Is it time to measure cognitive stunting?

For the past two years, I have been deeply engaged—through leading the Brookings Global Task Force on AI and Education—in examining generative artificial intelligence and what it means for students’ learning and development. Throughout this process—conducting interviews, reviewing studies, and debating with colleagues—one issue has consistently troubled me: how to accurately describe the risk AI poses to children’s thinking skills.

Generative AI is a learning tool like none we have ever seen. It can generate ideas, write, design, compose music and poetry, and train itself to improve. Precisely because of this power, many have sounded the alarm about its potential to encourage cognitive offloading among young people. Our task force raised this concern as well. In our report, we described the risk as follows:

“The risk stems from what study participants view as a straightforward progression: AI tools are more likely to foster dependence. As students increasingly use these tools, and ‘offload’ an increasing amount of their cognitive tasks to these tools, a positive feedback loop emerges where they see positive results in terms of grades and in time and effort saved. These outcomes then create increased dependence on AI tools, increased ‘cognitive offloading,’ and increased ‘cognitive decline’” (p. 56).

At the time, my coauthors and I debated how best to describe this phenomenon. In truth, we have never before encountered a moment in which, across so many domains, the task of thinking itself can be so easily outsourced. We ultimately used the terms cognitive offloading and cognitive decline because that is how participants described their experiences. But neither term fully captures the risk we are facing.

Human beings have always cognitively offloaded as we invent new tools. This is how I can walk into a grocery store rather than forage for food—something my distant ancestors surely could do better than I can. And, having one place where I can find food saves me so much time in my day for other tasks! From the wheel, which reduced the effort required to move people and goods, to the written word, which freed knowledge from the limits of memory and oral tradition, humans have continually adapted as tools reshaped how we use our minds.

What is different now is when and how offloading occurs.

When young people systematically use AI to complete schoolwork or other thinking tasks, they are not offloading skills they already possess. They are shortcutting the process of developing those skills in the first place. Nor is this best described as cognitive decline—a term typically used for age-related loss of function. Children are on the opposite side of the curve. They are building cognitive capacity and strengthening skills through sustained practice and experience.

Cognitive development, like learning to ride a bike, cannot be outsourced. Children cannot learn to ride by having an AI chatbot outline the steps or by having someone else pedal for them. In the same way, they cannot learn to think if AI routinely suggests ideas, writes text, designs solutions, or solves problems on their behalf. Struggle, practice, mistakes, observation, reflection, and adaptation are not inefficiencies to be optimized away; they are the core of learning. As the American Psychological Association defines it, cognitive development is “the growth and maturation of thinking processes of all kinds, including perceiving, remembering, concept formation, problem solving, imagining, and reasoning.”

Cognitive stunting

For these reasons, I have begun to describe the risk AI poses to children not as cognitive offloading or cognitive decline—or even cognitive debt or atrophy—but as cognitive stunting. I first made this argument publicly at the World Economic Forum in Davos in January and have increasingly found it to be an important framing.

The concern is straightforward: Overreliance on AI reduces children’s opportunities to engage in effortful thinking, thereby stunting their cognitive development. Students may never fully develop the suite of thinking skills that educators, and other caring adults in children’s lives, work to help them cultivate.

The analogy to physical stunting is instructive. Pediatricians in the United States routinely track children’s physical development, measuring height and weight against age-based norms—a practice that has been in place since 1977. Low growth rates are not diagnoses in themselves but signals that something may be wrong and warrants further investigation.

When children lack sufficient nutrients, particularly in early childhood, their physical growth can be impaired. Termed “stunting,” insufficient nutrition is a major cause, but chronic infection and lack of stimulation or caregiving can also cause stunting. When the body does not receive enough calories, protein, iron, or zinc, it prioritizes survival over growth. If unaddressed, childhood stunting can have long-term consequences, including lower physical stamina, reduced IQ, academic struggles, and lower lifetime earnings.

What if a parallel dynamic is emerging in relation to children’s technology use and cognitive development?

What if children’s wide use of AI—direct, unstructured interaction with general-purpose chatbots and AI companions not designed for learning, in effect “give-me-the-answer machines”—is reducing the number of effortful thinking moments required for healthy cognitive development and undercutting, for example, the develop of independent, critical thinking skills?

Learning scientists describe critical thinking as a set of interrelated processes, including building deep knowledge (you can’t think critically about nothing), analyzing ideas by identifying assumptions and relationships, evaluating evidence, and drawing reasoned conclusions. For young people who are still learning how the world works, these skills are difficult to acquire and need intentional, scaffolded practice. In our modern societies, education is one of the main ways in which children develop these higher order thinking skills, which help prepare them to navigate our complex systems.

