Humans in the Loop: The Key to Making AI Tutoring Work in Your School

AI is already in our classrooms – 67% of high school students are using tools like ChatGPT and AI tutoring to support their learning. But not all AI support is created equal, and without expert human oversight, these tools risk doing more harm than good.

In this article, I’ll share insights from my experience as a former math teacher and now Head of Academic at Third Space Learning, where we’ve developed Skye, our conversational AI math tutor. You’ll see why we believe the only effective AI tutoring is human-guided – and how schools like yours can harness it to deliver high impact tutoring and scalable support.

The promise of AI tutoring 

Despite ongoing concerns over AI math quality, many teachers are exploring AI in education for a good reason: it can broaden access to individual support and free up time in busy classrooms. In fact, 57% of teachers are already using tools like ChatGPT for school work, and more than one in ten (11%) have tried an AI tool in a lesson. Not all tools are made equal- read our blog on the best AI for math for a comprehensive review.

Gradually, it’s becoming more common to use AI tools as intelligent tutoring systems (ITS) to enhance student learning experiences.

While these artificial intelligence systems hold promise for improving personalized learning and educational experiences for a large number of learners, there are questions surrounding their impact on student performance compared to traditional human tutoring.

The effectiveness of AI tutoring compared to traditional tutoring

Unlike many online math tutoring companies, at Third Space Learning, we began developing our own curriculum and lessons for our tutors to deliver back in 2013.

Alongside every lesson, we created detailed tutor guides for every slide, highlighting key math misconceptions, Socratic questioning prompts, and model answers. All of these detailed features have become foundational in the development of our voice-based AI tutor.

This means we’ve over 10 years of data from 4,000 schools, 170,000 students and 2 million lessons that can show us what’s most effective in students making progress.

So we know what works in traditional tutoring and how essential our math experts are.

Up till now, my team of teachers has developed the diagnostic assessment and summative assessment, created the lessons and trained the instructors in best practice for teaching one on one and in incorporating the principles of formative assessment, instructional support and feedback in every lesson.

And all that remains exactly the same for the team working on AI tutoring. The only difference is that instead of traditional instructors delivering the math tutoring, we leave that part to Skye, the voice based AI tutor trained by us.

Benefits of AI tutoring 

There are many benefits to using AI tutors over traditional tutors. Conversational AI tutors like Skye address longstanding challenges in mathematics education that traditional tutoring models currently face:

Unlimited reach 

Traditional, human tutors can only work with one student or group at a time. For multiple learners to access tutoring simultaneously, you need as many tutors.

However, AI tutors run off of advanced algorithms meaning there are unlimited versions of them. This means they can support as many students who need it, simultaneously.

Scheduling flexibility

Unlike traditional tutors, AI tutors have unlimited availability. Schools can book voice-based AI math tutor sessions with Third Space Learning at 5-minute intervals before, during, or after school.

This flexibility removes the scheduling challenges often seen in traditional tutoring.

Consistent approach

Student outcomes with traditional tutors can vary depending on their experience, teaching style, and even day-to-day factors like mood or tiredness — even with consistent tutor training. AI tutoring sessions eliminate these variances, delivering a more consistent and reliable learning experience every time.

Every conversational AI tutoring session with Skye follows the same high-quality, instructionally supported structure. Skye uses a proven “I do, we do, you do” framework developed by experienced mathematics teachers to support students in developing their understanding of math concepts.

Personalized learning

AI tutors excel at adapting to individual needs through diagnostic assessments and continuous progress monitoring.

Working from advanced AI algorithms created by teachers with in-depth knowledge of common misconceptions for any topic whether that’s rounding or quadratic equations AI tutors such as Skye can identify misconceptions and provide appropriate targeted support.

Limitations of AI tutoring without human expertise  

Despite these advantages, AI tutors do face some challenges that many humans naturally overcome:

• Content expertise depth: AI tutors can explain mathematical concepts but they may have difficulty grasping the nuances of teaching more complex topics such as algebraic reasoning.
• Curriculum competence: AI tutors may struggle to assess at which level to introduce topics or determine how deeply to explore concepts based on a student’s prior knowledge and curriculum requirements.
• Emotional intelligence: Humans are well-equipped at reading emotional cues that may indicate a student is confused or disengaged and adjust their approach accordingly.
• Adaptive problem-solving: When students attempt unconventional problem-solving strategies, humans can follow their reasoning and provide appropriate guidance, unlike many AI algorithms.
• Reliant on their training data which can be biased: While AI tutors are a fantastic way to scale personalized support, they are only as effective as the data they are trained on.

These limitations suggest that AI alone is not a sufficient tutoring tool to improve student outcomes in math, or any other subject, and it requires human expertise.

Curriculum alignment

“Even the best AI tutor is ineffective if the lesson content isn’t grade appropriate or aligned with what students need to know.”

From the start, our academic team of former teachers designed Skye’s math curriculum using their deep understanding of classroom realities. They know how concepts build on each other in a spiral curriculum, how to introduce topics in an age‑appropriate way, and how to spot and address common misconceptions. As a result, Skye’s lessons:

• Are accurate and age‑appropriate
• Align with classroom teaching and state standards
• Cover relevant material with real‑world examples
• Remain accessible and equitable for all students.

An AI tutor can deliver this expert content using a conversational, voice‑based approach, but the underlying syllabus has to come from educators who know what students truly need at each level.

Scaffolded learning to address misconceptions

“If students hold misconceptions, they won’t learn or address them by simply being told the right answer.”

