Introduction

The fast developments in artificial intelligence (AI) technology have resulted in their growing integration into many spheres, including higher education. Various AI tools are continually being launched for multiple purposes, and even existing ones are being updated or upgraded to increase their efficiency. AI tools are the vast variety of technologies and applications developed utilizing artificial intelligence capabilities to assist individuals with tasks and activities. Some of these tools include Virtual Assistants– AI-powered chatbots or voice assistants capable of understanding natural language, answering inquiries, and doing chores like scheduling, information search, and device control (Mekni, 2021); Predictive Analytics – examines data to find trends, patterns, and generate predictions to help decision-making (Rustagi & Goel, 2022); Computer Vision – examine and decipher visual data such as image identification, object detection, and face recognition (Dhabliya et al., 2023); Natural Language Processors – allows computers to comprehend, interpret, and create human language, therefore enabling tasks such as sentiment analysis, text summarising, and language translating (Khurana et al., 2023); Intelligent Automation – AI-powered software capable of automating repetitive, rule-based tasks (Coombs et al., 2020); Generative AI – based on learnt patterns, creates new content such as text, graphics, audio, and even computer code (Feuerriegel et al., 2024); Intelligent Tutoring Systems – provide students with tailored training, feedback, and adaptive learning experiences (Phobun & Vicheanpanya, 2010); and Recommendation Systems - examine user preferences and behaviour to offer pertinent information, goods, or services (Isinkaye et al., 2015). Several AI-powered features draw attention to the adaptability and broad uses of AI tools in several spheres. However, some AI tools have some cross-functional capabilities. ChatGPT, for instance, may be used for task automation, offers summaries and insights equivalent to predictive analytics, and produces material fit for use in recommendation systems.

In higher institutions of learning, AI not only supports new ways of instruction but also gives a new perspective into the dynamics of instructions and brings transformative changes in evaluation techniques (Ivanov et al., 2024; Tuomi, 2018). As a result, many AI tools have been created and made accessible to support academics and undergraduates in various academic activities (Bubou & Job, 2020; du Boulay, 2016; Masry-Herzallah & Watted, 2024). Several factors influence AI tool use, including perceived usefulness, risk, ease of use, self-efficacy, reliance anxiety, and overall attitude, and these factors are also interconnected (Alanzi et al., 2023; Ayanwale, Frimpong et al., 2024a, Ayanwale, Sanusi, et al., 2024b; Falebita, 2024; Pan et al., 2024; Wang et al., 2021). Moreover, people appear excited and skeptical about the potential effects of new technological developments whenever they appear. This may explain their readiness to accept and use the new tools.

Undergraduate students may find utilizing these cutting-edge AI tools challenging unless they possess the necessary skills. To effectively engage with these applications, students need to demonstrate a certain level of preparedness. In this context, technological readiness refers to undergraduates' skills and willingness to navigate AI-powered tools. An individual's technological readiness, that is, their inclination to welcome and use new technologies, will determine their capacity to properly recognize, comprehend, and use these tools for learning (Bubou & Job, 2020). Furthermore, technological readiness may inform individuals' belief in their capabilities to utilize AI tools. As a factor shown to affect the adoption of AI, technological self-efficacy refers to the trust in one's capability to utilize new AI tools effectively for various reasons. Undergraduates with a high degree of technological self-efficacy are more likely to explore new AI tools as they are released into the public domain (Kong et al., 2021; Obenza et al., 2024; Wang et al., 2023). Similarly, attitude towards technology is another factor considered to influence the usage of AI, as it is explained as the general disposition of an individual towards the usage and adoption of new technologies, which could be positive or negative (Pan, 2020). This study considers the technological attitude as the undergraduate's general disposition, beliefs, and behaviours towards using and adopting AI tools. Studies have proven the presence of an interaction between attitude toward technology and the adoption of AI, which was mostly found to be positive (Adelana et al., 2024; Ayanwale & Ndlovu, 2024; Cai et al., 2017; Kwak et al., 2022; Obenza et al., 2024; Pan, 2020).

The perceptions about the usefulness and how easy it is to use AI tools may shape the attitude towards the use of AI tools possessed by someone. The Technology Acceptance Model (TAM) suggests that perceptions about usefulness and ease of use are considered valuable for accepting new technologies like AI tools (Davis, 1989). Several studies have affirmed that perceptions about usefulness and ease of use are related, emphasizing that a positive and robust relationship exists between them and that the level to which someone believes using technology effortlessly will influence the level of belief that using the technology will improve their performance (Ayanwale & Ndlovu, 2024; Davis, 1989; Mutambara & Chibisa, 2022; Toros et al., 2024).

