AI recommendations’ impact on individual and social practices of Generation Z on social media: a comparative analysis between Estonia, Italy, and the Netherlands

1 Introduction

This research investigates how mediated communication on Social media (SM) shapes the practices and behaviors of their users, based on a case study conducted in Italy, Estonia, and the Netherlands. Mediated communication using digital devices among users has become a common practice (Eurostat 2019a, 2020a; Digital Education Action Plan 2018–2020). In recent years, we have witnessed the rise of algorithms providing social value (Janssen and Kuk 2016), which follow, measure, and evaluate the data related to individuals and provide customized recommendations. Moreover, SM has become more important after the various restrictions settled by governments in 2020–2021 due to the COVID SARS-19 pandemic (Suárez-González et al. 2021), limiting physical contact between individuals. SMs, like Facebook, Instagram, TikTok, and others, allow users to communicate by creating, exchanging, and interacting with texts (Baruah 2012; Bhandari and Bimo 2022). In this case, the notion of text is assumed as a broader term, based on the definition of Lotman (1992), here used to identify digital representations in the form of verbal texts in natural languages (English, Italian, Estonian, Dutch, etc.) as well as non-verbal representations, such as images, video, and audio. Most SM use Artificial Intelligence Recommendation Systems (AiRS) to sort, order, and display the most relevant texts to users based on previously collected data about their users’ activities (Benrouba and Boudour 2023; Kim et al. 2023), often based on collaborative filtering methods (Zhang et al. 2022).

This research addresses the gap in the research on AiRS and investigates how users’ learning process is shaped by the affordances SM creates. AiRS are essential in helping people navigate the vast volume of texts published and created on SM and other digital platforms (Wu et al. 2019). As a result, AiRS play a significant role in how our society operates today, relying increasingly on digital tools for assisted decision-making. AiRS, which supply highly personalized content, services, and texts, are responsible for creating most digital experiences. They can, therefore, be viewed as an automated decision-making process that aims to resemble the user’s human reasoning closely. Several lines of research in academia and business are investigating AiRS, primarily concentrating on creating algorithmic structures (Duan et al. 2022; Kang and Lou 2022; Mariani et al. 2022; Sharma and Shafiq 2022), overlooking their effects on users. This study introduces a comparative text analysis as a method to map the practices among young adults based on their connection to regular interaction with AiRS-mediated texts in digital environments like SM.

Digital representations are the texts that AiRS manipulate to meet SM users’ needs to increase interactions. Users may receive various posts from AiRS that affect how they perceive the context of their surroundings. The uniqueness of human perception of digital images seen on SM rests in the ability to classify them in meaningful ways similar to how one would classify physically experienced objects, prompting responses accordingly (Damasio 1999; Hodzic et al. 2009; Zink et al. 2008). Following data from Eurostat (2019a), approximately 97 % of young adults aged 18–29 use SM. Young adults with basic or above basic digital skills are extensively involved in mediated communication for an average of 2 h per day on SM and other content-sharing digital platforms (Eurostat 2020a).

This research intends to develop a data-driven approach to the users’ learning process in AiRS-shaped digital environments based on theoretical discoveries and ideas about how users interact with SM through their portable devices. A commonly used top–down strategy that offers users pre-selected categories in a big data-based approach (Tinati et al. 2014) is constrained in its ability to give profound insights to engaged agents, e.g., designers, users or businesses (Grover et al. 2022; Hancock et al. 2020). With the use of big data, it is possible to quantify only a part of the interaction process while entirely ignoring behaviors and practices that are not quantifiably stated online, not to mention how these interactions influence behaviors in other digital and physical environments (Keles et al. 2020). To overcome the limitations of the big data approach, we used semi-structured interviews as the data collection method. This research used qualitative methods to generalize and model the practices that emerge on personal and social levels in digital natives. This research followed a bottom-up approach to SM practices and their reflection in social and cultural dimensions. The chosen methodology is based on data collected from the users instead of top–down methods set on design-centered pre-selected categories that impact users’ interpretations and practices.

