Artificial Intelligence and Desirable Futures: Visions of Scientific Researchers in Mexico

Inteligencia artificial y futuros deseables: visiones de investigadores científicos en México

Gabriela Elisa Sued gabriela.sued@iimas.unam.mx

Universidad Nacional Autónoma de México, México

Esta obra está bajo una Licencia Creative Commons Atribución-NoComercial-CompartirIgual 4.0 Internacional.

Although Artificial Intelligence (ai) originated more than six decades ago, it has only become widely diffused in the past decade. First emerging in the academic sphere in the 1950s, ai was initially dominated by a symbolic approach that sought to replicate human cognition through logical and mathematical rules. However, this approach was later questioned for being too fragmented. In the 1980s, the connectionist paradigm replaced it. This new approach emphasized learning and self-organization through neural networks inspired by the human brain (Weber & Prietl, 2021). Over the last ten years, ai has been defined by its ability to learn from large amounts of data and perform complex tasks autonomously (Kaplan & Haenlein, 2019).

Due to its ability to address challenges in various fields, including finance, healthcare, insurance, and climate change, learning-based ai has reshaped computer science by creating a new sociotechnical system that is still in the process of stabilizing. Within this evolving context, institutional and individual visions of ai’s future play a key role in guiding technoscientific development, social applications, and public policy. From the perspective of Science, Technology, and Social (sts) studies, such visions are not neutral forecasts, but rather social construction of meaning that orient collective action. The formation of sociotechnical imaginaries—shared visions of desirable futures—contributes to the stabilization of emerging technologies and legitimizes scientific interventions in social issues (Jasanoff & Kim, 2009). However, these imaginaries vary according to the actors and contexts involved in technological development.

In countries like Mexico, where ai has not yet been fully stabilized, visions of its future differ greatly from the dominant global corporate narratives. Large technology firms have played a pivotal role in shaping global visions of ai (Richter et al., 2023). As machine learning became associated with large-scale data collection, which is often linked to privacy concerns and discriminatory outcomes, companies decided to reframe ai in 2015 as the application of statistical models, automated classification, and prediction based on pattern recognition. This shift emphasized autonomous task solving rather than massive data use. These learning models were quickly adopted across commercial sectors, including finance, insurance, and advertising (Elish & boyd, 2018). By 2016, these models had expanded into areas such as urban planning, healthcare, environmental management, and sustainability. They were promoted under the banner of “ai for social good” (Hager et al., 2017).

By 2022, learning-based systems had evolved into foundational models— ai systems trained on vast amounts of unstructured data, usually via self-supervised learning. After minimal fine-tuning, these models can serve as adaptable bases for a wide range of downstream tasks applicable to diverse areas, thereby transforming ai research and development (Bommasani et al., 2022). Conversational chatbots, such as ChatGPT, derive from these foundational models and are optimized for natural language interaction. However, only a select few companies, collectively known as Big Tech, possess the technical expertise, computational infrastructure, and massive datasets necessary to develop such models, which results in high economic concentration and limited global competition (Nayak & Walton, 2024). These innovations have fueled social visions centered on enhanced productivity, routine work reduced or eliminated, and an inevitable technological transformation that increasingly shapes governmental sociotechnical imaginaries (Bareis & Katzenbach, 2022; Wang & Downey, 2025).

Unlike in the Global North, where ai development is concentrated in large technology corporations, public research centers in Latin America play a leading role in knowledge production and training specialized personnel (Red Iberoamericana de Indicadores de Ciencia y Tecnología [Ibero-American Network of Science and Technology Indicators], 2023). In the region, ai is still in its formative stages and has yet to be fully stabilized as a technology. In Mexico, regulatory standards have yet to be established, and efforts to promote public policies for innovation and technological development are in their infancy. Although systematic ai research has been identified (Chávez & Vizuete-Sandoval, 2025; Sued, 2024), little is known about its specific characteristics or the perspectives that guide it.

This study connects the importance of public ai research centers with the necessity of identifying their interpretive dimensions within the local context. The objective is to characterize the diverse and cross-cutting views of ai held in Mexico. The article posits that these views, while sharing certain common features, differ according to the type of research—basic or applied—being conducted and the institutional trajectories of the researchers’ affiliated organizations, which vary in history and purpose. To empirically examine this hypothesis, the study analyzes twenty-five interviews with researchers from two public institutions dedicated to computing research.