The Technology Acceptance Model (TAM) has been widely applied to examine the acceptance of AI-based tools across various domains. Studies have utilized TAM in combination with other frameworks to investigate factors influencing AI adoption. Many of these studies have examined the factors contributing to AI adoption in higher education using the theoretical framework. Some of the factors examined include self-efficacy (Obenza et al., 2024; Wang et al., 2021), anxiety (Falebita, 2024; Yin et al., 2023), attitude (Abdaljaleel et al., 2024; Obenza et al., 2024), intrinsic motivation (Lai et al., 2023), self-esteem and stress (Nja et al., 2023). It is important to note that most of these studies examined the interplay between two or more of these factors in determining the adoption of AI. However, none of these studies considered the relationship between self-efficacy, attitude and technological readiness jointly as they relate to the adoption of AI tools.

In many parts of the world, the adoption and integration of emerging technologies, such as artificial intelligence (AI) tools, within the undergraduate learning experience has been uneven, often constrained by a range of factors, including technological readiness, self-efficacy, and attitudes. However, the population from southwestern Nigerian, with its distinct socioeconomic, cultural, and educational characteristics, may yield findings that swerve from those observed in other countries. For instance, the level of technological infrastructure, access to resources, and exposure to AI technologies within Nigerian higher education system may differ significantly from more developed nations, potentially shaping the relationship between these key variables in unique ways. In addition, this study is important within the population of preservice teacher education programs. As the next generation of educators, preservice teachers' technological readiness, self-efficacy, and attitude towards AI tools will shape their future classroom integration. Focusing on preservice teachers in this study offers a unique perspective to inform the preparation of a teaching workforce equipped to navigate the opportunities and challenges of AI in education in the evolving the fifth industrial revolution (Education 5.0).

Investigating the structural links among these elements and their influence on using AI tools among undergraduates is essential. Identifying this interrelatedness may provide educational institutions with vital new perspectives to improve the efficient integration and acceptance of AI-based tools into the instructional process. Accordingly, our study examined how technological readiness (TR), self-efficacy (SE), and attitude (ATT) determine the usage of AI tools (AITU) among undergraduates via perceptions about usefulness (PU) and perception about ease of use (PEU). The conceptual framework representing this study can be found in Fig. 1.

Fig. 1
figure 1

Conceptual model

Full size image

Literature Review and Framework of the Study

Technological Self-Efficacy

Technological self-efficacy is a person's belief in their competence to use technology effectively. This belief has been associated with implementing and using various technological tools in higher education. Previous studies have revealed that university faculty and students who strongly believe in their technological capabilities tend to display more positive attitudes and greater openness toward adopting new technologies. For example, Pan (2020) found that undergraduates' technological self-efficacy influenced their attitudes toward adopting new technologies. Similarly, Kent and Giles (2017) demonstrated that technological self-efficacy was crucial for preservice teachers' integrating technology into their teaching practices. AI tool studies have shown that technological self-efficacy is a crucial determinant. Liwanag and Galicia (2023) noted that greater degrees of technological self-efficacy enabled more effective self-directed learning with technology, extending to the usage of AI tools. This association may also apply to the adoption of AI-powered educational tools, Masry-Herzallah and Watted (2024) showed a favourable association between technological self-efficacy and the adoption of online learning among higher education students. Nonetheless, the body of existing research emphasizes the critical impact of technological self-efficacy on attitudes and behaviours toward the adoption and use of cutting-edge technologies, including AI tools, in higher education institutions (Alanzi et al., 2023; Alharbi & Drew, 2019; Masry-Herzallah & Watted, 2024; Pan, 2020; Sanusi et al., 2024). Hence, in this study, it is proposed that:

  • Ho1: Technological self-efficacy determines the usage of AI tools

  • Ho2: Technological self-efficacy determines the perception of AI tools’ usefulness

  • Ho3: Technological self-efficacy determines the perception of AI tools’ ease of use

  • Ho4: Technological self-efficacy determines attitude toward AI tool usage

Technological Readiness

Technological readiness (TR) is the degree to which an individual is prepared and willing to use and integrate AI tools in various tasks. It is a factor that must be addressed when considering the adoption of AI, as its scope covers developmental, operational, integration, and sustainability readiness (Anh et al., 2024). It refers to an individual's preparedness and willingness to use and integrate specific technologies, such as artificial intelligence (AI) tools, into their work and daily activities. However, TR is often used interchangeably with the broader concept of technology readiness, which is not a user characteristic but relatively the overall level of preparedness and openness of an individual or an organization to adopt and utilize new technologies in general. While the two concepts are related, as an individual's technology readiness may influence their TR towards specific technologies, it is crucial to differentiate between them, as TR is a user-centric characteristic. In contrast, technology readiness is a more general organizational or societal-level attribute. Recognizing this distinction can provide valuable insights into the factors driving the adoption and integration of AI and other innovative technologies. The study by Anh et al. (2024) revealed that TR positively influenced the adoption of AI and perceptions about usefulness and ease of use. On the contrary, Tunmibi and Okuonghae (2023) , who described the concept of technological readiness as an individual's or group's capability, willingness, and resources to leverage and benefit from technology adoption effectively and utilization, revealed that TR does not influence the adoption of AI. Moreover, Labrague et al. (2023) established a connection between technological readiness and perceptions about usefulness, stating that TR predicts the perception of usefulness and the usage of AI tools. Also, Labrague et al. (2023) found that how individuals perceive their technological capabilities, or their "technological self-efficacy," is closely connected to their technological readiness level. Additionally, the attitude demonstrated by an individual towards the usage of AI tools could be predicted by their level of TR (Lazanyi, 2018; Lee & Naidoo, 2018). Since TR is considered a significant factor in predicting AI tool adoption, it is therefore proposed in this study that:

