Addictive GAI Use Among STEM Graduate Students: Dependency-Based Technological Diffusion
Plain Language Summary
This paper reports on a survey of STEM graduate students who use AI chatbots and coding assistants like ChatGPT or Claude to help write code. We found that nearly two-thirds of these students agree that using generative AI (GAI) "is addictive," and that students who feel this way are also more likely to worry about becoming dependent on AI, to second-guess their own abilities, and to hide their AI use from advisors and peers. Because graduate students already face elevated rates of anxiety and depression, we argue this pattern deserves attention, and we offer policy suggestions for how institutions, departments, courses, advisors, and students can respond.
Contribution
Rather than proposing a new addiction framework, we contribute descriptive survey evidence linking self-reported perceptions of GAI addictiveness among STEM graduate student coders to concealment behaviors, dependency worry, and self-judgment, and translate these correlational patterns into a five-level (institutional, departmental, course, advisor, student) policy model specific to the graduate apprenticeship context.
Research Questions
- To what extent do STEM graduate students who use GAI for coding perceive that use as addictive?
- What is the relationship between perceiving GAI use as addictive and other constructs such as dependency worry, self-judgment, confidence, and concealment of GAI use from advisors and peers?
- What policy guidance follows from these patterns for institutions, departments, courses, advisors, and students engaged in CS graduate education?
Methods
We administered a survey to STEM graduate students across six top-ranked US universities between Fall 2025 and Spring 2026, isolating a subgroup of 259 respondents (69.8% of the sample) who reported using GAI to code. The survey combined Likert-scale and short-answer items covering GAI perceptions, social dynamics, identity, and dependency, including the target item "Using GAI is addictive" rated on a 5-point scale. We analyzed responses using descriptive statistics and Spearman correlations, applying a Bonferroni-corrected threshold (α = 0.05/79 = 0.0006) to identify the items most strongly associated with perceived addictiveness.
Key Findings
64.6% of the graduate student coder subgroup (N = 259) scored ≥4 on "Using GAI is addictive" (M = 3.57, Mdn = 4, skew = −0.72), and Kruskal-Wallis tests found no significant differences across demographic categories. Perceived addictiveness correlated most strongly with worry about becoming dependent on GAI (r = .446) and with comparing one's GAI use to peers' (r = .431), and it was also associated, more weakly, with concern that GAI use would be visible in one's work, self-judgment when using GAI, worry about not learning programming concepts deeply, and beliefs that GAI is intelligent or capable of completing coursework. Students who found GAI most addictive were also the most likely to report being careful about how and when they mention GAI use to their advisor.
Implications
We argue these patterns point to unsettled and inconsistently supported norms around GAI use in CS graduate education, particularly given that concealment from advisors threatens the mentorship relationships central to the apprenticeship model of graduate training. We translate this into policy guidance at institutional, departmental, course, advisor, and student levels, emphasizing an ecological, feedback-driven model rather than one-size-fits-all rules, while acknowledging that longitudinal and qualitative follow-up work is needed since our measures rely on students' own sense of "addictive" rather than validated clinical instruments.