Background and Objectives: The arrival of new generative artificial intelligence tools and large language models (LLMs) has significantly altered how people access information and learn. As Generative AI tools become more common in academic environments, they bring both clear benefits and notable concerns. This technology offers powerful capabilities, like producing summaries and answering questions, to make learning faster and more personalized. The issue is complicated by AI's dual potential to both complement and replace cognitive activities vital for deep learning.
While previous research has touched upon AI in education, this study investigates an area not yet fully explored: how these new tools are actively changing student study habits. Existing studies have often been small-scale or focused on narrow audiences, leaving a recognized gap in the literature regarding a deep, systematic understanding of these impacts. This research addresses that gap by comprehensively examining the effects of generative AI on study habits through an innovative methodology. The goal is to provide a clear overview of the opportunities and challenges, offering evidence-based guidance for educators, students, and policymakers on the responsible use of these transformative tools.
Methodology: This study used a qualitative Delphi method to gather and synthesize expert opinions on this rapidly evolving topic. The Delphi method is well-suited for areas with limited empirical data because it provides a structured process for achieving consensus. In a novel application, the "expert" panel was composed not of humans, but of 10 of the world's leading generative AI models: Claude 3.5 Haiku, ChatGPT 4o, Microsoft Copilot, Google Gemini 2 Flash, Grok 3, Perplexity AI, DeepSeek V3, Llama 3.1 70B, Mistral AI 1, and Qwen AI 2.5 max. This approach was designed to leverage the collective knowledge and analytical capabilities of these systems, completing the process in a fraction of the time required for human panels.
The process involved three structured, iterative rounds. One model, Claude 3.5 Haiku, was assigned the role of "referee" to guide the process. In the first round, the panel answered an open-ended question, and the referee synthesized the responses into a 30-item Likert scale survey. The entire panel then completed the survey. The referee analyzed the results for consensus and disagreement, finding seven points of dispute. The process of revising and re-answering the questionnaire was repeated until, in the third and final round, the AIs reached a unanimous agreement on all topics.
Findings: The expert panel's consensus revealed that generative AI has a dual impact, with significant advantages offset by considerable drawbacks. The tools can enhance academic results and reduce study time by making learning more efficient and personalized. The primary benefit identified was personalized learning, where AI creates customized pathways that match a student's speed and style. Other key advantages included accelerated access to information, immediate constructive feedback for correcting errors, intelligent time management assistance, efficient text summarization, and increased student engagement through interactive environments.
These advantages, however, are balanced by serious risks. Chief among the concerns was the potential for over-dependence on technology, weakening students' intrinsic learning abilities and memory. This links to a potential decline in critical thinking and problem-solving, as students may rely on AI for answers rather than engaging in deeper cognitive effort. Finally, the panel highlighted important ethical issues, including student data privacy concerns and the risk that algorithmic biases could worsen educational inequality for certain student groups.
Discussion: The findings point toward the need for a thoughtful balance in integrating AI into education. AI should be used to supplement and support traditional methods, not replace them. Consequently, educational systems must prioritize developing strong digital literacy and critical evaluation skills. This requires teaching students to use these tools effectively, critically assess their output, and apply them in a responsible, ethical manner.
The study's methodology also demonstrates that using an AI expert panel in a Delphi study is an innovative and effective way to research emerging technologies. The authors acknowledge the study's limitations, noting the results are dependent on the specific AI panel selected. Given the rapid pace of AI development, future research could replicate the study with newer models. It is also suggested that future work could combine the AI panel with human experts for a more nuanced dialogue or evaluate the AIs' reasoning capabilities to better understand the "why" behind their consensus.