The paper is structured as follows. First, it reviews analytical tools from sts studies that address the discursive and narrative dimensions of scientific practice (Hilgartner, 2015; Jasanoff, 2015; Jasanoff & Kim, 2009; Kreimer, 2023). These tools are understood as meaning-making processes that shape and motivate research. Next, the methodology is presented, emphasizing a qualitative, interpretive approach grounded in researchers’ perspectives. The findings section outlines the range of views on ai; it begins with a set of common perspectives among interviewees, followed by divergent visions linked to cognitive and institutional factors that influence scientific practice. The discussion draws on Hilgartner’s (2015) concept of cutting-edge visions and identifies two main imaginaries: one centered on high-quality autonomous science (in a minority of cases), and another emphasizing the social utility of knowledge production (in most cases). The paper concludes that heterogeneous visions of ai coexist in Mexico yet converge toward niche imaginaries fostered by limited competitiveness with large-scale ais.

Several analytical tools have been developed in sts studies to examine visions of the future. One of the most prominent is the concept of sociotechnical imaginaries (Jasanoff & Kim, 2009), which illuminates how societies envision technological futures intertwined with governance dynamics. These sociotechnical imaginaries are collective, institutionally stabilized, and publicly articulated representations of desirable futures that can be achieved through scientific and technological advancement and vary across sociocultural contexts. For example, the United States has historically viewed nuclear energy as risky technology that requires strict state control, whereas South Korea has seen it as a driver of national modernization and economic growth (Jasanoff, 2015; Jasanoff & Kim, 2009).

The value of sociotechnical imaginaries lies in their focus on their social understanding and adaptation rather than the technical specificity of technologies. Unlike narratives derived from historical trajectories, these imaginaries are forward-looking, projecting visions of what is good, valuable, and desirable for a community. They are also tied to the exercise of power through the prioritization of development goals, allocation of funding, investment in infrastructure, and acceptance or suppression of political dissent (Jasanoff & Kim, 2009).

Vanguard visions can help us better understand the emergent and contested nature of visions that have yet to achieve institutional consolidation as a sociotechnical imaginary (Hilgartner, 2015). This concept bridges micro- and macrosocial levels by highlighting small collectives whose non-hegemonic projections of the future negotiate, compete with, and align to established dominant imaginaries. For instance, synthetic biology researchers in the United States articulate their visions of future to a broader national sociotechnical imaginary of innovation that encompasses entrepreneurship and iconic technological developments. Unlike stable and long-lasting imaginaries, avant-garde visions are heterogeneous and transient. They often reflect the early phases of a technological development process, when its meanings are still open to negotiation. Once adopted by a broader community, these visions can evolve into the category of imaginaries (Jasanoff, 2015).

As the findings will show, researchers often connect their work to the notion of the social utility of knowledge, a concept primarily linked to market engagement in the 1990s that, in the 21st century, shifted toward principles of development, equity, and sustainability (Bortz et al., 2022). Within this framework, techno-scientific promises play a central role in shaping desirable futures. By projecting prospective technological solutions into present research agendas, scientists can mobilize resources, build strategic alliances, and redefine their methodological approaches (Kreimer, 2023).

The study adopted a qualitative research method emphasizing the recording, categorization, and interpretive analysis of 25 semi-structured interviews with full-time researchers from two leading public university computing research centers in Mexico. Both institutions are recognized as pioneers in the field but differ in their orientation and research objectives.

Center A, established in the 1970s, encompasses several applied science departments, including the Computer Science Department, from which the interviewees hail. Center A prioritizes academic research and values publication in indexed journals as a key performance indicator. Center B, founded in the 1990s as a center of excellence in Computer Engineering, conducts research and provides teaching and knowledge transfer across various social sectors. Designed to achieve social impact, Center B emphasizes problem-solving, patent and utility model registration, and the creation of collaborative innovation networks. The center hosts thirteen laboratories organized into three research areas, one of which focuses on ai.

