Student-ai Collaboration Audit Reveals Preferences across 12 Academic Tasks

The growing integration of generative artificial intelligence into university coursework is rapidly changing the landscape of academic work, prompting questions about student engagement and the quality of learning. Nifu Dan from Georgia Institute of Technology, alongside colleagues, investigates how postgraduate computer science students are currently collaborating with these powerful tools. Their research presents a detailed audit of student preferences, examining the balance between the convenience of automation and the need for genuine academic agency. This study is significant because it moves beyond simple usage statistics to explore the nuanced perceptions of benefits, risks, and desired boundaries, ultimately informing the design of more effective and trustworthy AI systems for higher education. Through a mixed-methods approach utilising sequential surveys, the team identifies critical gaps between current AI capabilities and students’ expectations for collaborative learning.

As generative AI becomes increasingly embedded in higher education, it significantly shapes how students complete academic tasks. While these systems offer efficiency and support, concerns persist regarding over-automation, diminished student agency, and the potential for unreliable or hallucinated outputs. This study conducts a mixed-method investigation into student-AI collaboration, focusing on online graduate Computer Science students. The research aims to understand the patterns of AI usage, the nature of student interaction with these tools, and the impact on learning outcomes and academic integrity. Through analysis of code submissions, student surveys, and interview data, the study provides insights into the evolving dynamics of student-AI collaboration in a higher education context.

Student AI Collaboration Preferences in Higher Education

As generative AI tools become increasingly prevalent in higher education, researchers undertook a mixed-methods study to audit student collaboration preferences with these systems. The work focused on understanding the alignment between current AI capabilities and the desired levels of automation expressed by students completing academic tasks. Scientists recruited participants from the Georgia Tech Online Master of Science in Computer Science program to investigate this evolving relationship. The study employed a sequential, two-phase survey design to capture detailed insights into student perceptions, risks, and boundaries when utilising AI.

The initial survey leveraged an established task-based framework, assessing student preferences for and actual usage of AI across a spectrum of twelve distinct academic tasks. These tasks encompassed activities such as reading, writing, coding, studying, collaborative work, and assessment, providing a comprehensive overview of AI integration. Alongside usage data, the survey gathered primary concerns and motivations driving student adoption of these tools. This quantitative data informed the design of the subsequent qualitative phase, allowing researchers to delve deeper into specific areas of interest.

The second survey moved beyond simple preference measurement, utilising open-ended questions to explore how AI systems could be designed to address identified concerns. This phase sought to pinpoint specific system-level features, including transparency, confidence indicators, explainability, hallucination warnings, and pedagogical alignment, that would foster greater trust in AI within an educational context. Researchers analysed these qualitative responses to identify key themes and patterns in student expectations. The approach enables a nuanced understanding of how students envision trustworthy human-AI collaboration.

This study pioneered a methodology connecting quantitative measures of AI usage with qualitative explorations of desired system characteristics. By mapping tasks into alignment zones, Green Light, R&D Opportunity, Low Priority, and Red Light, based on desire and usage, the research team created a detailed, person-centered view of student engagement. The integration of the Human Agency Scale alongside the survey data provided a robust framework for analysing the interplay between automation and student autonomy, ultimately informing the development of more effective and trustworthy AI systems for education.

Students’ Calibrated Approach to AI Assistance

Recent research has revealed nuanced student preferences regarding the integration of generative AI into higher education. Scientists achieved a detailed understanding of how students negotiate the balance between utilising AI assistance and maintaining their own agency in academic work. The study employed sequential surveys to capture perceptions of benefits, risks, and desired levels of automation across twelve distinct academic tasks. Experiments revealed that students do not universally seek maximal automation, instead demonstrating a calibrated approach to AI interaction dependent on the specific task at hand.

Data shows a clear correlation between task characteristics and student motivations for using AI. Efficiency gains, such as saving time and reducing cognitive load, were primary drivers, but these were applied selectively. For writing, revision, and learning support, AI was largely viewed as a pragmatic assistant, with minimal expressed concerns due to the inspectable and editable nature of the outputs. Conversely, ideation and technical problem-solving tasks triggered distinct concern profiles, with students prioritising intellectual ownership and accurate information. Measurements confirm that concerns weren’t abstract fears, but grounded assessments of how AI assistance aligned with task goals and learning outcomes.

The research team measured student expectations for AI system design, finding a consistent call for features supporting verification, reflection, and skepticism. Transparency and verifiability emerged as dominant expectations, with students repeatedly requesting source citations and traceable evidence, particularly for tasks demanding correctness. Participants explicitly rejected persuasive presentation styles, instead favouring systems that visibly communicate uncertainty through confidence scores or reliability indicators. Students also expressed interest in explainable AI systems that expose reasoning steps and allow for user intervention, reinforcing their desire to remain active decision-makers.

Synthesizing these findings, the study demonstrates that students desire calibrated assistance, not maximal automation, from educational AI. The research identified a gap between current AI affordances and students’ normative expectations, highlighting the need for designs that prioritise trust, learning, and accountability. Results indicate that supporting learning requires AI systems that scaffold thinking, invite verification, and preserve student agency, rather than simply focusing on efficiency. The work acknowledges limitations, noting the study focused on computer science students, and future research should broaden the sample to encompass a more diverse student population.

Students Seek AI Support, Not Replacement

This research offers new insight into student perspectives on integrating generative AI into academic workflows. Through a sequential mixed-methods approach, the study identifies a nuanced relationship between perceived benefits, potential risks, and desired levels of automation across a range of academic tasks. Findings demonstrate students value AI’s potential for efficiency but express concerns regarding accuracy, originality, and the development of core skills like abstraction and conceptual understanding. The study highlights a preference for AI as a supportive tool rather than a replacement for human interaction, aligning with existing research suggesting the greatest benefits arise when AI complements pedagogical approaches.

Students indicated a need for transparency in AI systems, desiring visible limitations and opportunities for intervention, reflecting principles of Human-Centered AI and the importance of maintaining agency. The authors acknowledge limitations stemming from the specific student population studied, online graduate computer science students, which may not fully represent the broader student body. Future research should explore these preferences across diverse disciplines and student demographics to establish more generalizable findings. Further investigation into the design of AI systems that effectively balance automation with student control, and that clearly communicate their capabilities and limitations, is also warranted. This work contributes to a growing body of knowledge aimed at fostering responsible and equitable integration of generative AI in higher education.