Attitudes towards AI

There are broad ethical concerns regarding the sourcing of the information fed into generative AI databases. At its simplest level, information from a myriad sources is fed to an AI database to ‘teach’ the AI model. This information often generates an average from existing information to create new output based on a request, like a new image being generated based on existing images, or a new song being generated that sounds similar to previous songs. Apart from a few generative AI databases that ethically source the materials used to ‘teach’ their AI, authors and artists largely do often do not provide consent or receive compensation for their works to be harvested by the AI platform. This questionable sourcing of information has led to multiple lawsuits and concerns about the true originality of AI output in regards to written works, visual imagery/artwork, and music [4–6]. The rampant appearance of AI-generated content across many facets of life, especially without seeking it out, has been called ‘AI slop’ [7]. As dialogue about the use of AI in everyday life has increased, individuals have begun to develop attitudes towards AI. Fietta et al. [8] described how explicit and implicit attitudes towards AI often differ and demonstrated that women generally have more negative attitudes towards AI compared to men. Research [9,10] has supported that individuals who scored high in the areas of introversion, conscientiousness, and agreeableness tend to have more positive views on AI use, suggesting that personality traits may contribute to AI attitudes. Positive attitudes towards AI among those high in introversion might suggest that individuals who prefer to work in isolation might be more likely to utilize AI as a tool. Among psychology students, Gado et al. [11] found that the intention to use AI could be predicted by attitudes, perceived usefulness, social norms, and perceived knowledge about AI. The lack of understanding about AI contributes to increasing social tension about its advancements with arguments being made on both extremes: for and against AI use. Arguments endorsing AI suggest its use as benefiting humanity through automation of menial or dangerous work. Opposing that, arguments that condemn its use usually cite economic disparity resulting from automation of formerly human-performed jobs [12,13]. These arguments may extend to the use of generative AI as a tool for future scholars who might use AI to structure, write, and edit manuscripts that are submitted to journals like this one, possibly raising concerns of sterility and monotony in the writing style of future academics. Some have even claimed that AI may replace the roles of teachers and doctors [14]. Attitudes towards AI and their associated underlying personality traits may play a role in explaining different types of AI use [15,16].

Generative AI in academics

The role of AI in academic honesty is a major concern among academics [17]. Some concern may be warranted, as students have been found to understand blatant unethical use of AI in academics, but more subtle uses of AI that contribute to academic dishonesty may be misunderstood [18]. Most current generative AI models will often prioritize task completion over task accuracy, resulting in output that ‘hallucinates’ information that is not real or factually accurate. These ‘hallucinations’ [19] can include citing sources that do not exist, or listing real authors and journals while providing fabricated information. This results in not only plagiarism, but also AI contributing to the spread of misinformation, which students might not realize that they’re doing. A study by Murray and Williams [20] found that an individual student possesses multiple different ethical perspectives concerning the use of generative AI, rather than a single prominent perspective. Perceived ease of use and usefulness influenced student acceptance of AI systems in academic settings [21]. At the time of this research, there was a lack of studies exploring how students are using AI in academia, with only some preprints beginning to emerge [22]. Additionally, uncovering the individual qualities or situations that may mitigate or promote the use of AI need investigation.

With the rise of AI and its impact on the academic field and education, educators may find themselves questioning, evaluating, and adjusting their approaches to teaching. Fears about AI have led some teachers to adopt more reactive approaches to teaching such as implementing oral assessments to prevent AI use [23]. Others proposed proactive use by having students use generative AI to construct papers and become aware of the underlying reasons for plagiarism through workshops [24]. The focus on the negative potential of generative AI in academic writing means that the positive potential for ChatGPT and other AI platforms to enhance success and participation rates of some students is largely overlooked [25]. The role of AI in academic dishonesty dominates current arguments, but there is also discussion concerning the use of AI as a tool to reach learning outcomes [20]. For example, AI can be used to improve personalized instruction, increase students’ ability to learn independently, improve learning efficiency, and help students determine their skills and prospective employment opportunities [26]. There is an increasing push for universities to develop new policies, procedures, and regulations for their students’ use of AI in order to fully consider risks and rewards, as well as attempt to detect and prevent academic dishonesty [22]. AI may also begin to change the roles of both students and teachers [27], as perceived learning goals and different routes to obtain them emerge.

