Two years after the launch of ChatGPT, it's time to look at how AI has impacted education and tackle a burning question: Do we really need AI tutors?

First off, let me give you my take—yes, we (and our kids) do need AI tutors in the future. But I'm not talking about the current versions like Khanmingo, Duolingo Max, ChatGPT, or Codekidz.

I've had numerous conversations with friends who have deep experience in education. A common concern is whether AI tutors are truly effective. After nearly two years of observing AI trends in education, it seems like all we have are chatbots that mimic a teacher's accent. They can't solve educational problems or provide effective support for teachers, and many worry about job losses because of them.

Let's review some (maybe) successful AI applications in education like AnswerAI, MagicSchool, CourseAI, and Perplexity. Most of these are designed to help with specific tasks rather than replace human interaction. These tools can enhance learning experiences, but they still lack the nuanced understanding and empathy that a human tutor provides. I see them more as tools than true tutors.

A tutor isn't just someone who helps with homework. a tutor is a mentor, someone who can recognize when a student is struggling emotionally or academically. I created Codekidz to bridge this gap by integrating emotional intelligence into learning, but it's still far from perfect. This is partly due to my own limitations and partly due to current technological constraints.

What Should an AI Tutor Be Like?

An AI tutor shouldn't just be a chatbot. It should be like a real human but capable of doing things humans can't, or else what's the point?

Humans can't be available 24/7, but an ideal AI tutor would provide personalized support around the clock, adapting to each student's unique needs and learning pace.

Humans can't remember everything about a child, but an AI tutor can store and analyze vast amounts of data about a student's progress, preferences, and challenges.

But most importantly, an AI tutor should feel like a real tutor, not just a tool. It should exist as a truly engaging individual, just like Her.

Imagine in a classroom, when the teacher is teaching, the AI tutor, let's call her Tracy, quietly observes, ready to offer personalized assistance at any moment. When the teacher asks, “Tracy, could you provide some additional resources for those struggling with fractions?” Tracy springs to life, instantly generating tailored exercises and explanations, ensuring no student is left behind.

During breaks, students can chat with Tracy about anything, both one-on-one and in groups. Tracy listens, engages, and encourages curiosity, fostering a sense of community and belonging.

After class, when students go home, Tracy remains available, offering support through interactive quizzes and discussions. She helps students reinforce their learning at their own pace, providing a seamless bridge between classroom and home.

Tracy makes extensive use of visuals, voice, and graphics, and rarely uses text to interact with students, which is a human way of communication.

Should we create a visual look for Tracy? Definitely not. Tracy should live in everyone's imagination, not visualized by anyone specific. You can picture her as a teacher, a student like yourself, or a friendly peer.

So, as you can see, Tracy is not just a chatbot. She's a dynamic learning companion. You don't interact with her through boring one-on-one conversations. you interact with her through your voice, expressions, and actions. You need her in your interactions with teachers or classmates. Most of the time, she won't disrupt the learning process, but when you need her, she'll be right there.

Can This Be Realistic?

Sadly, not yet. While the vision of a truly effective AI tutor is exciting, there are many significant challenges that need to be addressed.

1. Emotional Recognition: One of the biggest hurdles is teaching AI to accurately recognize and respond to human emotions. Even the most advanced language models, like GPT-4 and potentially GPT-5, struggle with understanding and interpreting complex emotional cues. This is crucial for a tutor meant to provide emotional and academic support.

2. Large Context Data Management: An AI tutor would need to handle vast amounts of data, not just about academic content but also about each student's progress, preferences, and behaviors. Managing and correctly retriveing these data remains a significant challenge.

3. Network Latency: For an AI tutor to be effective, it needs to operate in real-time. Any noticeable delay can make interactions frustrating. The inference time cost of current large models is still very high, not to mention that the internet is still very slow in many places.

4. Privacy and Security: Storing and analyzing personal data about students raises serious privacy concerns. Securely managing this data while complying with regulations is crucial to protect students' privacy.

5. Diverse Learning Styles: Understanding and adapting to diverse learning styles is challenging for AI. Each student learns differently, and an AI tutor must be versatile enough to cater to these differences effectively.

6. Technological Limits: While technology is advancing rapidly, we are still far from creating an AI that can mimic the full range of human capabilities. Current AI lacks the depth of understanding and empathy that human tutors provide.

In conclusion, while the dream of a truly effective AI tutor is enticing, we're not quite there yet. With continued advancements in technology and a focus on overcoming these challenges, we might just make this vision a reality in the near future.