Evidence Matters | AI in K-12 Education: The Risks and Rewards

What if the smartest student in the classroom isn’t a student at all?

Today, students can ask AI to explain a difficult equation at midnight, brainstorm an essay in seconds, or create a personalized study guide with a few prompts. For teachers, AI and other educational technologies can provide new ways to personalize instruction, identify learning gaps, and give students additional opportunities to practice.

But there’s an important question beyond all of this technological progress: Is AI actually helping students learn, or is it simply helping them complete schoolwork more easily?

The answers emerging from recent research are more complicated than either extreme.

The Benefits: More Support and Personalized Learning

One major advantage of AI is individualized support. Students can ask questions at their own pace, receive alternative explanations, practice difficult concepts, and obtain immediate feedback.

A 2025 systematic review and meta-analysis of 68 experimental studies found a moderate positive overall effect of generative AI on learning outcomes, with particularly strong effects in areas such as behavioral engagement and self-regulated learning. AI can also support higher-order thinking when students use it as a learning partner rather than simply an answer generator. It can challenge students with questions, provide feedback, identify gaps in understanding, and help them approach difficult concepts from different perspectives.

EdTech can also help teachers personalize instruction. AI can analyze student performance and help educators identify where individual students are struggling. It can potentially reduce administrative work, giving teachers and counselors more time for direct interaction with students. Research on AI-supported instruction in STEM education also points to potential benefits. A 2026 meta-analysis found positive effects of AI-supported instruction on student learning outcomes, although the size of those effects varied across educational contexts and interventions.

The key is not replacing human interaction with AI, but using the technology to fill gaps that cannot reasonably be filled by one teacher managing a classroom full of students. Used well, it can give teachers better information and give students additional opportunities to practice.

The Risks: When Convenience Replaces Learning

On the other side of the coin, the same technology can create problems when students rely on it to do the thinking for them.

If a student asks AI to write an essay or solve a math problem without attempting the task on their own, the student may complete the assignment without developing the targeted underlying skill. And here, something important is then lost—completing schoolwork is not necessarily the same as learning from it. Students often need productive struggle, sustained attention, and practice before they truly master a concept.

Research reflects this concern. A 2025 meta-analysis of 57 studies examining university students found that generative AI improved academic achievement, language skills, motivation and higher-order thinking, but did not produce a statistically significant improvement in metacognition—the ability to monitor and regulate one’s own learning. Thus, a student might be able to perform better with AI assistance without necessarily becoming better at managing their own learning.

So, what happens if schools use AI to make learning easier at the same time the workplace is demanding stronger human expertise? A July 2026 analysis from KnowledgeWorks describes what it calls an “expertise paradox.” As AI automates routine and structured tasks, human value in the workplace is increasingly shifting toward judgment, oversight, problem-solving, empathy, quality control, and deep subject knowledge. At the same time, some of the routine tasks that once helped people develop those abilities—both in school and early careers—are increasingly being automated. A student who regularly asks AI to generate an answer may become very efficient at completing assignments but have fewer opportunities to develop the knowledge needed to determine whether that answer is correct. Yet those evaluative skills may become more important, not less, as AI becomes widespread.

The Key Is Achieving Balance

The central challenge is not whether students will use AI. Increasingly, they will. The challenge is whether they will use it in ways that build knowledge and independence, or in ways that allow the technology to do the learning for them. AI can provide support, personalization, and immediate feedback. But students still need opportunities to struggle, question, practice, and think independently. The goal shouldn’t be to make learning effortless. It should be to make learning more effective without removing the thinking that makes learning possible. That distinction may ultimately determine whether AI becomes a powerful educational tool or simply a faster way to complete assignments.

But another problem also persists. As schools try to figure out how students should use AI, the technology itself is changing faster than researchers can study it.

In Part 2, we’ll look at that evidence gap and what happens when AI-generated content becomes part of what children are actually learning from.

---

Kelly Gregory is the Riley Institute’s Director for Public Education Partnerships and Projects and previously taught for 11 years in South Carolina public schools. She holds a bachelor’s degree in Psychology and a master’s degree in Special Education. She also holds a National Board certification as an Exceptional Needs Specialist. She can be reached at [email protected].