Early evidence on AI and cognitive development

While it is still early in the development of generative AI and its uptake by children, emerging evidence raises legitimate concerns. Multiple studies suggest that heavy reliance on AI tools can negatively affect thinking abilities, particularly for younger users. One large study found that increased AI tool use was associated with lower critical thinking scores, especially among young people who relied on AI for cognitive tasks, compared with older participants. Educators are also reporting forms of “digital amnesia,” in which students struggle to remember information from assignments they have completed, even if asked about it shortly after they turn it in.

Another study monitoring brain activity during writing tasks found decreased neural engagement when participants used ChatGPT to write essays, compared with those who did not. The resulting work was of lower quality and reflected “diminished critical inquiry, increased vulnerability to manipulation, decreased creativity, and the internalization of shallow or biased perspectives.”

These outcomes, however, are not inevitable.

When AI is used narrowly—for example, intentionally deployed by teachers within high-quality pedagogy—it can support students’ thinking rather than replace it. A review of 84 empirical studies found that AI can be integrated as a support to student learning. Consistent with this, the Brookings Global Task Force found that narrow AI use, when embedded in strong instructional design and vetted content, can enrich learning. In contrast, wide AI use—unmediated interactions with general-purpose tools—poses greater risks to students’ learning and development.

Actions to limit cognitive stunting

No educator, parent, or pediatrician wants children to use AI in ways that undermine their cognitive development. Nor do employers or policymakers. Technology companies do not set out to harm learning either. Yet today there is no systematic way to assess which AI products, or patterns of use, may pose developmental risks.

This is where the concept of cognitive stunting can be useful. We can learn from the systems developed to track children’s physical stunting. Often referred to as a silent crisis or killer because it can be hard to detect, systematic measurement is an important foundation for being able to prevent and address stunting in children. Developing systematic ways to track and measure children’s healthy cognitive development, particularly in relation to AI use, is a good first step here too.   

Policymakers, particularly in the United States—where many widely used AI tools are developed—have an important role to play. Congress could convene hearings and request report language or studies to interrogate the feasibility of tracking cognitive stunting, including developing a clear definition, measurement framework, and reporting process. There are several questions that such an investigation could explore, including the below.

• How to define cognitive stunting? There are a range of existing measures tracking cognitive development that are used in different ways across the U.S. The NIH Toolbox has a set of validated and age-normed assessments, including for cognition, that is widely used by doctors, schools, and researchers for people ages 7 to 85+. They also have companion toolboxes for children between 3 and 6 years old. Are these, and other existing measures of cognitive development, sufficient for detecting the development of cognitive stunting? Would they capture the impacts of the possible decline in effortful thinking experiences from children’s “wide” AI use? Is there an existing measure or instrument that could serve as the equivalent of the height-for-age measure used to detect stunting in children? If not, should one be developed and if so, what should it be?
• How to systematically track cognitive stunting? Multiple organizations, programs, and existing longitudinal studies address the question of the healthy cognitive development of children. For example, the CDC runs the “Learn the Signs. Act Early. program” that provides caregivers, pediatricians, and early childhood educators a clear set of milestones for children’s development from 0-5 years, including their cognitive development. The goal is to notice any potential signs of developmental delays so that the appropriate screening, testing, and, where needed, intervention can be done early. In contrast, a consortium including multiple NIH institutes and centers runs the longitudinal Adolescent Brain Cognitive Development (ABCD) Study that tracks over 10,000 adolescents with the goal of better understanding the factors that shape adolescent development and creating “baseline standards for “normal brain development (similar to those that currently exist for height, weight, and other physical characteristics).” Are the current organizations and programs sufficient for systematically tracking cognitive stunting? Would a new consortium need to be developed or could cognitive stunting measures be included in existing mechanisms? How would data be accessed and acted upon if it did point to a problem of cognitive stunting in children?
• How to work with AI developers to limit children’s risk of cognitive stunting? AI presents a particular risk to children’s cognitive stunting given the ways in which general purpose AI chatbots and companions remove the friction in interactions, including at times the need for effortful thinking. What, therefore, are the appropriate approaches that could incentivize those developing AI products to ensure AI use doesn’t undermine children’s cognitive development? What patterns of AI use present the most risk to children’s cognitive stunting? Given that cognitive development is a complex process, how should we understand the role of children’s AI use within this process? How could we expect to see other factors, such as social media exposure or lack of physical activity or sleep, also contributing to children’s cognitive stunting?   

Children are resilient. With the right environments, guidance, and protections, they can adapt, grow, and flourish—even alongside powerful new technologies. But one thing is clear: we must ensure that, as educationalists like to say, children must learn to think before they learn to prompt.

The world young people in the U.S. are growing up in requires sophisticated thinking skills. Educators are on the frontlines of helping young people develop the knowledge, understanding and skills needed to successfully navigate our complex social, political, economic, and technological systems. Ultimately, whether in school or in activities outside of school, educators seek to help students not just learn to survive in our world but to thrive in it, deploying their skills to solve problems and improve their lives and communities. If children’s AI use does undermine their cognitive development, educators in and outside of school will be faced with a much harder job.