This research employs semiotics to explore the impact of AiRS on the cognitive and decision-making processes of the user and cultural value creation. These ideas are explored through a comparative study of content-sharing social networking services (SM) used by Italian, Estonian, and Dutch residents. This research aims to find how the use of AiRS can influence its users. Specifically, it focuses on users’ decision process in the context of interactions on SM as a tool for mediated communication, including leading digital platforms such as Instagram, YouTube, and TikTok, and others. This research examines users’ knowledge and interpretations, personal opinions and evaluations, and emotions. This research proposes a semiotic approach based on Uexküll’s (1972) works, applying them to the complexity of collecting data about human perception. In his practice, Thure Von Uexküll refers to a duality of self-descriptive qualities of interviewees’ speech, on the one hand describing their own conscious and unconscious experiences and feelings and their reasoning, and on the other hand, thanks to the description process, identifying own reaction through the conscious cognitive process of categorization. In this way, semi-structured interviews, as a descriptive method, can shed light on users’ needs, experiences, and affordances they find within AiRS-mediated digital environments. Affordances are functional qualities of an object within given environments that users can adopt based on their needs (Gibson 1977; Ingold 2009). Affordances of AiRS for SM users are mainly measured within the big data approach based on the indicators defined by designers. Still, users’ perspectives and emerging practices can be outside its scope. Users find their affordances based on their personal, social, and cultural needs. Repetition of the practices based on the AiRS’ affordances within SM can foster a scaffolding process, impacting daily learning.

At the core, this research aims to explore the role of AiRS in scaffolding the practices young adults use in their communication. AiRS used daily are connected to users by data they are trained on and influence their interpretations, actions, practices, and behaviors during the interactions (online) and after (offline). It adopts an interdisciplinary perspective connecting several branches of semiotics, cultural and cognitive, and cultural studies related to AI and digital transformation. First, at the methodological level, this research investigates the possibilities of employing semi-structured interviews as a data collection method, translating the users’ experiences into cultural texts. Second, at the analytical level, it aims to explore the correlations between the design of AiRS and their effect on users in particular sociocultural contexts, using semiotic tools for textual analysis. Based on semi-structured interviews collected in 2022–2023, with a sample group of 30 young adults aged 18–26 years from Italy, Estonia, and the Netherlands, this study aims to map the affordances and emerging practices of SM users. Ten participants were chosen from each country, respecting gender and social balance. Participants could express their experiences and reflection on SM use in 45–60 min interviews, which touched on their digital device background, their experience with SM and digitally mediated communication, and finally, their interactions with AiRS. The interviews were carefully transcribed to create a selection of texts for further semiotics text analysis. The methodology for text analysis is based on the isotopies as textual elements (Eco 1979, 1984; Greimas 1966: 96) and later analyzed according to intertextual connections, following theory on the semiosphere (Lotman 2005), a model of culture explained as a self-referential system.

Finally, this research proposes descriptive tools for comparative analysis of AiRS-mediated communication practices from individual perception to social and cultural levels. The main findings of this work provide an empirical analysis of emerging practices among young adults, which are highly shaped by AiRS used on SM.

2 Research settings

This research investigates SM practices among young adults aged between 18 and 26 years, also identified in the literature as Generation Z and digital natives. In cultural studies, digital natives are often categorized as a generation born after the introduction of digital technologies, which some scholars identify as the end of the 1980s (Verčič and Verčič 2013). However, digital technologies used by communities may vary significantly based on various criteria, including accessibility and frequency of use. Therefore, in this research, digital natives are considered as users of digital devices with access to the Internet in the age before and during adolescence. This specific age category is chosen due to specific aspects of human brain development (Lusk 2010). This brings a specific historical point of the development in HCI, marking a possibility for individuals to interact instantly with other human and non-human agents through the texts mediated by digital platforms. Therefore, a significant change in technology as a matter of practice can be connected to the introduction of search engines (e.g., Yahoo, Google) and digital platforms for creating, exchanging, and sharing various digital representations as texts (Myspace, Orkut, Hyves, Tumbler, etc.), and wearable devices which can be highly personalized with access to the Internet (e.g., iPhone smartphone by Apple). The common accessibility to this type of wearable device marks a specific historical timeline for individuals born between 1996 and 2010 (Dolot 2018; Turner 2015). This division is not solely based on the age difference but rather on the context in which individuals had grown and which shaped their experiences influencing the practices with a digital medium. In our work, we define Generation Z as a part of society with access to digital devices, e.g., personal computers and wearable devices, at an early age. They also could access SM in early adolescence, which might influence their communication practices.

A significant scope of literature dedicated to digital natives and Generation Z addresses educational specificities and their role in the innovation of learning and work environments. Their practices significantly impact the communication and identity representation change on digital platforms (Correa 2016; Williams et al. 2012). Importantly, people between the ages of 18 and 26 years experienced two years of restrictions against the SARS-19 coronavirus restrictions, which unavoidably affected their social life and communication practices more than any other previous generation due to the specific period of personal development in high school, college or early career relationships. Digital environments, including SM, mediated all these fundamental European young adults’ experiences. Therefore, the use of AiRS on these digital platforms might have affected Generation Z users’ cognitive, axiological, emotional, pragmatic, and other dimensions.