Of the 25 interviewees, 18 are men and 7 are women. Engineers represent the majority (17), followed by graduates in Computer Science or related fields (6) and Mathematics (2). All the interviewees hold doctoral degrees, most in Computer Science or related disciplines (16), while the rest hold degrees in Engineering (5), Cognitive Sciences and ai (2), and Mathematics (2). Fifteen researchers are primarily engaged in applied knowledge production, testing their own or third-party algorithms with various datasets. Five focus on basic research, developing new algorithms or optimizing existing ones, and another five combine both approaches. Their ai subfields were diverse, with pattern recognition, neural networks, machine learning, machine vision, and natural language processing (being less prominent). The most common area of automated machine learning technique application was medical data analysis (15 researchers), followed by environmental issues, natural language processing applications, and energy efficiency.

The interviews were conducted between May 2024 and April 2025 at the participants’ workplaces. Interviewees were selected based on institutional profiles, registration as specialists in bibliographic databases, and the snowball method, that is, referrals from other interviewees. Each session lasted between one and two hours and covered topics such as the participants’ academic backgrounds, training, areas of expertise, organizational practices, collaborative networks, and research focus. Participants were also asked about their perspectives on the future of ai in Mexico, including its potential, challenges, and limitations.

All recordings were transcribed and thematically coded using Taguette software (Rampin, 2018). Several themes emerged in the initial coding stage, including researcher training, academic collaboration, interest in ai, data acquisition processes, evaluation and funding mechanisms, and perceptions of the current and future landscape of ai in Mexico. References to technoscientific promises and the imaginary of social utility were primarily associated with interest in ai and data acquisition methods. These methods were often linked to health, environmental, or community-oriented projects rather than commercial applications. Visions of the future were identified in responses about Mexico’s ai development prospects and strengths. These emergent codes were consistently associated with the study’s core variables—the type of research (basic or applied) and institutional affiliation—which facilitated further refinement of the article’s central hypothesis.

The interpretation was validated using two main criteria. First was internal consistency of narrative, which was examined by looking at how similar patterns emerged across interviews and ai subfields. Second was external coherence with the analytical approach, ensuring that the empirically derived categories aligned with the theoretical framework without distorting their meaning. This validation process supported a situated interpretive analysis through which visions of the future were understood as discursive constructions shaping scientific practice while revealing structural constraints and potential actionable horizons.

The analytical frameworks highlight the variety of perspectives, especially during phases when technologies are still becoming stable. Among the interviewees, this diversity stems from the different motivations and positions associated with each ai subfield. These differences highlight the importance of situating their sociotechnical perspectives within their specific research contexts. Factors such as areas of application of developments, strategies for obtaining training data, research outputs, methodological orientations (basic or applied), and institutional affiliation contribute to shaping their views on ai.

Despite this diversity, shared perspectives emerge around certain themes. Unlike media and corporate imaginaries that portray ai as a disruptive force capable of radically transforming society, economy, and labor markets (Wang & Downey, 2025), researchers adopt a more cautious and reflective stance. While acknowledging ai’s importance, they refrain from attributing exceptional or immediate transformative powers to it.

This section identifies two main types of perspectives. The first type encompasses views common to all participants and focuses on general-purpose ai, its social applications, and the niche strategies researchers use to avoid direct competition due to limited large-scale resources. The second type comprises heterogeneous perspectives organized into two categories: (1) differences shaped by institutional affiliation, which influence how researchers envision ai and potential applications in the future in Mexico, and (2) distinctions based on research orientation—whether basic (typically aligned with logical-mathematical paradigms) or applied (emphasizing connectionist models).

Grounded in tools from the social studies of science and technology, this analysis examined the motivations and projections regarding ai among Mexican academic researchers. In a context of interpretive flexibility driven by the absence of public policies and strategic visions, the perspectives of local researchers are crucial for stabilizing ai and creating desirable futures. This is manifested through proposals, regulations, and potential policy design (Jasanoff, 2015). These perspectives are situated constructions shaped by social factors, such as research trajectories, experiences, and institutional affiliation, as well as cognitive factors, including the ai subfield, methodological orientation (basic or applied), and technical practices.

From a regional standpoint, the Ibero-American Network of Science and Technology Indicators (2023) notes that ai in Latin America remains in a constitutive phase dominated by public research centers and lacking stable regulatory frameworks—a situation mirrored in Mexico. Similarly, Chávez and Vizuete-Sandoval (2025) emphasize that ai capacity building in the region often perpetuates technological and data dependence. This is evident in local researchers’ difficulties accessing infrastructure, acquiring datasets, and competing with large-scale global models. Sued’s (2024) bibliometric study of Mexican ai production emphasizes applications oriented toward social utility, complementing the study’s qualitative evidence and reinforcing the importance of strategies that focus on the common good.

