSeek and Chat GPT, which also use inductive logic. Costa and Abbe (2000) reflect: "Inconsistency cannot be, at least directly, handled through the classical logic on which most programming languages are based" (Mello, 2024, p. 59). Formal systems, which follow principles of classical logic, when handling contradictory data, causes what is denominated the explosion principle in paraconsistent logic. In the sense that if one contradiction prevails, then everything else also prevails. To this end, it has now been necessary to seek alternatives, beyond classical logic to programming languages, for example (Mello, 2024). By reasoning from inconsistent logic, one can start from inconsistent premises and arrive at sensible results without causing total inconsistency, as not every contradiction unfolds into absurdity.
In this context, software such as DeepSeek reflects an open-source model capable of solving complex questions in a more critical and less generalist manner, as it utilizes paraconsistent logic propositions and handles contradictions effectively. This increases its reliability, allowing the user to understand the process as the inference progresses. Therefore, it goes beyond the binary nature of classical logic, which considers only two possibilities for any proposition: true or false.
As explained, symbolic paradigm technologies are strongly based not only on data but also on knowledge, a characteristic that allows for interdisciplinary relationships with information science and a consequent, and necessary, problematization and interpretation of its theoretical impacts and professional practice. Is the information science academic community taking full advantage of its interaction with this rapidly consolidating field of computing?
3 METHODOLOGICAL PROCEDURES
This research is a scientometric analysis conducted in the Web of Science (WOS) database, core collection. In September 2025, three searches were conducted in this database for each of the AI technologies discussed in this paper: ES, RPA, and ILP. The time period considered comprises five years, from 2020 to 2025. The fields used in the search were: Author Keywords (AK); Abstract (AB); Title (TI).
Furthermore, only academic articles were selected. Another filter was related to the restriction of knowledge areas, including fields of Computer Science and Information Systems: Computer Science Information Systems; Computer Science Software Engineering; Computer Science Artificial Intelligence; Computer Science Theory Methods; Computer Science Interdisciplinary Applications; Information Science Library Science.
In the first stage, the number of articles related to each technology was computed in relation to the authors' nationality, year of publication, and average number of articles per field of knowledge. These data allowed to verify which countries publish the most on these topics; Brazil's position in this scenario; whether IS is contributing satisfactorily compared to other fields; and whether the topics are on the rise or decline over the years.
DeepSeek was used to investigate correlations between the three AI technologies discussed and the most relevant IS theories/authors, requesting detailed explanations. DeepSeek was chosen as its use is free and presents superior performance compared to paid competitors, such as Chat GPT (CNN, 2025).
The Deepseek search used three similar search strategies, following the instruction: "Explain which theories and authors in the field of Information Science address topics addressed by the field of [...]. Provide bibliographical references at the end of your search." The 'areas' considered in the above instruction were, respectively: Inductive Logic Programming / ILP; Robotic Process Automation / RPA; Expert Systems. The responses were subsequently verified, and only those with complete bibliographical references were accepted. Finally, the number of mentions of each IS theory in relation to AI technologies was calculated.
It is noteworthy that Trindade and Oliveira (2024) recognize the use of generative AI as appropriate for scientific research, noting that the researcher must be aware of ethical implications, plagiarism, biases, and possible inconsistencies in the information, verifying and correcting any discrepancies. These authors highlight the need for a critical view of the use of AI in academia through information literacy. In this paper, DeepSeek responses were checked, and only analyses that indicated consistent and correct bibliographic references were accepted.
4 RESULTS AND DISCUSSION
This section presents the results obtained through searches in the WOS database, concerning the three symbolic paradigm technologies already discussed in this article: ILP, RPA, and ES. The analytical approach was eminently quantitative: in the first stage, some numerical metrics on the characteristics of the corpus of publications are presented; in the second stage, the correlation between the AI technologies and the corresponding IS theories/authors was performed. Chart 1 shows the general search criteria/filters used in all three searches conducted in the database.
The research was performed in September 2025, covering the time period from 2020 to 2025, selecting only articles from journals in areas related to computing and IS.