Current AI arguments overlook crucial components needed to understand changes in student behavior: when and why generative AI is being used, and what consequences that use has for the user. In academics, unethical use of AI may be consistent with previous research about cheating motivations. Intrinsic motivation has been found to have an inverse relationship with academic dishonesty [28–30]. Orosz et al. [30] found no relationship between extrinsic motivation and academic dishonesty, which contrasted work by Jordan [31] who found that cheating was related to extrinsic motivation and perceiving cheating as socially normative. Cheating has been found to have a negative relationship with self-efficacy as demonstrated by Nora & Zhang [32]. Eshun et al. [33] found that personality traits and self-efficacy had conflicting relationships with academic dishonesty. Clinciu et al. [34] took a different approach by looking at specific extrinsic pressures, and how they influence academic dishonesty. ‘Results pressure’ was defined as a student’s high interest in performance and grades, maintaining a scholarship or grant, finding a good job, and hypercompetitive behavior. ‘External pressure’ encompasses a low probability of being caught or penalized for fraud, pressure to cheat from their peer group, and/or as a response to uninteresting course content or a dislike of the teacher. Both results and external pressure were found to be positively related to academic dishonesty, so those were utilized in the current study.

Generative AI, body image concerns, and social media

Another growing area of generative AI utilization is on social networking platforms as a tool used in online interactions. Social media users can use AI to meet strategic self-presentation goals and help maintain social connections. Strategic self-presentation includes the intentional misrepresentation of the self for the purpose of displaying a preferred self-image on social media platforms [35]. AI can be applied to an image of the self to create completely new images including fictitious backgrounds, clothing, and body augmentations that do not represent the individual in real life. Image filters similarly do this [36], and allow the user to cover up blemishes, add makeup, or emphasize qualities that exaggerate them in ways that can tailor their features towards cultural beauty standards. It has been suggested that there is a greater need to understand the use of strategic self-presentation in the context of social media [37]. In response to this need, and in addition to the AI use in academia, our study looked at the motivational and psychological variables of AI use as they relate to ethically dubious uses of AI in online interactions, including digitally manipulating photos of the self, and more specifically, using AI to alter or fabricate images of the self.

Impression management is a major motive for being involved in social networking websites [38], and individuals that are highly approval-motivated are more likely to digitally edit their photos [35]. A high emphasis on self-presentation in a social media context has been associated with more mental health struggles and reduced quality of life [39]. For example, frequent use of social media websites where there is a greater perceived need for strategic self-presentation has been associated with lower levels of self-esteem and life satisfaction [39] as a result of an increased dependence on social approval [40]. In contrast, utilizing an honest self-presentation strategy has been associated with greater self-esteem [41]. A meta-analysis of 80 studies found that more social networking site (SNS) use was positively related to narcissism and loneliness, and negatively related to self-esteem [42]. Saiphoo et al. [43] conducted a meta-analytical review of studies researching the relationship between the use of SNSs and self-esteem, and they found that lower self-esteem was associated with increased SNS use.

A large part of addressing self-presentation goals comes from body image concerns. Saiphoo and Vahedi [44] conducted a meta-analytical review concerning the relationship between social media use and body image. Body image is defined as a complex, subjective perception of an individual’s physical appearance and attractiveness, including visual perceptions, behaviors, and attitudes towards the body [45,46]. Saiphoo and Vahedi [44] found that there was a small, positive relationship between social media use and body image disturbance due to the frequency in which social media users engage in appearance-based comparisons with societally-idealized bodies or aesthetics, such as being thin or muscular. Selfie editing was associated with negative self-evaluations [47]. The negative associations between body image and self-esteem were more pronounced for women than men [48]. Self-objectification has been positively associated with using photo editing to meet strategic self-presentation needs on social media [35]. Gender differences have been found, with women editing photos more frequently, resulting in more negative affect following upward social comparisons compared to men [49].