Human experts designed every lesson in Skye to follow the same “I do, we do, you do” structure used by effective tutors in live sessions. If a student answers incorrectly, Skye gives up to three targeted hints that help them identify, themselves, exactly where they’ve gone wrong. Crucially, these hints and prompts come from teacher‑crafted scripts reflecting real classroom misunderstandings.

Because Skye is voice‑based, not a text-based AI tutor, it can “listen” to student answers and detect misconceptions or half‑formed ideas. At a system level, it’s built with adaptive teaching in mind, so Skye can increase or decrease the level of help based on student responses – just as a human tutor would.

Preventing cognitive overload

“AI tutoring should be built to avoid cognitive overload too, breaking down learning into simple instructions to reduce distractions.”

Any experienced teacher knows how quickly students can become overwhelmed if too much new information comes at once. To combat this, our education experts made each online tutoring Skye lesson cover just one small step in the larger objective. This means:

• Familiar structure: Students learn the “I do, we do, you do” pattern quickly, so they can focus on the maths itself rather than deciphering a new format every time. 
• Clear, minimal slides: Each slide highlights only the content a student needs at that moment, using pointer tools and blurring out irrelevant areas.
  

Where human tutors might decide in the moment how to break a concept down, Skye’s authors preemptively craft the text, visuals, and question flow to minimise mental strain.

Timely feedback

“Many AI tutors, in the form of chatbots, simply mark answers correct or incorrect. This doesn’t help students learn.”

Instead of halting at right and wrong answers, Skye is trained to ask open‑ended questions and analyse student explanations. It works out what each student understands, and where they need more support, then keeps probing using effective questioning until they articulate their thinking and resolve the misconception.

This feedback loop is grounded in how real teachers prompt deeper understanding – rather than a yes/no approach. Whenever we notice a gap in Skye’s follow‑up prompts, our academic team updates the tutor’s responses to ensure the next student in a similar position will get the help they need.

Continuous refinement

“Third Space Learning’s academic team continually monitor and improve Skye based on real classroom implementation.”

Because Skye is an evolving AI platform (not a static question bank), teacher feedback and real student data feed directly into system updates. When schools, teachers, or students point out a confusing explanation or an awkwardly phrased question, our team investigates and refines Skye’s responses. Recent improvements have included:

• Reordering lessons to align more easily with in‑class teaching
• Guiding conversations more effectively to keep students focused
• Recognizing different accents better for smoother, more natural exchanges

Every tweak and fine‑tune relies on human observation building on evidence based teaching strategies, ensuring Skye remains a responsive learning tool rooted in expert pedagogy. By combining consistent AI delivery with real teacher oversight, we can scale personalized, high‑quality tutoring for every student who needs extra support.

Safeguarding with human oversight

“AI tutoring shouldn’t just be about delivering content; it must also protect every student’s wellbeing.”

Just like our human tutors, Skye’s interactions are guided and monitored by real educators at every step. Here’s how we ensure safety remains firmly in the hands of human experts and adheres to AI policy guidelines:

• Strict Guardrails: We use a highly moderated AI model trained by humans to stick to approved math content. Skye cannot wander off‑topic or display unverified materials; it engages students only as its teacher‑designed prompts allow.
• School‑Scheduled Sessions: All Skye sessions happen at times chosen by the school, on school premises, and under teacher supervision. There’s no video – only an audio connection plus our teacher‑created slides – so Skye can’t show anything unexpected or collect extra data.
• Automated Alerts and Recording: Every session is recorded in full, and Skye is trained to flag any potential safety concerns. These are immediately reported to the school’s safety lead or relevant staff member, who can review the flagged session and take action if needed.

By placing teachers in charge of scheduling, content guardrails, and any flagged issues, we maintain a robust “human in the loop” approach that ensures each student’s safety – just as carefully as if a live tutor were in the room.

Expertise of humans for the cost of AI 

Traditional tutoring is invaluable but we’ve heard time and time again that one of the biggest barriers to traditional tutoring is cost. That was one of the principal reasons we developed Skye: to make high‑quality, one‑on‑one math tutoring affordable for every school. At the heart of its effectiveness is a carefully designed model that combines advanced AI technology with deep human expertise.

Third Space Learning’s mission has always been to close learning gaps through tutoring without compromising on quality or putting too much strain on budgets or staff workload.

With Skye, schools can confidently expand one‑on‑one tutoring to all the students who need it most at a fraction of traditional costs.

The future of AI tutoring: combining human expertise and technology

The future of effective AI tutoring lies not in technology alone, but in thoughtful partnerships between the capabilities of AI and human expertise.

As we’ve seen with Skye, the most impactful AI tutoring solutions combine AI’s scalability and consistency with the irreplaceable judgment, creativity, and adaptability of experienced, traditional educators.

This partnership approach ensures that AI tutoring remains:

• Pedagogically sound, guided by qualified teachers’ expertise
• Classroom-aligned, complementing rather than competing with school curricula
• Responsive to individual student needs beyond algorithmic analysis

Human-guided AI tutoring makes high-quality education accessible. By making personalized support affordable and scalable, solutions like Skye enable schools to provide a tutoring solution to every student who needs it.

As AI tutoring continues to evolve, the role of the human element becomes more crucial, not less. The best AI tutors will always have human oversight to ensure they deliver not just information, but true education.

I was completely impressed and am looking forward to supporting Third Space Learning on their journey in developing their A.I. tutoring programs. Our students and teachers are also excited about this partnership and eager to see the growth that this program will bring to our students.

Sarah Oliver, Principal, Dutch Creek

AI tutoring will never replace human teachers, but with continued advancements in AI and pedagogy, it will become an even more powerful tool to support student learning.