  • Ho5: Technological readiness determines the usage of AI tools

  • Ho6: Technological readiness determines the perception of AI tools’ usefulness

  • Ho7: Technological readiness determines the perception of AI tools’ ease of use

  • Ho8: Technological readiness determines technological self-efficacy

  • Ho9: Technological readiness determines attitude toward AI tool usage

Perceptions about AI Usefulness

Perceptions about usefulness (PU) is an aspect of TAM that focuses on the extent of a person's feelings that adopting a particular system or product would improve their functioning (Davis, 1989). In this study, perceptions about AI usefulness are conceived as undergraduates' feelings of how much utilizing AI tools would improve their performance, productivity, or effectiveness in a specific task or situation. PU is based on the personal assessment of the possible advantages and benefits the AI system may provide (Ayanwale & Ndlovu, 2024; Esiyok et al., 2024). An individual's likelihood of developing a favourable attitude towards an AI tool and adopting the tool increases based on its perceptions of usefulness (Kim et al., 2020). According to studies, user acceptance and adoption of new technologies are much influenced by perceptions about usefulness (Adelana et al., 2024; Das & Madhusudan, 2024; Mutambara, 2023). Similarly, the attitude toward technology usage relates to how they are perceived as useful (Aljarrah et al., 2016; Alzahrani, 2023; Toros et al., 2024). Within AI acceptance, PU is a significant determinant of a person's desire to embrace and use AI technology (Li et al., 2024). This review provides valuable insights into the role of perceived usefulness (PU) in shaping user attitudes and adoption of new technologies, including AI tools. However, the review primarily focuses on the general concept of PU and its influence without delving into the nuances and complexities that may arise within specific technology domains or user populations. While the reviewed studies establish the significance of PU as a determinant of technology acceptance, there is a lack of in-depth exploration of how PU may interact with other individual and technological factors to influence the adoption of AI tools, particularly within the diverse higher education system. Undergraduates in higher institutions use several AI tools for various purposes and these tools include Grammarly, Bard, ChatGPT, Stable diffusion, Turnitin, personalized learning platforms, Generative AI, Dall-E, among others (Chan & Hu, 2023; Chng et al., 2023; Falebita & Kok, 2024). Additionally, the literature review lacks a critical examination of potential inconsistencies or contradictory findings across different studies, limiting the ability to draw comprehensive conclusions about the relationship between PU and AI tool adoption. For this study, it is proposed that:

  • Ho10: Perceptions about usefulness determine attitude towards the adoption of AI tools

  • Ho11: Perceptions about usefulness determine the usage of AI tools

Perceptions about AI Ease of Use

Perceptions about ease of use (PEU) are used in several TAM studies. It is often defined as the perception of an individual on how effortless it is to use technology (Davis, 1989; Dwivedi et al., 2023). In this study, PEU is envisioned as undergraduates' sense, thoughts, or feelings of using AI tools without being stressed or having negative emotions about how they can be used. In examining the connection between PEU and PU, TAM was theorized by Davis (1989). They opined that the direct impact of PEU on PU is because of the tool that can be easily and effectively used. Also, many studies have reaffirmed that PEU significantly influences PU (Labrague et al., 2023; Lestari & Indrasari, 2019; Sanusi et al., 2024) and this also has an impact on the adoption of AI technology (Alanzi et al., 2023; Al-Mughairi & Bhaskar, 2024; Chan & Lee, 2023; Wang et al., 2021). The consistent finding that PEU positively influences PU and that both constructs significantly impact the adoption of AI technologies suggests a strong theoretical basis for understanding user acceptance of innovative tools. However, the review primarily focuses on the general application of TAM without delving into the unique challenges and facilitating factors that may shape the adoption of AI-powered educational tools, specifically within higher education institutions. Other factors, such as psychological and demographic and AI tools' specific features and functionalities, could potentially moderate or alter the dynamics between PEU, PU, and adoption behaviours. In this study, we hypothesize that;

  • Ho12: Perceptions about ease of use determine attitude towards the adoption of AI tools

  • Ho13: Perceptions about ease of use determine the usage of AI tools

  • Ho14: Perceptions about ease of use determine the perception of AI tools’ usefulness

Attitude Towards AI

Attitude is viewed as the thoughts, emotions, reactions, and behaviours undergraduates exhibit toward using AI tools. This attitude can either be a favourable or unfavourable appraisal of how an individual perceives AI tools. Adopting technological tools is significantly impacted by an individual's or intending users' attitudes (Ayanwale, Frimpong et al., 2024a, Ayanwale, Sanusi et al., 2024b; Mutambara & Chibisa, 2022; Toros et al., 2024). Attitudes toward the usage of AI can be mixed, as some individuals embrace its potential to enhance efficiency and innovation in various fields, while others express concerns about ethical implications, job displacement, and the loss of human oversight, leading to complexity in its acceptance and scepticism (Mnguni, 2024). This study proposed that:

  • Ho15: Attitude towards AI determines the usage of AI tools

Methodology

We adopted the cross-sectional survey design for the study. A survey design is essential for collecting reliable and valid data because it effectively gathers meaningful data and insights that can inform decision-making while minimizing potential sources of bias and error (Creswell, 2014). With a survey, we obtained a quantitative-based picture of the undergraduates' technological readiness, self-efficacy, attitude, and usage of AI tools. A questionnaire was used in the study to get opinion-based data from undergraduates. Descriptive statistics were used to analyze all categories and variables in the data, and after that, the proposed model was tested using the partial least squares-structural equation model (PLS-SEM).

Participants

The study participants are undergraduates studying science, technology, and mathematics education from a public university in southwestern Nigeria, specifically from Ekiti State. The undergraduates studying science education have a speciality in agriculture, biology, chemistry, and physics education. The study was introduced to all the science, mathematics, and technology education students, but 176 out of 257 (68.5%) consented, completed and submitted the electronic version of the questionnaire. While participation was voluntary, the researchers made concerted efforts to encourage larger participation, ultimately securing the involvement of 176 students (68.5%), which helped to ensure the sample was more representative of the target population and reduce the risk of skewing the results towards those with a particular interest or experience in technology. Among the participants were 67(38.1%) males and 109(61.9%) females (see Table 1). Regarding age, 85(48.3%) were below the age of 21years, 70(39.8%) were of age 21 – 23years, 16(9.1%) were 24 – 26years and 5(2.3%) were above the age of 26years. On the course of study of the undergraduates, 33(18.8%) were studying agricultural education, 72(40.9%) studying biology education, 23(13.1%) chemistry education, 17(9.7%) mathematics education, 20(11.4%) technology education, and 11(6.3%) were studying physics education.

Table 1 Participating undergraduates’ demographic characteristics
Full size table

Research Instrument

A modified version of the TAM questionnaire was the instrument used for the study. Some items were slightly modified from previous studies, while others were adapted to meet the study’s objectives. Six constructs were considered: technological readiness (TR), self-efficacy (SE), attitude (ATT), perceptions about usefulness (PU), perceptions about ease of use (PEU), and AI usage (AIU). The items of PU and PEU, which have reliability indices of 0.852 and 0.855, respectively, were adapted from Mutambara and Chibisa (2022), while SE, TR, and ATT, which have reliability indices of 0.785, 0.888, and 0.835, respectively, were adapted from Ayanwale, Frimpong et al., 2024a, Ayanwale, Sanusi et al., 2024b). The content was refined by adding and removing certain questions to effectively capture the construct, ensuring that the items were relevant and meaningful to the target population while fitting the study context. Each construct comprised five items developed on a 5-point Likert scale of “strongly agree” to “strongly disagree.” This provided a balanced capturing, meaningful grading, and efficient data collection and analysis (Adelson & McCoach, 2010).

Data Analysis Technique

The PLS-SEM was used to analyze the data and test the research hypotheses. The use of PLS-SEM is well-suited for this study. It is a robust and flexible approach that can handle complex models with latent variables, even in non-normal data and small sample sizes, making it an appropriate choice for the study (Hair et al., 2014). The SmartPLS 4 SEM application was used for the analysis (Ringle et al., 2024). The SEM analysis was carried out in two stages (Hair et al., 2017). During stage 1, we focussed on assessing the measurement model. This involved testing the measurement scales for the reliability and validity of the study's operationalized latent constructs; specifically, the confirmatory composite analysis (CCA) was used in this regard (Hair et al., 2020, 2021). In addition, Cronbach’s Alpha (CA), factor loading, Average Variance Extracted (AVE), Composite Reliability (CR), and Fornell-Larcker Criterion for discriminant validity were also used. The purpose was to ensure that the observed variables accurately reflected the underlying hypothetical concepts they were intended to measure (Hair, Risher et al. 2019a, Hair, Sarstedt et al. 2019b). In stage 2, we assessed the structural model. We specifically examined the hypothesized association and effects among the latent variables. The structural model analysis allowed for the evaluation of the path, significance, and strength of the paths between the various constructs, supplying empirical evidence to either support or not support the proposed hypotheses (Sarstedt et al., 2017). Following this established two-step SEM analytical approach, the psychometric properties of the measurement instruments were confirmed first before proceeding to the testing of the substantive hypothetical relationships specified in their conceptual model. This rigorous analytical strategy is considered best practice in quantitative, theory-testing research utilizing latent variable techniques (Hair et al., 2021).

