AI Can Add, Not Just Subtract, From Learning

The role of artificial intelligence (AI) in education continues to change as teachers and parents learn how it can be used in classrooms and other learning environments. For Assistant Professor Ying Xu, studying its impact and possibilities has become essential to her research.

Xu, who believes this is a “critical moment for us to emphasize evidence-based research,” has delivered new research in recent months regarding AI’s impact on children and learning. She recently co-authored a study detailing how children perceive their interactions with generative AI, and another exploring how AI-powered chatbots can help bilingual student learning and parent-child dialogues.

While AI’s advances can create enormous potential for learning, the goal for Xu is to explore how the technology can make better use of existing learning time rather than replace other activities.

“Rather than aiming to ‘introduce’ AI to children — which could take away from valuable time with their families or outdoor play — our starting question is always, Can AI make the time children already spend on media more enriching and engaging?” says Xu, who joined HGSE in 2024. 

One recent study looked at how AI-enabled TV characters can interact with students and how digital media use impacts language development in early childhood. As that work has weighed what’s possible with AI and how these students perceive the tools they’re learning with, Xu points more to what AI can enhance in a student’s learning ecosystem.

“I think that a lot of the worries and concerns we have are mostly based on replacement. We think about what kids lose if they engage with AI: They’re losing out on interacting with students, they’re losing out on this productive struggle,” explains Xu. “But what I see is addition, and what AI could add to students’ everyday learning landscape.”

Below, Xu details how her educational research began to explore how artificial intelligence impacts learning and why she feels the technology can be harnessed to improve student outcomes.

When did AI become the focus of your educational research?

I started doing research on AI’s impact on education before it became as prominent in the public sphere as it is today. It stemmed from my earlier research on how children learn from educational media in general, including television, interactive apps, and electronic books. I observed how children interacted with these technologies, measured their learning, and explored if these technologies brought about meaningful benefits.

I walked away feeling that there was still room for improvement — I wondered whether technology could better support the rich, interactive learning that happens in natural social settings, such as with teachers, caregivers, or peers. That was why devices powered by conversational AI, like Siri and Alexa, caught my attention as they introduced technology capable of enabling natural dialogue. What if we could leverage this capability and turn it into focused educational opportunities? We used evidence-based instructional approaches to explore this question, to see if AI could simulate those instructions in reading, science learning, and creative activities. And we found that in many cases, AI could be quite effective.

Then, as you know, the introduction of ChatGPT brought AI to the forefront of public discussion. On top of that, the most widely used AI products are typically designed for general purposes rather than being specifically created for education or children. This has added urgency to the research, as families, educational systems, and policymakers now face many pressing questions. At the same time, it has pushed much of the research beyond lab settings and into real-world environments that are evolving rapidly.

How do you navigate the question of the possibilities of AI to aid learning versus what the technology can do right now?

I think this is a critical moment for us to emphasize evidence-based research more than ever. With all the excitement around AI, it’s easy to get caught up in using it just for the sake of it. That’s why we need to shift the narrative — not by asking how we can fit AI into education, but by starting with the end goal: What learning outcomes do we want to achieve, and can AI meaningfully contribute to them?

Most of my research has focused on children. Rather than aiming to “introduce” AI to children — which could take away from valuable time with their families or outdoor play — our starting question is always, Can AI make the time children already spend on media more enriching and engaging?

Let me give you an example: Think about television. On the one hand, TV is one of the most accessible educational resources for children. On the other hand, it may not be the most effective way for them to learn, as it primarily delivers information in a one-way format, while research consistently shows that children learn better through interactive engagement. To bridge this gap, we collaborated with PBS Kids to integrate AI into their programs, allowing children to have science-related conversations with media characters as they watch TV.

When evaluating whether this approach works, we focus on the important learning outcomes: Are children acquiring science concepts? Are they developing a curiosity-driven mindset to explore the world around them? Through a series of studies, we found that AI can, in fact, amplify learning opportunities in everyday media use. For example, by enabling interactive dialogues with TV characters, we observed significant improvements in children’s scientific reasoning and engagement. These findings reinforce the idea that AI’s role in education should not be about replacing traditional learning experiences but about enhancing them in ways that are backed by research.

That speaks to something you mentioned at the Askwith Education Forum last fall, which is that AI is just another tool, one that can bring positive change when used correctly.

Yes, there has been much debate — and even fear — about whether AI will replace educators. To be fair, these concerns are not unfounded. Data shows that AI tutors can indeed successfully mimic certain behaviors of human educators. In some cases, evidence suggests that children achieve similar scores whether coached by an AI tutor or a human tutor. However, while this evidence highlights AI’s potential, it does not justify the conclusion that AI can replace human teachers.

We need to recognize that learning and development occur through a long and complex process. It is not just about receiving information — something AI might be good at — but also about social interactions, building trust, and forming relationships, all of which are crucial to a child's growth. These fundamental aspects of education are challenging for AI to replicate.

Ultimately, the question is not whether AI or human teachers are better but how they can work together to improve education. AI could bring automation, access to information, and efficiency, while human educators offer subject matter expertise, experience, and personal connection and support that students need to thrive.

It’s a really fascinating time to study AI and how it can have an impact on students. What kinds of questions are you trying to answer when you pursue research on how AI impacts education?

There are different layers to how we can approach this issue.

The first area is to understand the phenomenon — how students are engaging with AI, what specific tasks they use it for, and under what circumstances. This also involves exploring questions such as what motivates children to use AI and what goals they seek to achieve. Such research could provide a baseline description of how AI is integrated into students' learning experiences.

The second layer of research focuses on the impact of AI on student engagement and learning. The key question is whether AI improves learning or developmental outcomes compared to traditional methods or other educational resources. To answer this, we need comparison studies where one group of students uses AI while another learns without it. We can then determine whether AI provides measurable benefits and in what ways. We should also focus on potential negative consequences, such as over-trust or reliance.

The third layer focuses on how we can maximize the benefits AI can bring, including different ways AI is designed and implemented. For example, questions arise about whether AI code-switches based on children's home language or engages them in multimodal interactions that encourage multiple forms of engagement. It also involves considering instructional activities that complement AI usage — for instance, determining at which stage of learning AI might be most beneficial. This line of research has direct implications for both design and practice.

I think that by addressing these issues, we will be in a good position not only to significantly deepen our understanding of AI’s role in education but also to identify actionable strategies for its effective implementation.