The role of AI in strategic self-presentation is just one use of AI in a social media context. Another use is meme creation. Digital memes are images or videos with variations of text that are typically used in a humorous manner [50], though usually, memes are based on a singular visual photo. There are several functions of memes, which include communicating social and political beliefs, unifying collective identities, and developing culture [50]. They can make ideas more compact, providing images in a shareable, easily processed format. A study by Huntington [51] found that a high level of agreement with the message communicated in a political meme results in decreased scrutiny of the message, and increased perceived effectiveness of the message. The acceptance of certain memes is dependent on whether the viewer relates to attitudes in the meme. Memes can be used to increase social-relatedness on social media through humorous and/or relatable images, gifs, and videos [52]. Memes are also used as a coping mechanism when experiencing stress [53–56] and in coping with depression [56]. Research by Maclean et al. [57] suggested that image sharing may have beneficial effects for most Instagram users, as individuals high in social connectedness reported lower levels of loneliness when frequently sharing images.

The current study

Six research objectives were pursued. The first was to see if individuals that are using generative AI in unethical ways were aware of the misuse (i.e., if they knew what constituted unethical use and use it that way anyway), or if the unethical use was due to a lack of knowledge about AI. The second was to explore motivational correlates (intrinsic motivation, extrinsic motivation, results pressure, external pressure, time pressure, a desire for popularity, personality) related to potentially dishonest or ethically dubious behaviors across academics (academic dishonesty, writing dishonesty, AI plagiarism, using AI to cheat in academics) and online interactions (digitally manipulating photos of the self, using AI to create new images of the self). The third was an exploratory investigation into how the aforementioned behaviors relate to each other (e.g., Are those being dishonest in their academic endeavors also being dishonest in their online interactions?), which would provide evidence about unethical uses being domain-specific or found across domains. The fourth looked at how the dishonest or ethically dubious behaviors related to consequential psychological variables including self-esteem, self-efficacy, satisfaction with life, loneliness, objectification of the self, body image concerns, narcissism, and personality traits. The fifth goal was to explain the variance in dishonest behaviors using the motivational variables that were positively related. Finally, the sixth goal was to explore the role of creating memes with and without AI, and to see how frequency of using social media websites and meme creation relates to psychological and motivation variables. Each objective will contribute to the overall goal of understanding if unethical uses of AI are life domain-specific or applied across life domains, what related motivations for those uses are, and what psychological consequences are associated with those uses.

Participants

Participants were recruited through posting advertisements in American college and university social media groups, and through postings on public bulletin boards across the University of South Carolina Beaufort (USCB) campus from February to May 2024. Some USCB students received extra credit points for their completion of the survey. Participants outside of the United States were removed from the sample, leaving a total of 185 participants. 59.5% of the sample identified as female, 37.3% as male, and 3.2% as nonbinary or genderfluid. Average age was 31.72 (SD = 11.99). Most of the sample were white/ Caucasian (78.4%), followed by black (8.1%), Latino (4.9%), Asian (4.3%), Native American (1.6%), or other (2.7%). Most were single and never married (53.5%), followed by married or living as (42.2%), separated/ divorced (3.8%), or widowed (.5%). Education levels differed with most having attended some college (25.4%) or had obtained a bachelor’s degree (31.9%). 41.6% identified as first-generation students. 51.9% attended college and came from a family where their parents also attended college. 69.7% identified themselves as middle economic class, followed by lower economic class (22.7%), and upper economic class (7.6%).