Results

Descriptive Statistics

Descriptive statistics of construct items are presented in Table 2. This includes the mean and standard deviation scores of the items. Table 2 also summarizes the distributional characteristics of the measured variables of the dataset used in the study. With the analysis, we provide information about the central tendency and dispersion of the observed data (Hair et al., 2017; Hair, Risher et al. 2019a, Hair, Sarstedt et al. 2019b; Tabachnick & Fidell, 2013). Technological self-efficacy was found to be relatively high, with SE1 showing a mean of 4.074 and a low standard deviation of 0.905, which shows that the participants had a steady and high level of technological self-efficacy. Similarly, SE2 had a mean score of 4.091 and a low SD of 0.834, and SE3 had a mean of 4.023 and SD of 0.866, both showing a steady and high level of technological self-efficacy. However, SE4 has a mean of 3.716 and a higher SD of 0.988, and SE5 has a mean of 3.915 and a high SD of 0.922, indicating variability in technological self-efficacy.

Table 2 Descriptive analysis of manifest variable
Full size table

Regarding Technological Readiness (TR), TR2 and TR5 had a mean of 3.989 and 3.886 with high standard deviations of 0.879 and 0.941, respectively, which suggest a slight variation in the technological readiness of the respondents. The remaining items (TR3 and TR4) have high levels of technological readiness but differ in degrees of agreement. For AI perceptions about usefulness, three items (PU1, PU2, and PU3) had higher mean scores and lower standard deviation, demonstrating a higher level of consistency in agreement with the items. However, two items (PU4 and PU5) had lower mean scores with higher standard deviations, indicating more variability and less consistency among the respondents regarding the items. On perceptions about ease of use, all the items (PEU1 to PEU5) had very close high mean scores with a range of 0.079, and the standard deviations are also low, indicating a high degree of homogeneity and consensus among the participants on the measured items. Regarding attitude towards adopting AI, all the items of the construct had relatively high mean scores and moderately low standard deviations, which indicate agreement among the respondents regarding the items of the construct. Lastly, on the usage of AI tools, AITU1 has a high mean score of 3.443, which implies agreement with the item with a moderately low standard deviation, which shows consistency in the agreement with the items; this is also consistent with other items. AITU2 to AITU5. However, when compared with other constructs, the mean scores are lower and the standard deviation higher, indicating that variation in agreement with the items of this construct is higher. These findings demonstrate positive technological self-efficacy, favourable technological readiness, positive perceptions about AI usefulness and perceptions about ease of use, and favourable usage of AI tools among undergraduates. The higher standard deviations on certain items suggest more variability in responses, which emphasizes the discrepancy of viewpoints among respondents.

Confirmatory Composite Analysis

The CA, CR, and AVE were considered for the test reliability and validity of the construct (Adelana et al., 2024; Hair, Risher, et al., 2019a). The factor loading above 0.70 is considered suitable (Hair et al., 2014), and outer loadings below 0.70 are deemed unsuitable. Therefore, some indicators (TR1=0.650 and ATT1=0.698) that had their outer loading below 0.70 were considered unsuitable for measuring the constructs and excluded from the model. The items’ multicollinearity was measured with the Variance Inflation Factor (VIF) approach, values of VIF less than 5.0 are generally acceptable (Hair Jr et al., 2010). Table 3 shows that the VIF values of all the items are less than 5 and range from 1.357 to 2.858 for each item which indicates no collinearity issue among the items. The AVE had high values above 0.50 which ranged between 0.574 and 0.674, demonstrating strong convergent validity and is thus adjured good for the model (Ayanwale, Sanusi, et al., 2024b; Fornell & Larcker, 1981). Also, for CA and CR, which are measures of the reliability and validity of the constructs, Fornell and Larcker (1981) and Hair et al. (2014) suggested a value ≥0.70 to be an acceptable value for measuring the constructs’ internal consistency and the overall consistency of the latent constructs with CA and CR respectively. The six constructs’ CA and CR values shown in Table 3 are all >0.70 which then establishes all constructs’ internal and overall consistency.

Table 3 Confirmatory composite analysis
Full size table

The Discriminant Validity was determined using the Fornell-Larcker Criterion approach. According to Fornell and Larcker (1981), comparing the construct’s square root of the AVE with the highest correlation of any of the constructs, the AVE square root should be greater. This suggests that the concept has discriminant validity as it reveals less variance with its correlated indicators than with other model constructs (Hair et al., 2017). The square root of the AVE of each construct is greater than all the correlations with any other construct (see Table 4). Consequently, this suggests that the latent variables differ from one another, supporting the discriminant validity of the model. Overall, all constructs were shown to be reliable and valid. As a result, the structural model was evaluated.