Chart 2 shows the results of the search for ILP technology. The keywords used were author (AK), abstract (AB), and title (TI). A total of 837 articles were retrieved, with an average of 139.5 articles per knowledge area. The IS is below this level, with only two published articles.
Chart 3 highlights the production on the topic by country. China leads the research, with 268 articles published, the United States is in second, with less than half that number. The 10 countries with the highest production were separated for analysis, with Brazil ranking 10th, with 37 publications.
China's performance reflects the country's government investment in AI as a strategic measure for scientific, technological, and economic development, as shown by Chang, Arcesati and Hmaidi (2025). According to these authors (2025), this performance is the result of a clear development plan through public-private partnerships. China surprised the stock market in the early 2025, causing a true "seismic shock", knocking down the US Big Tech company shares with the launch of DeepSeek, which uses open-source technologies and low-cost chips (Bloomberg, 2025). As previously mentioned, the Brazilian government has also made efforts to develop a national AI ecosystem (CNN Brasil, 2024). The countries should invest in their own technologies to reduce monopoly and consequent dependence on foreign solutions. It is evident that dependence in the field of AI can negatively impact not only the economic aspect but also information security, as AIs 'learn' from induction on information/data generated by humans and other sources.
The graph in figure 3 provides a visual dimension of academic production in ILP per country.
Chart 4 shows a compilation of scientific production in ILP from 2020 to 2025, showing relative stability with a slight downward trend since 2023. The sharp decline in 2025 is explained by the fact that this research collected data up to the beginning of the second half of 2025. However, it is possible to project that the publication level reached in the previous year will be reached or even surpassed by the end of 2025, given that by September 2025, 80% of the 2024 production had already been achieved. This choice of time frame was strategic, aiming to obtain the most recent data possible and infer possible impacts of the international AI race, which gained unprecedented momentum at the beginning of this year with the entry of new players.
The graph in figure 4 provides a visual dimension of the evolution of the theme over time, with the abrupt drop in 2025 for the reasons already mentioned above, and growth can even be projected until the end of the year, given that 80% of the previous year's production volume was already reached at the beginning of the second half of 2025.
The next analyzed topic is academic literature on RPA technology, as shown in Chart 5. Using the filters presented in Chart 1, a total of 185 papers were retrieved, with an average of 30.83 per field. In this context, IS's output on the topic is quite close to the average, with 25 papers.
In this production niche, Brazil ranked seventh among the countries that publish the most, tied with Italy and Spain, as seen in Chart 6. China once again emerges as the leader, followed by India in second, and the United States in third. Once again, the importance of massive government investment and public-private partnerships in strategic research areas is evident, reflected in strong market performance.
Figure 5 represents the production data per country, discussed above.
As for the popularity of the research topic over time, Chart 7 shows a stronger growth trend, although the total number of articles is still relatively modest compared to ILP technology, for example.
Figure 6 shows a clear upward trend in the data, with an apparent decline in the last year, justified by the fact that the data were collected at the beginning of the second half of 2025, that is, before the end of the year. Production at the beginning of the second half of 2025, however, already represents 63% of the total production for the previous year, demonstrating a projected growth forecast.
Finally, Chart 8 displays the results of published articles on ES technology, using the same filters as the two previous surveys, as described above. From a total of 471 articles, an average of 78.5 papers per field of knowledge was obtained, with IS well below this parameter, with 12 published papers.
As shown in Chart 9, China once again leads with 118 publications, the United States in second with 53, less than half of the number of first place. Brazil ranks 17th, with 9 papers.
The graph in Figure 7 displays the above data, showing that the country with the most publications has more than twice the production of the second-placed country.
Chart 10 shows the evolution of production per year; out of the three themes, it is the one showing the most pronounced downward trend, with production in 2024 being 69% lower than that at the beginning of the analysis period, in 2020.
The graph in figure 8 shows a sharp downward trend from 2022 to 2023 with a slight recovery in 2024. In the beginning of the second semester of 2025 had obtained 70% of the amount published in the previous year, pointing to a stabilization trend.
