Who Makes the Better Teacher, Humans or AI?

Co-authored by Hannah Feeney, Jessica Murphy, Kodie Curran, and Michael Hogan.

It is 9.30 p.m., and you are a student teacher tasked with developing a lesson plan integrating maths, science and engineering for your iSTEM class tomorrow morning. You have two options. One is your class teacher, whose three years in the classroom have attuned her to what students need, but she left hours ago, and her strength is practical experience, not the precise theoretical framework your assignment calls for. On the other hand, you can consult ChatGPT. While artificial intelligence (AI) has never taught in a real classroom, it can draw on thousands of lesson plans built around your exact pedagogical framework and synthesise them in layman’s terms in under a minute. Which do you reach out to?

Since the release of generative AI in 2022, this has been the question facing students, teachers, and wider society. While AI allows for unprecedented productivity, providing its users with on-demand responses to an almost unlimited range of prompts, many fear its advances come at the expense of human skill and judgment. This is especially profound within education, where schools and universities act as a bridge between current theory and future practice—and where a critical question follows: If an AI tool can plan a lesson in seconds, what is left for the human teacher?

AI in and of itself is neither a harmful nor beneficial force, with its impact wholly dependent on the context in which it is embedded. Both humans and AI have their unique strengths. AI can capably complete well-defined, information-rich tasks while human intelligence excels in socio-emotional interactions requiring contextual sensitivity and empathy. While AI can mimic this contextual understanding, its outputs rest on data patterns rather than lived experience of subtle social cues and a deeper understanding of human activity, history, and culture that becomes part of long-term memory as we develop. This absence of deeper contextual understanding can leave gaps in AI’s response to subtle and complex problems, particularly in education where, as Christoulouka & Verdis (2025) argue, context is not merely a static backdrop to learning but an active force that shapes learning experiences and is shaped by them.

Different strengths, different roles

These complementary strengths are evident in a 2024 study conducted by Li et al., where 23 Chinese students pursuing a master’s degree in educational technology collaborated with either ChatGPT or an in-service teacher in an iSTEM course. The study examined the strengths and limitations of both collaborators, acknowledging that AI may struggle to mirror the relational aspects of teaching, while in-service teachers may lack the time and capacity to mentor student teachers on top of their commitments in the classroom. Participants were randomly assigned to either a teacher-student pairing (TSPL) or a ChatGPT-student pairing (CSPL) condition. Each group then had a one-hour orientation where the TSPL students discussed strategies for effective collaboration and exchanged contact details while the CSPL students were introduced to ChatGPT’s uses and limitations for teaching.

Over the next eight weeks, all participants engaged in a weekly three-hour iSTEM course that combined theoretical iSTEM content with practical teaching case design. In four of these weeks, students completed an individual learning task based on that week’s content, which informed the researchers of their learning trajectory. These tasks built progressively toward the final open-ended assessment, where students were tasked with developing a lesson plan incorporating the interdisciplinary knowledge and implementation strategies they learned throughout the course. These lesson plans were anonymised and evaluated by the course’s lead professors and two teaching assistants, none of whom had a conflict of interest with the participants. The study also measured participants’ critical thinking before and after the course, along with the cognitive load they experienced. Cognitive load was assessed using a single-item measure, where participants rated their mental effort on a scale from 1 to 9, while critical thinking was assessed through a 28-item questionnaire that encompassed five dimensions: truth-seeking, open-mindedness, analyticity, systematicity, and inquisitiveness.

Different strengths, different results

The study demonstrated interesting findings. Participants in the AI condition reported significantly lower cognitive load, noting that ChatGPT could answer their questions at any time. More surprisingly, the AI group’s critical thinking improved over the course—an effect that was not observed in the teacher-supported group. Students put this down to knowing that ChatGPT can produce plausible but misleading answers, which pushed them to critically reflect on its outputs and verify them against reputable sources. Meanwhile, in-service teachers were available only within school hours; while this left room for independent thinking, it also may have encouraged students to wait for answers rather than seek them out independently.

In terms of learning performance, the findings were mixed. Students in the AI condition significantly outperformed students in the teacher-supported condition in the four post-lesson tasks. However, students in the teacher-supported condition performed better in the lesson plan design task, developing comprehensive and applicable lesson designs for a real-life classroom. Student reflections noted that collaborating with an experienced teacher contextualised the practical realities of cognition, lesson design, and implementation in the classroom, providing them with a deeper context that ChatGPT could not replicate.

Returning to the opening dilemma as to who makes a better teacher, humans or AI, the answer seems more complex. While ChatGPT can develop unlimited amounts of lesson ideas, it cannot apply these findings to a real-life classroom. It cannot develop lessons that will resonate with a tired group of 9-year-olds at 1 p.m. on a Friday afternoon. It cannot identify the shy student who requires extra encouragement or the capable student who needs that extra challenge. These judgments come from experience and an intuitive sense of what it means to teach in a real classroom. Teaching requires more than simply imparting knowledge; it requires one to read a room, respond to emotions, adapt to challenges, and build the trust that makes learning happen. This is where human expertise shines.

The future of education is unlikely to belong to AI alone or to humans working without it. Instead, it may belong to those who know when to ask AI for support, when to seek the wisdom of an experienced teacher, and, most importantly, when and how to best combine both. In this way, maybe the best teachers are not human or AI. The best may be a partnership between them.

Education Essential Reads

Why Universities Keep Lowering Their Standards

Hiding Grades at College Isn't the Answer

Hannah Feeney, Jessica Murphy, and Kodie Curran are researchers working at the University of Galway.