Table 4 Discriminant validity—Fornell-Larcker criterion
Full size table

Evaluation of Structural Model

The structural model was evaluated using a set of criteria: effect size (f2), predictive relevance (Q2), R2 of endogenous variables, path coefficient size and significance (β & p), and structural model collinearity (VIF). The issue of collinearity was addressed using the VIF, and as per Hair et al. (2021), a VIF value exceeding 5 indicates a potential collinearity issue between the constructs. In our study, the VIF value was consistently below 5, ranging from 1.000 to 3.229 (see Table 5), affirming the reliability of our methods and the absence of collinearity among the latent variables. The path coefficient and its significance, t-statistics, as shown in Table 5, revealed that six out of the 15 hypotheses tested were rejected, while nine hypotheses were accepted. The hypotheses were rejected based on the obtained p-values, which were greater than 0.05 and the t-values less than 1.96, adhering to established criteria for rejecting such hypotheses (Hair et al., 2017).

Table 5 Hypotheses testing
Full size table

We evaluated the model’s explanatory power using the coefficient of determination (R2) and effect size (f2) values. The R-squared (R2) values are used to determine the strength of the relationships between the variables in the model. R2 values of ≥0.19, ≥0.33, and ≥0.67 are commonly interpreted as indicating weak, moderate, and substantial levels of the explained variance, respectively (Hair et al., 2021). In other words, these thresholds guide the magnitude of the relationships between the constructs in the structural model, with higher R2 values suggesting stronger explanatory power. The explanatory power (R2) was used to assess the models, ranging from 0 to 1. Explanatory power increases with R2 value, with values close to 0.50 indicating moderate power and those above 0.50 indicating high power (Adelana et al., 2024). Also, according to Cohen (1988) and Hair et al. (2021), f2 values of ≥0.02, ≥0.15, and ≥0.35 implies small, medium, and large effect sizes, respectively. In this study, the R2 values of SE(0.389), PEU(0.453), PU(0.623), ATT(0.632), and AITU(0.508) are all considered to contribute to the explanatory power of the model moderately. Similarly, it was found in this study that the effect sizes considered very small were for the paths PEU to AITU(0.003), SE to PU (0.013), and TR to ATT (0.001). The paths found with small effect sizes are ATT to AITU (0.061), PU to ATT (0.096), PU to AITU (0.035), SE to ATT (0.027), and SE to AITU (0.041), medium effect sizes SE to PEU (0.179) and PEU to ATT (0.172) while the effect size of PEU to PU (0.507) was considered significant. The Q2 was used to determine the predictive relevance of the model; a Q2 value greater than zero is considered satisfactory in the predictive relevance (Chin, 1998; Mutambara & Chibisa, 2022). The Q2 values of the endogenous variables ATT, AITU, PEU, PU, and SE were 0.286, 0.240, 0.345, 0.350, and 0.382, respectively, more significant than zero. This indicates that all the variables are related and relevant factors determining the usage of AI tools among undergraduates. Furthermore, Figure 2 shows the output structural model generated from the SEM analysis via SmartPLS establishing the relationships between the variables of the study.

Fig. 2
figure 2

Structural model

Full size image

Discussion

The outcome of the SEM analysis revealed the existence of interrelationships among technological readiness, self-efficacy, attitude towards AI tools, perception of AI tools’ usefulness, perception of AI tools’ ease of use, and usage. The analysis showed that technological self-efficacy determined the usage of AI tools. This indicated that the undergraduates’ belief in their ability to use AI tools efficiently strongly predicted their actual usage of AI tools. This finding aligns with Sanusi et al., (2024) and Kwak et al., (2022), who all agreed that self-efficacy in using technologies predicts the usage of AI. Similarly, it was revealed that the perception of AI tools’ ease of use is determined by technological self-efficacy. Self-efficacy, the belief in one’s ability to carry out a specific action, has been linked to influence the usage and acceptance AI tools such as Chatbots(Esiyok et al., 2024). However, it was further revealed from this study that technological self-efficacy does not determine the perception of the AI tools’ usefulness. This supports the findings of Esiyok et al., (2024) , who revealed that self-efficacy does not influence undergraduates’ perceived usefulness of AI chatbots. This finding suggests that undergraduates' belief in their technological capabilities may not necessarily translate to perceiving AI tools as useful. The disconnect between self-efficacy and perceived usefulness highlights the need to consider other factors, such as the specific features and capabilities of the AI tools, as well as the context of use, that may shape undergraduates' perceptions independent of their general technological self-confidence. Also, this study showed that technological self-efficacy does not determine the attitude of undergraduates toward AI tool usage. This indicates that undergraduates’ belief in their ability to use AI tools does not predict their disposition or reactions towards the use of AI tools, which shows that other factors could trigger the attitude of undergraduates towards using AI tools. Against this, Toros et al. (2024) revealed from their model that self-efficacy is a strong predictor of undergraduates’ attitudes toward using new technologies like AI tools.