After quantitatively analyzing the literature on ILP, RPA, and ES technologies, several aspects stand out: China's leadership with a significant advantage in all scenarios; the rise of RPA technology amid the decline and/or stabilization of other technologies; the moderate to weak contribution of IS to academic production on these topics; and Brazil's weak performance in all three scenarios. It can be inferred that IS must make real efforts to foster more intense research on topics so relevant to current and future contexts, also considering the central role of data, information, and knowledge in these technologies.
4.1 Intersection between the symbolic paradigm and IS theories
After the quantitative analysis, it is questioned how, exactly, IS would fit into the context of the three technologies to gain greater insights on the interdisciplinary dialogues that can be built. Since the IS theoretical framework is quite broad and complex, and the possible theory/technology combinations potentially vast, AI was used to assist in carrying out this Herculean task. Therefore, DeepSeek was adopted as the methodological tool. However, care was taken to ask the AI to provide bibliographical references to support its analyses, and citations that did not explicitly provide their sources were discarded. Chart 11 displays the results.
As observed from the chart, regarding ILP technology, the following topics were related: Knowledge Representation and Logic; Ontologies and the Semantic Web; Information Retrieval and Text Mining; Machine Learning and Information Science. Among the cited authors are B. Hjorland, R. Baeza-Yates, and L. Floridi. The connection revealed by DeepSeek follows the reasoning that, by dealing with logic and mathematical models, ILP is consistent with IS theories on knowledge representation, ontologies, and information retrieval.
The role of the semantic web for modern society, constantly connected to the internet for both work and leisure purposes, stands out. Nunes, Maculan and Almeida (2020) discuss the importance of this field for information society, addressing misinformation and informational ambiguity that plague this era, giving meaning to data, improving its processing, representation, retrieval, and validation.
Regarding RPA, the following topics were covered: Automation Theory and Information Management; Human-Computer Interaction and Cybernetics; Process and Information Flow Management; Data Science and Knowledge Representation. Featured authors include L. Floridi, P. Outlet, T.H. Davenport, P. Checkland, S. Suton, and others. The link highlighted is that RPA, by using robots to perform repetitive tasks, can be used in information management and knowledge organization (KO).
For Marcondes (2021), KO emerged to manage the post-war explosion of scientific knowledge and, today, faces new challenges with big data and the web, starting with classification systems such as CDD and UDC, and advancing into the realm of thesauri and computational ontologies. According to the author (2021), in addition to technical aspects, KO must focus on solidifying its theoretical foundations to make sense of this new panorama, using contributions from ontologies and semiotics.
Regarding ES technology, due to its ability to store knowledge and process data to assist, or even replace, human experts in some tasks, it can be useful in organizing and representing knowledge, knowledge management, and information retrieval. Specifically, the following theories are mentioned: Knowledge Representation and Ontologies; Information Retrieval and Intelligent Systems; Artificial Intelligence and Information Science; Knowledge Management and Knowledge-Based Systems. Cited authors include Hjorland, Belkin, Salton, Takeuchi, and Nonaka, among others.
Chart 12 counts the number of occurrences of specific IS areas and related topics in the Deepseek analysis. It shows that knowledge representation, ontologies, and information retrieval are the relatively most prominent topics, while the other areas/topics appear only once. However, the difference between the most and least mentioned areas is small. In total, 17 IS theories/technologies were listed, with the most frequently mentioned ones receiving three to two mentions.
Knowledge representation, an area of IS, as seen in the table above, has a deep intersection with the symbolic paradigm of AI, as it aims to represent the document's main concepts in the form of symbols, condensing its content, in such a way to making its storage, organization, retrieval, and access more efficient. It is therefore natural that this topic appeared more frequently in the analysis generated by DeepSeek.
The original document interpretation to be included in the system, its description as an object, its origin, and the abstract of its content, based on the concepts contained therein, encompassing its conceptual essence as perfectly as possible. In this sense, representation would be a substitute for the document stored in the system, aiming at its retrieval. In this representation process, the document, or a set of documents, may be replaced by a condensed set of information, favoring its location and use by users (Alvarenga, 2003, p. 23).
Closely related to knowledge representation is the concept of ontologies, which also stood out in the analysis. Ontologies refer to structured models of knowledge representation that highlight the relationship between concepts within specific domains, with implications for information science, computing, and philosophy (Almeida, 2014).
