How the Science of Learning Shapes MathVoice
A correct answer does not always reveal understanding. Here is how the science of learning informs MathVoice’s approach to helping teachers see how students reason.
A student writes down the right answer. Does she understand it?
The answer to this question is not as straightforward as you might hope. A correct answer can hide a shaky idea, while a confident explanation may rest on a misconception. Skilled teachers know this, but diagnosing student thinking and adjusting instruction in the moment takes experience. Even the most experienced teacher cannot be part of every conversation in a busy classroom.
That's the challenge we set out to solve. Rather than starting with technology, we began by asking what learning research tells us about recognizing genuine understanding.
MathVoice grew out of that question. After completing collaborative, play-based math activities, students record short explanations of their thinking. MathVoice captures those explanations and uses AI to help teachers understand the reasoning behind them, so they can decide what to reinforce or revisit next.
What the science of learning tells us
The science of learning is the body of cognitive science research on how human memory and understanding work. It tells us that:
- Spacing practice out beats cramming;
- Retrieving an idea beats rereading it;
- Prior knowledge does a lot of the heavy lifting; and
- Durable learning depends on students thinking about meaning.
Decades of research have supported these principles across laboratory and real classroom settings. One important guide for our work is the recently updated second edition of Deans for Impact's The Science of Learning, which translates this research for educators.
Learning means thinking about meaning
Information becomes durable knowledge when students put effort into understanding its meaning. They explain how or why. They reorganize an idea or connect it to something they already know. Work that draws a student's attention to meaning moves learning out of working memory and into something that lasts.
In math classrooms, student talk can provide important evidence of that meaning. When a child explains how she approached a problem or why she chose a particular strategy, she is actively making sense of the math. These moments happen in dozens of conversations throughout the day, often beyond what one teacher can hear.
MathVoice captures and analyzes student math talk and surfaces the reasoning that usually gets lost in a busy room, giving teachers a window into how students are actually thinking, not just whether they reached the right number. The AI is purpose-built and entirely teacher-facing, helping educators interpret student thinking and deliver their next steps with confidence. The AI never interacts with students.
Research also shows that people do not always judge understanding accurately, whether they are learning or observing someone else learn. Fluency can resemble mastery. A student may complete a familiar procedure smoothly without understanding why it works. Researchers describe these misleading signs as “illusions of learning.”
MathVoice is designed to help teachers look beyond those surface signals. A tool focused only on engagement or correct answers can reinforce the surface-level signals teachers are trying to see beyond. MathVoice instead helps teachers look for evidence of genuine understanding, including misconceptions that may not be obvious from an answer alone.
One example from the research illustrates why this matters. A student reads "Lena has 7 apples and buys 5 more" and writes 7 − 5 = 2. The student appears to have selected an operation before making sense of the problem’s structure. The example points to the importance of helping students identify a problem’s structure before choosing a strategy.
MathVoice is designed to help teachers notice this kind of reasoning. By capturing what students say, MathVoice can help reveal the reasoning behind a wrong answer while there is still time for a teacher to respond.
“Sometimes I'll watch their video two or three times because it surprises me. They know things I didn't know they knew and I learn new ways of their thinking. And once I know their way of thinking, I can capitalize on that and really help them grow.”
— First grade teacher, Houston, TX
Holding edtech to a higher standard
This research also offers a useful standard for evaluating educational technology. Features that create the appearance of engagement are not valuable unless they support meaningful learning. A tool can keep students busy, light up an engagement dashboard, even nudge a score upward in the short term, without deepening understanding. By that standard, activity alone is not evidence of learning.
The central question is whether a tool helps teachers recognize and strengthen genuine understanding. That question guides how we develop MathVoice. Our goal is to give teachers a clearer view of student understanding, not simply more measures of activity.
Ultimately, the standard returns us to the question that opened this article. A student writes down the right answer. Does she understand it? Every teacher wants to know, and no teacher can ask thirty times at once.
MathVoice is not about using AI for its own sake. It is about extending teachers’ ability to notice how students are reasoning. Learning research shows the value of asking students to explain their thinking. MathVoice helps teachers make use of those explanations. When a student talks through a problem, her teacher can hear the reasoning even across a room full of other voices.
MathVoice does not grade students or replace teachers’ professional judgment. It helps educators notice student thinking they may not otherwise have had the opportunity to hear.
That is the standard guiding our work: technology should help teachers understand students more fully and respond more effectively.
Further Reading
- Deans for Impact's The Science of Learning (2nd ed., 2026) is the most accessible overview of this research.
- For a peer-reviewed companion, see Weinstein, Madan, and Sumeracki, "Teaching the Science of Learning" (2018).
- For the foundational evaluation of which study strategies actually hold up, Dunlosky and colleagues, "Improving Students' Learning With Effective Learning Techniques" (2013).
PowerMyLearning is a national nonprofit advancing evidence-based math education products and professional learning for educators.