Data on digital skills can highlight possible affordances that digital environments present to their users (Correa 2016). However, in constantly changing environments, such as AiRS-mediated SM, users are reinforced continuously to learn and adapt to adequately regulate their perception of digital stimuli and express contextually coherent behaviors. Today’s significant scope of digital experiences is created by AiRS, delivering highly personalized content, services, and products (Zhou et al. 2012). They can be thus seen as an automated decision-making process aiming to be close to the human reasoning of the user. Many scholars have studied AiRS, mainly focusing on developing AI algorithms. In contrast, this research uses a semiotic approach to explore how AiRS influence human decision-making and interpretative processes. This approach focuses on the users’ cognitive, axiological, emotional and pragmatic dimensions to overcome the limitations of a narrow big data-based approach and provide the main cultural insights on AI-driven practices online and offline. Semiotics of culture can explore the impact of AiRS on the cognitive and decision-making processes of the user and cultural value creation.

Easy access to the Internet has a significant impact on cultural integrity and the practices of groups not only around Europe but around the world. Accessing the same information simultaneously can be categorized as factual news and opinions. The Internet has provided a platform for most people to share their opinions, and SM granted free and easy access to these opinions. Accordingly, AiRS of SM made it possible to valorize messages shared as opinions to become culturally valuable, sometimes overcoming the messages related to representing the facts. Finally, using AiRS on SM may increase social value by augmenting the contextual connection among algorithmic categories of the digital representation of facts and opinions.

To provide a sufficient dataset highlighting similarities and differences among users of SM and their AI-mediated communication practices, Italy, Estonia, and the Netherlands were selected as case studies. These three EU countries proposed various policies, including digital literacy, institutional level of familiarization with digital competencies, attitudes towards SM and statistical data on Internet users and daily time spent online. All three countries selected for this research defined equal access of the users to facilities, such as Internet connection, digital devices, and relatively similar cultural backgrounds in social structures and communication. Italy, Estonia, and the Netherlands are the countries of the European Union located in one economic zone with equal access to digital resources and norms and standards for evaluating the variables. Each country has a unique position regarding digital literacy, institutional level of familiarizing residents with digital competencies, basic or above basic digital skills among individuals and online social engagement practices (e.g., Ferrari et al. 2012). These three datasets (Eurostat 2019a, 2019b, 2020a, 2020b; Digital Education Action Plan 2018–2020; OECD 2019) were used to understand through a comparative analysis how AiRS influence human decision-making and interpretation processes within different social contexts (Table 1).

The report for the EU Commission in the chapter “Graduate Outcomes” (Digital Education Action Plan 2018–2020: 32) states that high school graduates have higher skills in SM and digital literacy than university graduates. It suggests that there is a difference in the practices used by these two categories in the digital sphere. According to data provided by Eurostat (2020a), there is a significant gap in SM use between the age categories of 16–24 and 65–74 among those who use the Internet with the purpose of posting messages to SM sites or instant messaging, participating in social networks by creating a user profile, posting messages or other contributions, participating in social or professional networks, uploading self-created content to any website to be shared: in Estonia (65 % of the population involved overall, 94 % in the age category 16–24 y.o., 24 % in the age category 65–74 y.o.), in the Netherlands (71 % of the population involved overall, 92 % in age category 16–24 y.o., 43 % in the age category 65–74 y.o), and in Italy (48 % of the population involved overall, 79 % in age category 16–24 y.o., 15 % in the age category 65–74 y.o.). Therefore, it is possible to conclude that most SM users are young adults aged 26–27 and younger (by 2023) who spend a significant amount of time daily on SM. Comparative analysis of data collected in Italy, Estonia, and the Netherlands can provide valuable insight into how culture-specific patterns in communication might be reflected within digital platforms.

3 Data collection and analysis methodology: aiming for a holistic approach to users’ experience in AiRS-mediated environments

4 Comparative analysis: analysis of young adults’ practices in social media environments

5 Findings and discussion: generalizations on social and cultural levels

Correspondence between AiRS categories and human-user categories is a dual learning process in which, on the one hand, algorithms are intentionally improved frequently via machine learning, while, on the other hand, users engaging with AI agents in digital environments adapt in various ways. Therefore, algorithmically high-valued categories can impact how central values are formed within a culture by shaping users’ practices.