Furthermore, it was found that technological readiness determined the perception of AI tools’ usefulness. Similarly, it was found that undergraduates’ technological readiness determined their perception of AI tools’ ease of use. In addition, undergraduates' technological readiness level determined their technological self-efficacy. That points to the fact that technological readiness is a viable predictor of the perception of the AI tools’ usefulness and ease of use (Anh et al., 2024) and technological self-efficacy (Labrague et al., 2023). This suggests that developing undergraduates', especially preservice teachers’ technological readiness may be a crucial first step in fostering their self-efficacy and positive perceptions of integrating AI tools. This could, in turn, influence their future classroom practices and ability to leverage these technologies effectively to support student learning. However, the analysis shows that technological readiness does not determine the usage of AI tools. This does not agree with the findings of Ayanwale, Frimpong et al. (2024a), Ayanwale, Sanusi et al. (2024b) and Anh et al. (2024), who revealed from their developed SEM that technological readiness strongly predicted the adoption of AI. Also, the results revealed that technological readiness does not determine attitudes toward AI tool usage. This also contradicts Lazanyi (2018) and Lee and Naidoo (2018), who showed that technological readiness predicted attitudes toward adopting AI. This advocates a better understanding of the complex factors influencing AI adoption in the educational system, highlighting the importance of further investigating other potential mediating and moderating variables that may influence undergraduates' readiness and attitudes towards engaging with and utilising emerging technologies like AI in their learning and future teaching practices.

In addition, the result from the analysis showed that perceptions about usefulness determined attitudes towards adopting AI tools. People who feel that new technologies like AI tools will be beneficial for efficient task completion would have a positive attitude towards such tools. In contrast, those who feel it is useless tend not to have a positive attitude towards their usage. This indicates that perceptions about usefulness predict attitudes toward the usage of AI tools (Chibisa et al., 2022; Kim et al., 2020; Toros et al., 2024). In support of this finding, Nja et al. (2023) showed from their developed model that perceptions about usefulness which was identified as expected benefits have impact on attitudes towards the adoption of AI. However, the analysis showed that perceptions about usefulness do not determine the usage of AI tools. This contradicts the findings of Anh et al. (2024), Chibisa et al. (2022), Kim et al. (2020), and Mutambara and Chibisa (2022), who all revealed that perceptions about usefulness are a good predictor of the adoption of new technologies. In this study, the undergraduates’ perception of the usefulness of new technologies, such as AI tools, does not predict their usage of the AI tools. This is an indication that some moderating variables, such as social influence (Mutambara & Chibisa, 2022) and demographical characteristics (Joseph et al., 2024; Nouraldeen, 2022), though not considered in this study, might be a significant influence.

A further finding of the study revealed that perceptions about ease of use determined attitudes toward the adoption of AI tools. This is in agreement with several studies that have suggested a relationship between perceptions about ease of use and how likely people are to accept new technology (Chibisa et al., 2022; Kim et al., 2020; Toros et al., 2024). Perceptions about ease of use are crucial in technology adoption and utilization, according to the TAM (Davis, 1989). A more favourable attitude towards adopting and utilizing technology is likely to emerge when people regard it as effortless and straightforward to use. This shows that if undergraduates view AI tools as user-friendly and uncomplicated, they are more likely to have a positive inclination toward adopting these technologies. Similarly, perceptions about ease of use were found to determine the perception of AI tools’ usefulness. This is also consistent with TAM, which suggests that perceptions about ease of use predict perceptions about usefulness (Davis, 1989; Kim et al., 2020; Toros et al., 2024). On the contrary, Mutambara and Chibisa (2022) and Chibisa et al. (2022) found from their studies that perceptions about ease of use did not strongly predict the perceptions about the usefulness of new technologies. Additionally, this study’s result showed that perceptions about ease of use did not determine the usage of AI tools. This is contrary to TAM, which suggests that perceptions about ease of use predict the adoption of new technologies (Anh et al., 2024; Kim et al., 2020). This finding also corroborates Mutambara and Chibisa (2022) and Chibisa et al. (2022), who suggested that perceptions about ease of use are not a good predictor of adopting new technologies such as AI tools.

Also, the result of SEM analysis in this study showed that attitude towards AI determines the usage of AI tools. This conforms with the TAM, which suggests that an individual’s attitude toward technology is a crucial predictor of their intention toward actual usage of the technology (Ayanwale, Frimpong et al., 2024a, Ayanwale, Sanusi, et al., 2024b; Chibisa et al., 2022; Kim et al., 2020; Mutambara & Chibisa, 2022). This study showed that for AI tools to be more generally embraced and deployed in higher educational institutions, it is necessary to foster favourable attitudes among students towards these technologies.

Considering the effects of the paths to the development of this study’s SEM, the paths SE -> PU, TR -> ATT, TR -> AITU, and PEU -> AITU have a very small effect on the model. Likewise, the paths ATT -> AITU, PU -> ATT, PU -> AITU, SE -> ATT, SE -> AITU, TR -> PEU, and TR -> PU have effects on the model, but the effect of each path was considered minor. In addition, the effect of paths PEU -> ATT and SE -> PEU were moderate. However, the paths that greatly affect the model are PEU -> PU and TR -> SE.

