Subject-specific takeaways
We saw examples of powerful benefits and concerning risks in math and literacy instruction.
Literacy
Benefits
More writing practice and faster, more actionable feedback on writing:
Writing assignments are so laborious for teachers to review, provide feedback on, and grade that students are often given few writing assignments (or only short writing assignments). To benefit from feedback, students need the chance to apply it relatively quickly, but the time consuming nature of writing assignments for teachers often makes that impossible. We saw multiple products increase the volume of student writing, the speed at which students receive actionable and applicable feedback, and the number of at-bats students were given to improve their writing to an extent that would be impossible for teachers to deliver on their own.
Precise diagnosis of letter sound combinations.
AI-powered products can identify the specific letter sounds and letter combinations that individual students have mastered more precisely than a teacher could possibly do for 25+ students at the same time.
More consistently high-quality close-read questions:
Facilitating a discussion of a rich text is a complex teaching task requiring intimate understanding of the text, common misconceptions, and questions ready for students who come close but don't articulate the full answer. Even when done well, it is challenging to engage all students in the class at the same time. Some AI-powered products demonstrated that they could pose text-based, close-read questions to individual students or small groups without reducing the expectation of the question or inadvertently giving students the answer more consistently than we see in paper-pencil instruction.
Risks
Drills for skills and standards in isolation:
We saw a number of products offer teachers lesson planning or practice worksheets in literacy that were aligned to a skill or a standard but not connected to a text or a plan for building knowledge. This is a deeply concerning pattern because extensive research confirms that reading instruction needs to be taught within the context of texts, not as comprehension skills in isolation.
Text releveling:
Lowering the complexity level of a text is a practice with a long history and intensely debated merit. There are some cases when it may be appropriate (e.g., when the goal is to build students' background knowledge on a topic by exposing them to a variety of texts and media sources before engaging in a close read of a complex text). Generally, however, lowering the complexity of the core text a student is reading has been found to limit long-term growth. We saw a number of AI-powered products that made it easy for teachers or students to change the complexity of texts, which we think is likely to decrease, not increase, cognitive demand and long-term learning.
Novelty without follow-through:
Some AI-powered products we saw allow students to generate their own custom decodable book. We saw examples of students who were excited to use AI to generate their own stories but skipped the part where they actually read them—cases where the product missed an opportunity to provide instructional benefit.
Math
Benefits
Representations of thinking, not just answers:
Several products allowed students to write out or verbally record responses, allowing teachers to see into not only answers but also students' reasoning. This is a real contrast to the input modes of previous edtech products (e.g., students put their final answer in a box) and a meaningful acceleration in how quickly a teacher could collect data and track patterns across a full class of students at once.
Accurate, fast identification of student misconceptions:
AI-powered products can identify not only whether students got the right answer but where thinking is strong, where there are misconceptions, and what feedback is needed. We saw a few AI tools identify misconceptions in student thinking with more accuracy than we see in paper-pencil instruction and more quickly than teachers can walk around the room.
Easy ways to share student work:
AI-powered tools can make it easy for teachers to showcase students' work and explanations, so classes can see multiple examples of peers' work and thinking, helping students develop their own understanding.
Risks
Students juggling analog and digital materials:
Though some products allowed students to write or talk out their responses, many captured only the final answer. This both limited teachers' access to student thinking and created a clunky experience for students, juggling paper-pencil materials and devices simultaneously.
Misdiagnosed misconceptions:
AI-powered math products occasionally misdiagnosed the skills gap driving a wrong response and asked students to spend time relearning a topic they already clearly understood, which not only poses an opportunity cost in terms of time spent needlessly reviewing material, but could also lead to student disengagement and muddled teacher understanding of students' needs.
Misalignment with a curricula:
Some products offered a wide array of activities for students to work through, but limited—or even no—alignment with particular curricula. This lack of alignment with the overall instructional experiences increases the risk of student learning experiences that feel disconnected, asking students to juggle competing product schemas and taxonomies, without ever connecting the dots.