The field of information retrieval also encompasses the concepts of information representation and ontologies, carrying an intrinsically strong connection with computing. Essentially, information retrieval aims to provide users with effective access to content of interest, usually through computerized systems, also encompassing document representation, storage, and organization (Silva, 2023).
In short, the study observed that technologies of the AI symbolic paradigm can serve not only more deterministic, formal applications of IS, such as knowledge organization and representation, but also to support more humanized and less 'exact' activities, such as knowledge management. It is hoped, therefore, that the systematization of the relationship between IS and the aforementioned technologies can serve as a map for the scientific community, aiming at a consistent increase in interdisciplinary production on the topic addressed in this article, helping IS to more closely monitor the pace of technological development.
5 FINAL CONSIDERATIONS
The discussions revealed the significant relevance of AI symbolic paradigm technologies for the scientific community, IS professionals and researchers, and society at large, due to their applications, which influence a wide range of everyday life. The emphasis of AI symbolic paradigm technologies on data and knowledge confers on them a special value for rich interdisciplinary dialogues with IS.
The first phase of quantitative analysis revealed a rise in interest in RPA technology, a relative stabilization in ILP, and a decline in interest in ES technology between 2020 and 2025. It would be interesting to analyze a longer period to see if these trends are confirmed. The relatively weak performance of IS in this research niche was also noteworthy, despite its close relationship with the topics of information, knowledge, and data. This undoubtedly serves as a warning sign for further work in this area.
China's leading role in all analysis scenarios was evident, indicating that the current massive investment in market-oriented AI technologies may be supported by strong investment in scientific research. This is a lesson to be learned by Brazil, which is just beginning to strengthen public policies focused on AI ecosystems.
As limitations of this study, the analysis focused on WOS exclusively could have limited the richness of the results. The countries with lower production rankings for the technologies covered in this study might also be investigating these topics more diligently, but publishing in journals indexed by other sources, whether international or national.
Acknowledgments:
Not applicable.
References
-
AHMADI, M.; ABADI, M. Q. H. A review of using object-orientation properties of C++ for designing expert system in strategic planning. Computer Science Review, Amsterdam, v. 37, n. art. 100282, 2020. DOI: https://doi.org/10.1016/j.cosrev.2020.100282
» https://doi.org/10.1016/j.cosrev.2020.100282 -
ALMEIDA, Maurício Barcellos. Uma abordagem integrada sobre ontologias: ciência da informação, ciência da computação e filosofia. Perspectivas em Ciência da Informação, Belo Horizonte, v. 19, n. 3, p. 242-258, 2014. DOI: https://doi.org/10.1590/1981-5344/1736
» https://doi.org/10.1590/1981-5344/1736 -
ALVARENGA, Lídia. Representação do conhecimento na perspectiva da ciência da informação em tempo e espaço digitais. Encontros Bibli: Revista Eletrônica de Biblioteconomia e Ciência da Informação, Florianópolis, v. 8, n. 15, p. 18-40, 2003. DOI: https://doi.org/10.5007/1518-2924.2003v8n15p18
» https://doi.org/10.5007/1518-2924.2003v8n15p18 -
BLOOMBERG. Deepseek atualiza modelo de IA barato que surpreendeu big techs ocidentais. O Globo, Rio de Janeiro, 28/05/2025. Tecnologia. Disponível em: https://l1nk.dev/Ykpiv Acesso em: 03 fev. 2025.
» https://l1nk.dev/Ykpiv -
CHANG, Wendy; ARCESATI, Rebecca; HMAIDI, Antonia China’s drive toward self-reliance in artificial intelligence: from chips to large language models. Mercator Institute for China Studies, Berlin, 22/07/2025. Disponível em: https://l1nq.com/RH0TP Acesso em: 03 fev. 2025.
» https://l1nq.com/RH0TP -
CNN BRASIL. Entenda o que é DeepSeek, IA que derrubou ações de tecnologia nesta segunda. CNN Brasil, São Paulo, 27/01/2025. Disponível em: https://sl1nk.com/pWu77 Acesso em: 03 fev. 2025.