The isotopies indicated from the interviews point to common effects noted by the users. All of them are highly connected and can be based on perception and affordances that users may find within digital environments. This way, AiRS’ use of SM induces Generation Z users to experience stress due to social pressure, adapting to standardization of representations and identity, and employing AiRS to their own needs by aiming to emotional regulation by purposeful stimulation, even creating coping mechanisms to deal with misinformation. Users 18–26 years old have a high awareness of the echo chamber and filter bubbles effect (Terren and Borge-Bravo 2021; Wolfowicz et al. 2021) when an internet user is exposed only to the selected information, which corresponds to a narrow topic and often aligns and to amplify user’s own views and opinions (Guess et al. 2018). They purposely attempt to shape digital environments and stimuli to create a desired distorted world, shaping their perception of it. Young adults claim to “train algorithms” to show only desired messages. To achieve this goal, they may carefully select the AiRS, using the function “do not show anymore,” and intuitively search for prospective messages that interest them. The experience of stress and exhaustion when the environment is not treated mindfully, the effects of being exposed to unexpected information can have adverse effects. As a solution, some mention a process of training algorithms, purposely searching for the relevant digital representation to create a digital environment that should correspond to desired results. In this case, young people do not look for AiRS to show them what they like but rather prefer to create an environment that would bring them to their desired future version of themselves. This way, some stated that they would be following a person they are not interested in as an individual but from whom they would like to learn certain skills. This way, the highlighted isotopies suggest that young people aged 18–26 years old feel social pressure to acquire new skills through digital platforms as a result of social pressure mediated by AiRS.

AI is acknowledged and accepted as an element of the environment. Most Generation Z users who participated in the interviews stated that they accept the data collection process about their activities online as unavoidable. However, they prefer not to overshare personal information online, feeling exposed and vulnerable. Even the representation of their own face can be considered a potential threat that can be used in the future. Young adults are open to AiRS used on SM, but they feel the need to control it with the tools they have. These tools are based on own perception of a digital environment. The commonly used answer “I go on Social media just to distract myself from something” in most cases implied to the situations of physical surroundings when respondents experienced emotions and feelings marked as socially negative or unpleasant, e.g., boredom (“… when I wait in a line,” “… while in the bus”), anxiety (“… before the exam,” “… during self-isolation period”), anger, sadness, loneliness (“… I need a community of people who share my beliefs”). This way, individuals may use AiRS as a tool to impact their own identity and perception of themselves in the digital environment. Furthermore, as a feeling of identity is holistic, it necessarily affects general attitudes about oneself.

Generalization analysis of highlighted isotopies into general practices on social and cultural levels underlines the connection of the processes, where one is necessarily interdependent with another. In this research, we hightailed the following general social practices uniting them into a complementary model of operative categories between AiRS and users, as presented in Table 7.

6 Conclusions and further research

This research contributes in various ways to the literature on AiRS and social practices on SM. This research highlights the contextual significance of AiRS as a mediator in creating new communication practices from individual perception and finds that the interpretation and decision-making results in changes on the social and cultural levels.

First, this research used a bottom-up approach through semi-structured interviews as a data collection method and qualitative analysis to map practices that incorporate online and offline behaviors. On the theoretical level, this research contributes a piece of knowledge on how the textualization process of SM users can be interpreted with respect to AiRS as mediating agents in communication. On the practical level, this research proposes tools for qualitative analysis by employing a semiotics methodology that allows the modeling of complex communication processes on SM.

Second, this study compares Generation Z practices on SM in Italy, Estonia, and the Netherlands. The differences in digital literacy suggested by statistical data (Eurostat 2019a, 2019b) did not find a significant correlation with the data collected during interviews. The interviews, translated into text and analyzed through the isotopy theory, highlight homogeneous findings. These findings helped to map repetitive meaningful elements in users’ self-description process. Meaningful elements identified four main practices that directly relate to affordances AiRS provide on SM: 1) stress avoidance from social pressure, 2) adaptation to standardization of representation, 3) emotional regulation via digital environments, and 4) emotional and physical detachment from digital representations.

Finally, this research suggests social and cultural interpretations of how AiRS used on SM influence Generation Z users. By spending significant time online, young people learn about their identity based on what they see in texts of digital environments and adapt to AiRS-mediated narratives. They often choose practices that would enhance their digital representations according to AiRS’ dominant patterns. Mediation of communication practices on SM leads the 18–26 year-old age category to have unified practices learned from AiRS, affecting their embodied interpretations.

This research proposes an exploratory mapping of AiRS’ impact on SM users and suggests that further research is needed. The communication process on SM is complex. It includes several involved agents, designers, users, and AiRS, which impact each other in various ways. To improve AiRS as a tool and allow it to serve better, it is crucial to see it from the users’ perspective. Further studies can extend findings on cognitive and bodily changes in SM users by creating a bigger scale approach, including other age groups, and adding other data analysis methods.