%20(1).png)
Many of us have faced the same dilemma: we’re eager to give each student targeted support that meets them where they are and gets them to the next level. We want to give students enough safety and confidence so that they can ask questions and share when they are unsure. And we want to do these things for all students, not just a small subset of the class. There are decades of research and insights into what makes for an impactful learning experience that leads to deep learning and longterm retention. We know how to do it. We’ve seen the differences in outcomes for students who do and don’t have access to these types of experiences. The question is not how to do it. The question is how to do it at scale. Some people are already benefiting from these strategies for impactful learning. We want many more people to gain access to these benefits.
The excitement and scepticism about the potential for AI to realise the benefits of known pedagogy at scale is well founded. While AI offers the potential to deliver robust pedagogy at scale, there are risks. Technology deployed without pedagogical intent, or simply for its novelty, often results in poor learning experiences and undermines trust in digital education (World Economic Forum, 2024).
At Next Gen Learning (NGL), our goal is to keep students at the centre of every decision. We leverage robust pedagogy and deep faculty expertise to design learning experiences where technology is an enabler, not the driver. AI tutors, when integrated into thoughtfully designed learning environments and drawing on expertly curated faculty knowledge bases, can expand the reach and impact of faculty expertise. However, the effectiveness of AI tutors depends on their integration into a design that combines pedagogy and subject matter expertise.
Benjamin Bloom’s work is foundational to educational practice. His research on structured, scaffolded learning, progressing from lower-order to higher-order thinking, remains central to effective course design. In his landmark 1984 study, Bloom demonstrated that students who received individualised, one-on-one tutoring achieved learning gains two standard deviations above those in traditional classroom settings (Bloom, 1984). This phenomenon, known as the “2 sigma problem,” highlights the transformative potential of personalised instruction.
Elite universities have long relied on tutorial methods as a core pedagogical approach to bring these benefits to their students. However, scaling one-on-one tutoring in resource-constrained environments has remained a challenge.
Recent advances in generative AI have changed how we, as educators, can address this problem. A 2024 Harvard Gazette report provides compelling evidence for using AI to deliver some of the benefits of one-on-one tutoring to a wider audience. In a large introductory physics course, students using a tailored AI tutor doubled their learning gains compared to peers in conventional active-learning classrooms (Harvard Gazette, 2024). The AI tutor provided real-time feedback, adaptive problem sets, and 24/7 support, replicating many of the benefits of human tutoring while accommodating diverse learning paces. These results closely mirror the gains observed by Bloom and suggest that, when thoughtfully implemented, AI tutors can help address the 2 sigma problem at scale.
Central to Bloom’s findings is the concept of mastery learning. Mastery learning is an instructional approach that requires students to demonstrate a high level of proficiency in a topic before moving on to new material. Unlike traditional models that progress students based on time or coverage, mastery learning emphasises personalised feedback; opportunities for repeated practice and revision; formative assessments to diagnose and address gaps; and flexible pacing to accommodate individual learning needs.
These principles align closely with the practices of effective human tutors, who adapt their instruction to each student’s needs, identify misconceptions, and provide targeted practice (Bloom, 1984). Mastery learning has been shown to significantly improve student outcomes, particularly in higher education and professional programs where foundational knowledge and skills are critical (Park University, 2025).
AI tutors are uniquely positioned to operationalise mastery learning at scale. For example, in the Harvard physics course, the AI tutor generated adaptive problem sets and required students to demonstrate mastery before progressing. Real-time feedback and unlimited practice opportunities allowed students to learn at their own pace, building both competence and confidence (Harvard Gazette, 2024). This approach enables faculty to use analytics from the AI system to identify common challenges and adjust their teaching accordingly, further enhancing the learning experience (Miller, Kestin, & Mazur, 2024).
The effectiveness of AI tutors is not inherent in the technology itself, but in how it is set up and how it is integrated into a broader learning design. Pedagogically sound learning experiences, grounded in faculty expertise, are essential for realising the benefits of AI tutoring.
The pedagogical intent of the learning experience as a whole should be built into the AI tutor. Spaced repetition, scaffolding, and formative assessment with feedback are key. Courses can include critical thinking, problem-solving, and support to develop self-regulation skills. AI tutors can augment these processes by providing immediate feedback, prompting students to reflect on their reasoning, and encouraging metacognitive strategies.