» https://sl1nk.com/pWu77 -
CNN BRASIL. Governo vai investir R$ 23 bilhões em inteligência artificial até 2028. CNN Brasil, São Paulo, 30/07/2024. Economia, Macroeconomia. Disponível em: https://sl1nk.com/vAi6Z Acesso em: 03 fev. 2025.
» https://sl1nk.com/vAi6Z -
COREA, Francesco. AI knowledge map: how to classify AI technologies: a sketch of a new AI technology landscape. Medium, San Francisco, 29/08/2018. Disponível em: https://sl1nk.com/RHzvj Acesso em: 03 fev. 2025.
» https://sl1nk.com/RHzvj -
COSTA, Newton C. A; ABBE, Jair Minoro. Paraconsistência em informática e inteligência artificial. Estudos Avançados, São Paulo, v. 14, n. 39, p. 161-174, 2020. DOI: https://doi.org/10.1590/S0103-40142000000200012
» https://doi.org/10.1590/S0103-40142000000200012 -
CROPPER, Andrew et al. Inductive logic programming at 30. Machine Learning, New York, v. 111, p. 147-172, 2022. DOI: https://doi.org/10.1007/s10994-021-06089-1
» https://doi.org/10.1007/s10994-021-06089-1 -
FERNEDA, Edberto. Redes neurais e sua aplicação em sistemas de recuperação de informação. Ciência da Informação, Brasília, v. 35, n. 1, p. 25-30, 2006. DOI: https://doi.org/10.1590/S0100-19652006000100003
» https://doi.org/10.1590/S0100-19652006000100003 -
GARNELO, Marta; SHANAHAN, Murray. Reconciling deep learning with symbolic artificial intelligence: representing objects and relations. Current Opinion in Behavioral Sciences, Amsterdam, v. 29, p. 17-23, 2019. DOI: https://doi.org/10.1016/j.cobeha.2018.12.010
» https://doi.org/10.1016/j.cobeha.2018.12.010 -
IMPERIAL COLLEGE LONDON. Inductive Logic Programming. London, 2024. Disponível em: https://www.doc.ic.ac.uk/~shm/ilp.html Acesso em: 03 fev. 2025.
» https://www.doc.ic.ac.uk/~shm/ilp.html -
JANJANAM, Durgaprasad et al. Design of an expert system architecture: An overview. Journal of Physics: Conference Series, Bristol, v. 1767, n. art. 012036, 2021. DOI: https://doi.org/10.1088/1742-6596/1767/1/012036
» https://doi.org/10.1088/1742-6596/1767/1/012036 - JOVANOVIĆ, Saša Z.; ĐURIĆ, Jelena S.; ŠIBALIJA, Tatjana V. Robotic process automation: overview and opportunities. International Journal Advanced Quality, Belgrade, v. 46, n. 3-4, 2018.
-
KALOTA, Faisal. A primer on generative artificial intelligence. Education Sciences, Basel, v. 14, n. 2, n. art. 172, 2024. DOI: https://doi.org/10.3390/educsci14020172
» https://doi.org/10.3390/educsci14020172 -
MADAKAM, Somayya; HOLMUKHE, Rajesh M.; JAISWAL, Durgesh K. The future digital work force: Robotic Process Automation (RPA). Journal of Information Systems and Technology Management, São Paulo, v. 16, n. 1, n. art. e201916001, 2019. DOI: https://doi.org/10.4301/S1807-1775201916001
» https://doi.org/10.4301/S1807-1775201916001 -
MARCONDES, Carlos Henrique. Fundamentos da organização do conhecimento. Ponto de Acesso, Salvador, v. 15, n. 3, p. 249-282, 2021. DOI: https://doi.org/10.9771/rpa.v15i3.47468
» https://doi.org/10.9771/rpa.v15i3.47468 - MELLO, Márcia Regina Gabriel de. Verdade, pós-verdade e desordem informacional: aportes teóricos e filosóficos à Ciência da Informação. 2024. 243f. Tese (Doutorado em Ciência da Informação) - Faculdade de Filosofia e Ciências, Universidade Estadual Paulista, Marília, 2024.