In this context, Mexican researchers agree on using specific-purpose ai solutions to address local problems, particularly those related to health, the environment, and energy efficiency. This strategy allows them to avoid direct competition with large-scale models. These discursive constructions serve as technoscientific promises (Kreimer, 2023) that legitimize research projects and guide daily practices by articulating motivations and objectives.

However, the visions identified in the interviews lack institutional stabilization, which is a characteristic of fully developed sociotechnical imaginaries (Jasanoff & Kim, 2009). They are not embedded in formal documents or public policies, but rather, they reflect individual, non-hegemonic, and emergent avant-garde visions (Hilgartner, 2015). Rooted in the potential social utility of knowledge or significant academic contributions (Nowotny et al., 2003), these cutting-edge perspectives contrast with corporate imaginaries, which prioritize the increase of productivity and the inevitability of technological implementation (Wang & Downey, 2025).

This study contributes to a deeper understanding of meaning-making processes in sociotechnical systems by highlighting the situated and context-dependent nature of ai research in Mexico. By differentiating between global and local visions, the paper provides insights that can inform the design of public policies and strategic plans that are grounded in local practices, capabilities, and strengths and oriented toward the common good and the social utility of knowledge.

The study reveals the coexistence of various sociotechnical perspectives among Mexican ai researchers. Participants generally express ambivalence toward general-purpose ai, emphasizing specific, context-sensitive developments as niche strategies that align with the common good and the availability of local data. These perspectives vary according to researcher profiles and institutional contexts. Those engaged in basic science value academic autonomy and the production of fundamental knowledge. In contrast, researchers focused on applied projects emphasize social utility. Center A favors interdisciplinary collaborations for applied purposes, while Center B emphasizes ai’s potential for scientific and technological development, as well as the necessity of greater national funding and coordinated efforts.

These findings challenge corporate and media narratives that portray ai as inherently disruptive. They also expose the tensions within Mexican and Latin American scientific fields, such as those between global scaling and local problem-solving and between theoretical research and technology transfer.

Based on this evidence, the study outlines science policy recommendations aimed at enhancing projections and applicability. In applied research, strengthening collaboration among researchers, health professionals, engineers, local communities, and government agencies is critical to transform academic prototypes into effective, socially impactful solutions. Simultaneously, supporting basic science through low-cost computing infrastructure and solutions that reduce reliance on specialized hardware can sustain local innovation. Together with a national strategy for coordination and funding, these actions can boost technology transfer and the production of fundamental knowledge. This will narrow the gap with global corporations and foster an ai ecosystem that is responsive to national needs and capacities.

This study makes three key contributions: it identifies alternative visions to dominant corporate ai narratives, critically examines the concepts that guide researchers’ motivations and discourse, and provides empirical evidence to inform context-sensitive, socially oriented science policy. However, the qualitative design and sample, which includes leading ai researchers, limit the generalizability of the results to the broader academic community in this field.

A notable challenge was the reluctance of some researchers to expose their work to social science analysis. This restricts access to information, thereby hindering the development of studies with greater validity and scope. This resistance underscores the tension between scientific practice and its social approach and highlights the need to foster trust to expand research horizons. Future studies may identify the cutting-edge visions of researchers affiliated with institutions with diverse profiles or in other Latin American countries to enable comparative analyses of ai visions and motivations.

Understanding future visions of ai at public research centers paves the way for a research agenda rooted in practice, where technological developments respond to real possibilities, local contexts, and social needs. In order to achieve ai with a positive social impact, local capabilities must be strengthened, and non-hegemonic ways of envisioning the future of ai must be cultivated.

How to reference: Sued, G. E. (2025). Artificial Intelligence and Desirable Futures: Visions of Scientific Researchers in Mexico. Trilogía Ciencia Tecnología Sociedad, 17(36), e3568. https://doi.org/10.22430/21457778.3568

redalyc-journal-id: 5343

https://revistas.itm.edu.co/index.php/trilogia/article/view/3568 (html)

The author gratefully acknowledges the support of the Secretariat of Science, Humanities, Technology, and Innovation (SECIHTI) in preparing this manuscript through a postdoctoral fellowship. Special thanks are also extended to the researchers who generously gave their time and willingness to participate in the interview process.