AI tutors achieve their greatest impact when paired with high-quality, faculty-authored content and intentional instructional design. There are three critical reasons for this:
Domain Expertise: AI algorithms require accurate, discipline-specific knowledge to generate valid explanations and problems. Faculty play a vital role in curating and validating the content that underpins AI tutors, ensuring relevance and rigor (Harvard Gazette, 2024).
Contextual Relevance: AI tutors must be aligned with course objectives and learning outcomes. When faculty collaborate in the design and implementation of AI tutors, the resulting experiences are more likely to reflect the intended rigor and depth of the curriculum (Luo et al., 2025).
Human-AI Collaboration: Faculty insights are essential for refining AI functionality. By analysing data generated by AI tutors, faculty can identify class-wide knowledge gaps and adjust their teaching strategies accordingly. This dynamic interplay ensures that technology serves pedagogy, not the reverse. At NGL, student insights from AI tutors inform both individual feedback and faculty-driven adjustments to didactic content delivery.
The NGL team is using these principles to guide the thoughtful integration of AI tutors in online learning experiences.
Foster Human Connections
Use AI to free up faculty time for higher-order tasks such as mentoring and leading discussions. Technology should enhance—not replace—human interaction.
Integrate into a Pedagogically Sound Learning Design
AI tutors should be intentionally embedded into the course design, not added as disconnected or superficial features.
Mastery Before Momentum
Structure learning so that students can practice at their own pace and demonstrate understanding before progressing.
Align with Learning Goals
AI tools must directly support clearly defined course outcomes and learning objectives.
Prioritise Faculty Leadership
Faculty and learning designers should curate the content, shape assessments, and ensure ethical oversight through regular review of AI outputs.
Ensure Transparency, Explainability, and Compliance
Students and faculty should understand how AI systems make decisions. The AI's reasoning should be accessible and auditable to foster trust and accountability.
Iterate Based on Data
Use engagement metrics and evidence of learning to continuously improve AI interactions and personalise learning pathways.
Design for Equity and Accessibility
Ensure AI-supported learning is inclusive—providing multimodal support, accommodating disabilities, and enabling access across a range of devices.
AI tutors represent a breakthrough in addressing Bloom’s 2 sigma problem, making personalised, mastery-based learning accessible to many more students at a more reasonable cost than has been possible before. Their effectiveness depends on seamless integration into pedagogically rigorous, faculty-curated learning experiences. By centering the expertise of faculty and the experience and needs of students, we can harness AI to create effective, adaptive learning experiences.
Bloom, B. S. (1984). The 2 Sigma Problem: The Search for Methods of Group Instruction as Effective as One-to-One Tutoring. Educational Researcher, 13(6), 4–16.
Dialzara. (2024, April 29). AI Tutoring Implementation: 6-Step Guide for Educators. Retrieved from https://dialzara.com/blog/ai-tutoring-implementation-6-step-guide-for-educators/
Goel, A., & Polepeddi, L. (2024). Jill Watson: Scaling AI Teaching Assistants in Higher Education. AI in Education Journal, 8(3), 45–59.
Harvard Gazette. (2024, September). Professor tailored AI tutor to physics course, engagement doubled. Retrieved from https://news.harvard.edu/gazette/story/2024/09/professor-tailored-ai-tutor-to-physics-course-engagement-doubled
Johnson, D. W., Johnson, R. T., & Smith, K. A. (2014). Cooperative learning: Improving university instruction by basing practice on validated theory. Journal on Excellence in College Teaching, 25(3–4), 85–118.
Luo, T., Young, D., & colleagues. (2025). Exploring instructional designers’ utilization and perspectives on generative AI tools: A mixed methods study. Educational Technology Research and Development.
Miller, K., Kestin, G., & Mazur, E. (2024). AI-Driven Mastery Learning in University Physics. Physical Review Physics Education Research, 20(1), 010145.
Park University. (2025, February 14). AI in Education: The Rise of Intelligent Tutoring Systems. Retrieved from https://www.park.edu/blog/ai-in-education-the-rise-of-intelligent-tutoring-systems
Prodigy. (2021, December 2). One-On-One Tutoring: Top Benefits for Student Success. Retrieved from https://www.prodigygame.com/main-en/blog/one-on-one-tutoring
Roll, I., Aleven, V., & Koedinger, K. R. (2024). Metacognitive Tutoring: Teaching Students How to Learn with AI. International Journal of Artificial Intelligence in Education, 34(1), 78–95.
World Economic Forum. (2024, January 13). 7 principles on responsible AI use in education. Retrieved from https://www.weforum.org/stories/2024/01/ai-guidance-school-responsible-use-in-education/