-
NUNES, Fabrício Rodrigues do Espirito; MACULAN, Benildes Coura Moreira dos Santos; ALMEIDA, Maurício Barcellos. Os fundamentos da Web Semântica como ferramentas de auxílio para as demandas da Sociedade da Informação. Em Questão, Porto Alegre, v. 26, n. 3, p. 224-249, 2020. DOI: https://doi.org/10.19132/1808-5245263.224-249
» https://doi.org/10.19132/1808-5245263.224-249 -
SEN, Prithviraj et al. Neuro-symbolic inductive logic programming with logical neural networks. Proceedings of the AAAI Conference on Artificial Intelligence, Washington, v. 36, n. 8, p. 8212-8219, 2022. DOI: https://doi.org/10.1609/aaai.v36i8.20795
» https://doi.org/10.1609/aaai.v36i8.20795 -
SYED, Rehan et al. Robotic process automation: contemporary themes and challenges. Computers in Industry, Amsterdam, v. 115, n. art. 103162, 2020. DOI: https://doi.org/10.1016/j.compind.2019.103162
» https://doi.org/10.1016/j.compind.2019.103162 -
TRINDADE, Alessandra Stefane Cândido Elias da; OLIVEIRA, Henry Poncio Cruz de. Inteligência Artificial (IA) generativa e Competência em Informação: habilidades informacionais necessárias ao uso de ferramentas de IA generativa em demandas informacionais de natureza acadêmica-científica. Perspectivas em Ciência da Informação, Belo Horizonte, v. 29, n. art. e-47485, 2024. DOI: https://doi.org/10.1590/1981-5344/47485
» https://doi.org/10.1590/1981-5344/47485 -
SILVA, Patrícia Nascimento. Recuperação de informação na Ciência da Informação: produção acadêmico-científica brasileira (2012-2021). Transinformação, Campinas, v. 35, e237336, 2023. DOI: https://doi.org/10.1590/2318-0889202335e237336
» https://doi.org/10.1590/2318-0889202335e237336 -
ZWASS, Vladimir. Expert system. Encyclopedia Britannica, Chicago, 22 mar. 2024. Disponível em: https://www.britannica.com/technology/expert-system Acesso em: 22 jan. 2025.
» https://www.britannica.com/technology/expert-system
Edited by
-
Editor:
Gildenir Carolino Santos https://orcid.org/0000-0002-4375-6815
Not applicable.
Source: 
Source: 
Source: Prepared by the authors (2025).Description: Figure 3 (Publications on ILP by country): A column chart illustrating scientific production in Inductive Logic Programming (ILP) by nation. China leads by a wide margin (268 publications), followed by the United States and India, with Brazil in tenth place among the analyzed countries.
Source: Prepared by the authors (2025).Description: Figure 4 (Publications on ILP by year): A line graph showing the temporal evolution of ILP publications between 2020 and 2025. It displays a production peak in 2021, followed by stability with a slight downward trend, recording 106 articles by the second half of 2025.
Source : Prepared by the authors (2025).Description: Figure 5 (Publications on RPA by country): A column chart detailing academic production in Robotic Process Automation (RPA). China maintains the lead (41 publications), followed by India and the United States, with Brazil in seventh place, tied with Italy and Spain.
Source: Prepared by the authors (2025).Description: Figure 6 (Publications on RPA by year): A line graph with marked points showing the growth of RPA publications. There is a continuous upward trajectory from 2020 (21 articles) to a peak in 2024 (41 articles), with a growth projection for the end of 2025.
Source: Prepared by the authors (2025).Description: Figure 7 (Publications on ES by country): A column chart with data labels on Expert Systems (ES) production. China stands out with 118 publications, more than double the United States (53) and India (52), while Brazil shows a modest production of 9 papers.
Source : Prepared by the authors (2025).Description: Figure 8 (Publications on ES by year): A line graph showing the annual evolution of ES publications. It reveals a sharp decline after the 2022 peak (110 articles), reaching 67 papers in 2024 and showing signs of stabilization in 2025.