Our study shows that the undergraduates’ technological self-efficacy is an essential determinant of their perceptions and usage of AI tools. When undergraduates believe in their ability to navigate through AI tools effectively, they are more likely to see them as useful and easy-to-use tools. However, as technological self-efficacy does not directly determine undergraduates' attitudes toward using AI tools, one must keep in mind that self-efficacy may be more closely associated with functional, usability-related perceptions than with higher-level attitudinal variables among undergraduates. Therefore, undergraduates must be provided with targeted training, support, and practical opportunities to build their confidence and comfort in using AI tools since several higher education institutions are regularly integrating adaptive learning tools for instructional activities.

Also, we have demonstrated in this study that undergraduates’ technological readiness is a strong determinant of perceptions on the usage of AI tools but does not strongly determine the actual usage of AI tools. Specifically, the undergraduates’ general openness and inclination towards new technologies do appear to positively influence their perceptions of AI tool usefulness and ease of use. This indicates that technologically ready undergraduates are more likely to perceive AI tools as beneficial and usable. Similarly, as technological readiness positively determines undergraduates’ technological self-efficacy, it implies that a general openness and comfort with AI tools would likely translate into a stronger belief in their technological capabilities. However, as technological readiness does not directly determine the attitude towards and the actual usage of AI tools, undergraduates’ technological readiness alone cannot sufficiently determine their attitude towards and usage of AI tools. To foster broader technological readiness among undergraduates, initiatives that raise awareness and excitement about new AI tools should be implemented by both developers and institutions to promote their usage. Undergraduates should also keep in touch with the realities of the 4IR.

In addition, our model demonstrated that perceptions about usefulness and ease of use strongly determine the undergraduates’ attitudes toward using AI tools. However, their direct impact on the usage of AI tools is unclear. This indicates that the perceptions do not have a significant direct impact on the usage of AI tools but rather on the attitude, which could later translate into actual usage of the AI tools. With the critical role of perceptions on attitude, as shown in our model, it is essential to strategically communicate the benefits of AI tools to undergraduates (users) so that they can develop the right perspectives, which will later influence their attitude towards the usage of AI tools.

Our model also shows that attitude towards AI is a strong determinant of the usage of AI tools among undergraduates. Beyond just perceptions of usefulness and ease of use, this emphasizes the crucial role of attitudinal factors in determining the adoption of AI tools. It is, therefore, important to prioritize proactive measures in understanding and managing undergraduates’ attitudes toward using AI tools. Also, increasing the actual use of AI tools encourages good attitudes about them, which can be done by raising awareness, stressing advantages, and resolving issues relating to AI tools. Factors contributing to establishing positive attitudes, such as perceptions of usefulness, ease of use, and compatibility with learning objectives, should be emphasized in designing and implementing AI-based educational interventions.

Implications

The findings of this study have some important implications for students, educators, and AI tool developers. The findings provide a blueprint for teachers and developers of AI tools to enhance undergraduates' engagement with AI tools. Educators can encourage its usage under well-coordinated guidance rather than forbidding students from using them for various academic tasks. The students should be empowered through AI literacy so that they may gain confidence and competency in using these innovative tools. By strengthening undergraduates' technological self-efficacy through targeted training and support, their innate curiosity and willingness to explore the vast potential of AI tools could be ignited. This could also foster a culture of technological readiness that aids the seamless navigation of AI tools. When developers strategically design AI tools so that they are perceived as useful and easy to use, this could reshape undergraduates' attitudes, transforming initial apprehension into enthusiastic adoption. Notably, the interrelated nature of these factors emphasizes the need for a holistic, coordinated approach to helping undergraduates utilize AI tools effectively. Educators and developers should collaborate to address technological readiness, self-efficacy, perceptions, and attitudes cohesively. It is then that undergraduates and educators can unlock the true potential of AI within higher education, empowering the next generation of innovators and problem-solvers.

Limitations and Future Studies

The study's focus on a single public university and a specific academic department may limit the generalizability of the findings. Collecting data from only one university department could potentially overlook variations in technological readiness, self-efficacy, and AI tool perceptions that may exist across different disciplines and academic programs. This narrowed scope raises questions about whether the observed relationships and patterns are representative of the broader undergraduate population or specific to the chosen STEM department. Also, undergraduates from other academic backgrounds may exhibit divergent opinions and behaviours concerning integrating AI tools. This disciplinary variable could mediate the effects of the constructs examined in the structural model. To enhance the robustness and applicability of the findings, we suggest that future research expand the study to include a more diverse sample across multiple higher education institutions and academic departments. This would allow for a more comprehensive understanding of the phenomenon and enable the identification of potential disciplinary or institutional differences. Additionally, a larger overall sample size should be considered for future studies. This would strengthen the statistical power of the analysis and increase the confidence in the generalizability of the results, potentially uncovering the degree of the relationships or patterns that the current sample limitations may have concealed. In summary, longitudinal studies should be considered for future studies to examine changes in undergraduates' attitudes and usage of AI tools over time, or comparative analyses should focus on different demographic groups, academic disciplines, or types of higher education institutions to uncover potential issues and provide more actionable insights to guide the effective integration of AI tools within preservice teacher education and other undergraduate programs.