The Rise of Generative AI in Academic Research: A Double-Edged Sword
When OpenAI introduced ChatGPT-5 last August, many academics raised their eyebrows at the company’s assertion that this artificial intelligence model demonstrated “PhD-level” intelligence. How could an AI, often criticized for its hallucinations, inconsistent reasoning, and tendency to flatter, compete with the intellectual rigor of the world’s brightest minds? Yet today, an array of scholars, including renowned mathematician Terence Tao, are integrating tools like ChatGPT into their research practices, using them in ways similar to how they might have once relied on PhD students.
A Skeptic Transformed
My own journey from skepticism to belief happened gradually over several months as I engaged in a research project employing various generative AI tools. Initially, I doubted whether these technologies could replicate the collaborative, transformative experience inherent in working with PhD candidates. However, as I implemented these tools to support tasks I typically would have entrusted to students, it became clear that generative AI holds significant promise.
While my experience with AI systems heightened my productivity and efficiency, it also illuminated a critical issue: the potential dangers of depending on AI at the expense of traditional learning mechanisms.
The Engine-Room of Research Production
PhD students are undoubtedly the backbone of research in academia. They not only assist in executing experiments and writing papers, but they also undergo a vital educational process that helps form the next generation of research leaders. Through guidance from experienced supervisors, PhD candidates learn to formulate hypotheses, critique findings, and ultimately, take responsibility for their scientific output.
In my research, I focus on developing mathematical models to understand computer security. This work typically involves a collaborative dynamic with PhD students, who have been essential for tasks like proof-writing and programming. Involving them helps me refine my ideas and fuels their academic journey, enabling them to ask vital questions and engage with scientific debates.
More Read
Using AI as a Complementary Tool
During my recent project, I shifted gears and began leveraging ChatGPT as my “collaborator.” Using this AI, I articulated and refined my key mathematical definitions and theorems in what I would typically call “pen-and-paper” style. After that, I utilized Anthropic’s Claude Code to translate these mathematical concepts into a digital format, ensuring I could validate each reasoning step with a proof-checking program.
Moreover, tools like Claude Code and ChatGPT allowed me to implement Python programs demonstrating how these mathematical concepts could operate in practice. Through careful verification, I checked that the programs aligned with the mathematical definitions I developed. Remarkably, what would have taken a year with a PhD student, I achieved in about six weeks with generative AI assistance.
The Hidden Danger of AI Dependency
Despite these productivity gains, it’s essential to recognize a lurking danger: the potential diminishment of the learning experience for budding researchers. The efficiency boost came at a cost—the invaluable educational opportunities for students were sacrificed. During this project, I failed to teach a student essential skills, such as how to identify meaningful research problems, rigorously test hypotheses, and articulate ideas clearly. Learning to navigate scientific inquiry is a critical aspect of a PhD, one that contributes to developing well-rounded researchers.
An essential part of cultivating excellent researchers is developing their awareness of knowledge gaps, maintaining a healthy skepticism toward their hypotheses, and fostering a sense of intellectual responsibility for their work. Unfortunately, AI simply cannot replicate these human experiences effectively.
The Need for Cautious Adoption
As universities increasingly adopt generative AI tools, it becomes imperative to approach their use cautiously. The apprentice model of training graduate students has been foundational to the advancement of knowledge in academia. If we lean too heavily on AI, we risk undermining this model and the quality of research it sustains.
While generative AI can enhance productivity, especially in the early stages of research development, it should be viewed as a complementary tool rather than a substitute for mentorship and hands-on training. The challenge for academics lies in balancing the efficiencies offered by AI with the irreplaceable value of human guidance and learning.
By thoughtfully integrating generative AI into the research landscape, we can harness its benefits while preserving the traditional learning pathways that foster the next generation of thinkers and innovators.
Inspired by: Source
