Generative AI Use and Ethical Perceptions among University Students in Brazil and Canada

Description

This exploratory mixed-methods study examines how university students in Brazil and Canada perceive and use generative artificial intelligence (GenAI), with particular attention to the functional, social, and ethical dimensions of adoption. The Unified Theory of Acceptance and Use of Technology (UTAUT) and Perceived Ethical Legitimacy (PEL) were used as organizing frameworks to guide the analysis of students’ experiences and perceptions. The study addressed two main questions: how students perceive the impact of GenAI on learning and whether these perceptions differ between the Brazilian and Canadian contexts; and what patterns emerge in students’ ethical reasoning about GenAI use and how these patterns relate to their reported use practices. The dataset contains coded information derived from questionnaire responses provided by university students in Brazil and Canada. The questionnaire was administered in Portuguese in Brazil and in English in Canada, with minor contextual adaptations to country-specific questions, such as those concerning the participant’s first language. The instrument included questions about GenAI use and frequency, ease of use, perceived academic impact, applications in academic work, ethical and equity concerns, social influences, and personal reflections about GenAI. Open-ended responses were coded by one researcher with AI assistance using a predefined coding framework, and the coding was subsequently verified by an independent researcher. Multiple codes could be assigned when a response expressed more than one relevant idea. The publicly available dataset contains these coded representations rather than the students’ original written responses, which are withheld to protect participant privacy. The findings indicate high levels of GenAI adoption, but comparatively limited sustained integration into students’ academic work and substantial ethical ambiguity surrounding its use. Brazilian and Canadian students showed broad similarities, although differences emerged in areas such as degree of integration, perceived effects on grades, peer influence, and the framing of fairness and equity. The analysis also identified recurring tensions between the perceived benefits of GenAI and concerns about ethical use, between continued use and personal discomfort, and between social encouragement and individual hesitation. The coded data can therefore be used to examine the prevalence and co-occurrence of different perceptions and reported behaviors, compare patterns across the Brazilian and Canadian samples, and explore relationships among functional, social, and ethical dimensions of GenAI adoption. The codes should be interpreted as researchers’ categorizations of participants’ responses rather than as the original responses themselves.

Steps to reproduce

The data were collected as part of an exploratory mixed-methods study investigating university students’ use and perceptions of generative artificial intelligence (GenAI) in Brazil and Canada. An online questionnaire comprising 21 closed- and open-ended items was developed to capture students’ experiences, patterns of use, perceived academic impacts, social influences, usability perceptions, and ethical considerations. The instrument was informed by the Unified Theory of Acceptance and Use of Technology (UTAUT) and Perceived Ethical Legitimacy (PEL), which served as organizing frameworks for data collection and analysis. Participants first completed an online consent form and then responded anonymously to the questionnaire through a secure web link. The questionnaire was administered in Portuguese in Brazil and in English in Canada, with minor contextual adaptations between the two country versions. The final sample comprised 308 university students, including 100 participants from Brazil and 208 from Canada. Because slightly different versions of the questionnaire were used during data collection in Canada, some questions were not administered to a subset of participants; the corresponding values are therefore structurally missing and should not be interpreted as participant non-response. Open-ended responses were coded using a theory-informed codebook based on a framework analysis approach. A large language model was used to assist the initial assignment of codes. The coding procedure was developed iteratively: an initial sample of 30 responses was used to evaluate the coding rules, the rules were refined to reduce misclassification and ambiguous classifications, and the revised framework was then applied to the full dataset. Multiple codes were permitted when a response meaningfully reflected more than one theme, whereas items representing a single categorical response were assigned a single primary code. The AI-assisted coding was supervised by one researcher, who reviewed the generated assignments and retained responsibility for the final coding decisions; the resulting coding was subsequently verified by an independent researcher. The published dataset contains the resulting coded information rather than participants’ original free-text responses, which are not shared to protect participant privacy. The coded data can be used to reproduce descriptive distributions, cross-national comparisons, cross-tabulations, chi-square tests of association, and Cramer’s V effect-size calculations reported in the associated study.

Institutions

• University of Toronto

  Ontario, Toronto
• Universidade Federal do Rio Grande do Sul

  Rio Grande do Sul, Porto Alegre

Categories

Licence