Materials

Cronbach’s Alpha was used to qualify the reliability of each scale used. Measures of AI attitudes, literacy, and ethics were included to gauge the relationship between unethical uses of AI and the knowledge of and the knowledge that those uses are unethical. Attitudes towards AI were measured using the AI Attitude Scale (AIS-4) [58], a 4-item measure that is used to gauge participants’ agreement with to what extent AI will improve their life, their work, be utilized in technology that they will use, and will be positive for humanity. It was selected based on its generalizability across diverse populations [58]. The AIS-4 was found to have strong reliability (α = .89). Understanding when using AI constitutes academic dishonesty was measured with the 11 ‘AI-giarism’ items developed by Chan [18]. The scale showed strong reliability (α = .88) and had participants state to what extent they agree or disagree that a series of different uses is academic misconduct (e.g., “The student input a prompt into an AI system, copied the generated response, and submitted it to the teacher.”). Chan [18] described ‘AI-giarism’ as “the unethical practice of using artificial intelligence technology, particularly generative language models, to generate content that is plagiarized either from original human-authored work or directly from AI-generated content, without appropriate acknowledgement of the original sources or AI’s contribution” [18, p. 4]. The artificial intelligence literacy scale [59] was used to measure four dimensions of AI literacy: awareness, use, evaluation, and ethics. The measure consists of 12 questions that participants self-report the extent to which each statement describes their literacy with different aspects of AI. Awareness including familiarity with AI (e.g., “I can identify the AI technology employed in the applications and products I use.”), knowing how to use AI effectively to complete tasks (e.g., “I can skillfully use AI applications or products to help me with my daily work.”), evaluating which AI to use for a task and what its limitations are (e.g., “I can choose the most appropriate AI application or product from a variety for a particular task”), and familiarity with ethical questions surrounding AI (e.g., “I am always alert to the abuse of AI technology”). The awareness, use, and ethics subscales demonstrated unsatisfactory reliability, but the AI evaluation subscale demonstrated satisfactory reliability (α = .77). The AL literacy scale was not used for analyses due to its overall low reliability scores.

Section III of the Academic Ethics Questionnaire [34] was used to measure external and results pressure, which consists of 23 items each scored on a 5- point Likert-type scale. Each item asked participants to rate the extent to which each scenario would influence them to engage in dishonest academic behavior (e.g., “You needed to get a good grade”, “You were pressured by other students”, “You do not like the instructor”). Results pressure (α = .89) and external pressure (α = .89) both had strong reliability.

Intrinsic and extrinsic motivation was measured using the 13-item Intrinsic Motivation and Extrinsic Motivation scale (IM-EM) [60]. The IM-EM scale had participants rate the extent to which a series of statements about motivation described them (e.g., “I prefer activities that challenge my knowledge or abilities”). Reliability was strong for intrinsic motivation (α = .85) and moderate for extrinsic motivation (α = .68).

Time pressure was measured by asking participants how frequently they use artificial intelligence to save time on a task. The desire for likes/reactions on social media posts was measured by asking participants how frequently this statement described them: “I hope that my photos, videos, and/or memes get many likes/reactions.” The importance of the popularity (virality) of social media posts was measured by asking how frequently this statement described the participant: “It means a lot to me when my photos, videos, and/or memes go viral.” Participants responded with ‘never’, ‘sometimes’, or ‘frequently’ for each question.

Section II of the Academic Ethics Questionnaire [34] was used to measure academic dishonesty and writing dishonesty, which consists of 22 items. Each item asked participants how frequently they performed each behavior during their time in school or university (e.g., “Copy someone else’s assignment”). The 13-item academic dishonesty measure (α = .95) and 5-item writing dishonesty measure (α = .89) had strong reliability. Three additional items were asked from Bashir & Bala [61] that encompassed additional forms of academic dishonesty (e.g., “Pay someone else to complete homework or an exam for me”), plus three additional items about AI use (i.e., “Use artificial intelligence on an assignment without saying that you used AI”, “Use AI to write a paper”, “Use AI to cheat on an exam or assignment”).

Specific use of photo manipulation techniques for online and mobile media was measured using the Self Photo Manipulation Scale (SPM) [62]. The SPM scale is a 10-item, self-report measure with strong reliability (α = .89). Participants were asked how frequently they used various photo manipulation techniques for self-imagery shared online and through mobile devices (e.g., “Get rid of red eye”; “Whiten your teeth”). Participants were also asked to what extent they “use AI to create new images of the self” and “use AI to create a dating profile.”

