I do think artificial intelligence can be useful to teachers in the classroom, though I definitely don’t think it is or can be “revolutionary” and its benefits to students are limited (though I do think it has lots of benefits for English learners).
Other teachers, though, feel more positive about AI’s use as a “tutor” to students, and you can read more about their ideas here.
In this post, two educators share their ideas about using AI tutors in education.
AI Tutors: Benefits vs. Challenges
Svetlana Kandybovich is a teacher, teacher trainer, and learning designer with more than 20 years of international English Language Teacher (ELT) experience. She shares her insights on teaching, learning, and professional development with fellow educators and teacher trainers through her blog, ELTcation:
The ed-tech world is full of promises about AI-powered chatbots or “tutors” designed to imitate the best teacher in the world. These tools are said to transform learning, language learning in particular, by making it more engaging, personalized, adaptive, and inclusive. By and large, these promises resonate with me as both a teacher and a parent. The idea of giving every learner round-the-clock access to personalized support and language practice is certainly exciting, but as we all know, promises don’t mean much unless they actually deliver.
I’ve tried a range of chatbots across different platforms and even developed my own from scratch using developer sandboxes. Each chatbot was designed with specific language learners or groups in mind, with objectives ranging from broad to highly focused. However, I noticed several key challenges that need to be addressed, beyond privacy and ethical concerns, for AI tutors to evolve from being just one-off activities into effective learning aids.
Learner Readiness
An instrument designed to imitate teachers doesn’t automatically make it one. Like any other learning tool that requires a certain level of learner autonomy, learners need to know how to learn and how to use the tool effectively to support their learning. AI tutors often rely on learners’ critical thinking and AI literacy as essential prerequisites for using the technology effectively.
However, in practice, many learners, regardless of their AI literacy or critical and creative thinking skills, struggle to gain meaningful value from chatbot interactions unless they are prepared to engage and learn with it. Simply telling learners to “chat with the chatbot” or equipping them with a set of prompts, without further guidance, is unlikely to result in consistent and meaningful learning.
For example, one of my earliest experiments was the Q-Chart Assistant, a feature of the Question Chart module I made accessible online. The process works like this: Learners create questions using a Question Chart and then run them through the chatbot. The chatbot helps learners identify mistakes, clarify their questions, and suggests additional practice options based on the areas it flags for improvement.
After reviewing the recorded interactions with the chatbot, I noticed that learners who weren’t ready to engage with the chatbot quickly lose interest and fail to use it as a learning tool. Many stop after a quick check or immediately ask the chatbot for the correct answers. In these cases, the learning they gain is essentially the same as simply looking at an answer key.
Customization
For a chatbot to provide genuinely personalized practice, it needs to offer the right customization options to align it with the learners’ needs, skill levels, learning objectives, and tasks, as well as the teacher’s approach.
When setting up or assigning an AI tutor, it’s essential to have access to key features, including the ability to view the system instructions that guide how the chatbot processes prompts, how to upload relevant data, how to switch between AI models and choose the most appropriate model for the task, and how to adjust settings like the model’s temperature to align the chatbot’s responses with specific tasks. Playlab is one of the few platforms for educators I’ve come across that provides teachers with a sandbox offering these functionalities to create educational AI chatbots for their own contexts.
A Teacher in the Loop
AI models powering AI tutors lack complete consistency and reliability, meaning there’s always a chance they could go off course. I’ve had cases where the chatbot would pick up on a perfectly acceptable answer or give feedback to the learner that didn’t seem relevant at all. This makes it necessary for the teacher to stay involved, keeping an eye on both what the learner is doing and how the chatbot is behaving, stepping in and tweaking instructions as needed. This also means that using AI tutors often requires more time and effort from the teacher.
That said, at the current rate of development, AI tutors can benefit language learners as a supplement to classroom practice, especially when they have a narrow focus and are paired with motivated learners who can manage their own learning with the chatbot. However, it’s still too early to view them as tools that will completely transform language learning and help achieve broader objectives, such as developing fluency beyond scripted conversations. More advanced models, like real-time voice agents and empathic models, could change that, though. Time will tell.
AI Tutors Tutoring Human Tutors?
Michael Pershan is a math teacher and writer in New York City who writes about teaching at http://pershmail.substack.com:
Silicon Valley’s long-standing dream is for technology to free students from the shackles of schooling. Once freed, children will pursue their talents and interests without restraint, yielding a population as bold and innovative as the best of the tech industry. (Steve Jobs dropped out of college, if you haven’t heard.)
The particular technology fueling this revolution often changes (e.g., MOOCs, Khan Academy, flipped classrooms, gamified learning). But the dream never dies. It is currently fueled by AI tutors, admittedly the closest any product has come yet to the vision.
Like most anyone who works directly with children, I think this dream is absurd. Of the thousand or so students I’ve taught, I struggle to think of any that could have learned much without the presence of a teacher, someone to set expectations, provide accountability, and guide learning.
The truth is that self-motivation for academic learning is a rare quality. Very few adults, let alone children, possess it. Even the most intelligent of us go through our lives ignoring books, websites, and videos that we might learn a great deal from. It’s not that we lack curiosity; we need teachers. We don’t worry about ignoring our YouTube tutors.
Similarly, most will walk away from a chatbot without thinking twice. But this is precisely what a tutor is supposed to be—someone we can’t easily walk away from. Maybe one day people will care about offending robots. Until then, AI tutors are an educational dead end, a bad idea.
Despite all the above, I’m not exactly an AI skeptic. I can imagine a future where intelligent software makes a big impact on human learning. But it won’t be by replacing human tutors—it’ll be by improving us.
Good tutors are dynamic. They quickly assess a student’s abilities and create a plan on the fly for what they might be able to learn next. Weak tutors, on the other hand, are inflexible. When they realize that a student can’t solve an equation, their only recourse is to solve it for them (while explaining each step), then asking the student to solve the next one. A bad tutoring session feels like running into a brick wall, over and over again. There is no progress, because there is no progression.
But what if we could provide tutors with live, in-the-moment feedback? Would that improve the quality of their tutoring?
In a recent article, Mike Goldstein described an experiment he helped organize along these lines:
An AI bot would patrol a tutorial, and then, roughly 20 minutes into a tutorial, a little box would pop up on the screen. It told teacher and student what the talk ratio was, just like a Fitbit offers your step count when you glance at it. If either party was talking too much, they’d adjust.
The Stanford researcher that Goldstein enlisted to study this experiment is Dora Demszky. Many of her recent publications are in the same vein, looking at AI-generated teacher feedback in a variety of forms. In one study, she and her colleagues study a piece of software they call “Tutor CoPilot.”
This seems a very promising direction to me. But, at least for now, Silicon Valley and its admirers have no interest in it, for the simple reason that it leaves schools intact. The dream has always been to empower the individual by weakening the power of standardized, age-graded schooling. The enthusiasm for this individualizing project has produced a massive overhype of the prospects for AI tutors, the latest great hope for disrupting education. I have to imagine that its failures will become apparent extremely soon. Nevertheless, I’m holding out hope that AI will make a difference for students by improving teachers.
Thanks to Svetlana and Michael for contributing their thoughts!
Consider contributing a question to be answered in a future post. You can send one to me at [email protected]. When you send it in, let me know if I can use your real name if it’s selected or if you’d prefer remaining anonymous and have a pseudonym in mind.
You can also contact me on Twitter at @Larryferlazzo.
Just a reminder; you can subscribe and receive updates from this blog via email. And if you missed any of the highlights from the first 12 years of this blog, you can see a categorized list here.