Psychological variables that were measured included self-esteem, self-efficacy, satisfaction with life, loneliness, body image concern variables, narcissism, and personality traits. Self-esteem was measured using the Rosenberg Self-Esteem Scale (RSE) [63]. The RSE is a 10-item, self-report measure. Participants rated to what extent they agreed that the series of statements described the self (e.g., “I feel that I have a number of qualities”; “I wish I could have more respect for myself.”). The RSE was found to have strong reliability (α = .88).

The General Self-Efficacy Scale (GSE) [64] was used to measure self-efficacy. The GSE is a consists of ten item scale with strong reliability (α = .90) that measure the extent to which an individual believes they can perform novel or difficult tasks and cope with adversity across multiple domains of human functioning. It reflects goal-setting, persistence in the face of obstacles, and recovery from setbacks.

Satisfaction with life was measured using the 5-item Satisfaction with Life Scale [65]. Participants responded to items stating to what extent five descriptions accurately described how they felt about life (e.g., “In most ways my life is close to my ideal”, “I am satisfied with my life”). The measure had strong reliability (α = .89).

The 8-item version of the UCLA Loneliness scale (ULS-8) [66] was used to measure participants’ sensation of being separated from others. Participants responded to the items with the frequency in which they experienced similar feelings (e.g., “I lack companionship”; “I feel left out”). It had strong reliability (α = .82).

Four measures were used to evaluate social and physical comparisons, including objectification of the self. Frequency of making social comparisons was measured using the Iowa-Netherlands Comparison Orientation Measure (INCOM) [67], which had strong reliability (α = .83). The INCOM is an 11-item, self-report measure. Participants indicated to what extent they agreed with a series of statements concerning their experience with social comparisons to other people (e.g., “I often compare myself with others with respect to what I have accomplished in life.”).

The valuation of physical appearance was measured using the Self-Objectification Beliefs and Behaviors Scale (SOBB) [68]. The SOBB scale is a 14-item, self-report measure consisting of two factors. Participants rated to what extent they agreed the series of statements best described their valuation of their physical appearance (e.g., “How I look is more important to me than how I think or feel”; “I often think of how my body must look to others.”). The SOBB consists of two factors. The first factor contains 7 items concerning the internalization of the observer’s perspective of the body (e.g., “I choose specific clothing or accessories based on how they make my body appear to others.”), which had strong reliability (α = .91). The second included 7-items about prioritizing and valuing the self through physical appearance (e.g., “Looking attractive to others is more important to me than being happy with who I am inside.”), which also had strong reliability (α = .94). Participants rated to what extent they agreed the series of statements best described their valuation of their physical appearance (e.g., “How I look is more important to me than how I think or feel”; “I often think of how my body must look to others.”).

The tendency to compare one’s physical appearance to others was measured using the Physical Appearance Comparison Scale (PAC) [69]. The PAC scale is a 5-item, self-report measure. Participants rated to what extent they agreed the series of statements described their experiences concerning physical comparison to others (e.g., “At parties or other social events, I compare my physical appearance to the physical appearance of others.”). The PAC had satisfactory reliability (α = .73).

The 10-item version of the Big Five Factor Personality Inventory (BFI-10) [70] was used to measure personality as it has been found to be as predictively capable as the 44-item BFI [71]. None of the big five personality factors had satisfactory reliability scores, so the measure was not included in the analyses. Narcissism was measured using the Narcissism subscale of the Short Dark Triad [72]. The Narcissism subscale is a 13-item, self-report measure. Participants rated to what extent they agreed that a series of statements, concerning self-importance and self-admiration, were true (e.g., “People see me as a natural leader”; “I insist on getting the respect I deserve.”). It had satisfactory reliability (α = .73).

Six additional questions were included to ask about the frequency of specific behaviors related to daily social network use (i.e., how many hours per day) and frequency of meme-related behaviors (i.e., “create memes without AI”, “create memes with AI”, “post memes online”, “use memes to improve your friendships”, “use memes to improve your relationships”).

Procedure

The survey was completed by participants online after clicking on an invitation to complete the survey through social networking advertisements or by scanning a QR code on a poster. Following an informed consent page, participants completed items about the frequency of artificial intelligence behaviors. All other measures were presented in randomized order to reduce the possibility of order effects. Participants completed demographic items and were then debriefed.