A small brain imaging study has found that six- and seven-year-olds show less communication within and between the brain networks responsible for self-regulation and focused attention when brainstorming with a conversational AI than adults do. While both groups performed similarly on a standard creativity test outside the scanner, only adults showed a link between higher creativity and stronger coordination among these cognitive control systems during the chat with the AI.
The research, published in Cerebral Cortex, suggests that young children may not yet have the mature brain wiring needed to actively direct and evaluate a collaborative dialogue with generative language models. The study looks at how children and adults engage with a tool similar to ChatGPT as they brainstorm ideas in real time, a process known as co-creativity.
Generative AI programs, like the one used in the study, rely on large language models trained on vast text to predict how words fit together and to generate ideas. As such tools increasingly appear in homes and classrooms, the ability to plan, stay focused, remember what has been said and filter distractions becomes essential to productive collaboration with AI—especially for younger users still developing executive functions.
Past neuroimaging work in adults has shown that more original ideas tend to emerge when three major brain networks work in concert: the network linked to spontaneous thought and memory, a network that flags important new thoughts, and a cognitive control network that handles goal-directed planning. When these systems coordinate, a person can brainstorm freely and simultaneously judge whether an idea makes sense.
By contrast, cognitive control networks remain maturing through adolescence into the mid-twenties. A 2019 study by the same team found that older children show stronger connections within and between these networks while listening to stories, and a 2023 study reported developmental shifts in how these networks relate to creativity, with stronger control-region connectivity linked to higher creativity in adolescents but lower creativity in adults.
The current study focused on first graders—average age about 6.4 years—and a comparison group of adults averaging 38 years. All participants had typical verbal and nonverbal intelligence and came from average to above-average socioeconomic backgrounds. Before scanning, children completed a Divergent Association Task, asking them to name 10 words that are as unrelated in meaning as possible.
Inside an fMRI scanner, participants engaged in a five-minute spoken conversation with ChatGPT. They were given the same open-ended goal for both groups: plan the opening of any kind of store, anywhere and for anyone, by talking aloud with the AI. Children were told they would be speaking to “the smartest computer in the world,” while adults were simply told they would be speaking to ChatGPT.
The researchers recorded and transcribed every conversation to assess how original and varied the dialogue was. They measured divergent semantic integration—the extent to which meanings between words diverge within a conversation—and, on the brain scans, mapped functional connectivity across seven major networks while accounting for head movement inside the scanner.
The seven networks included the frontoparietal and cingulo-opercular networks (cognitive control), the dorsal and ventral attention networks (focused and automatic attention), the salience network (idea selection), the default mode network (imagination), and the memory retrieval network.
On the outside-the-scanner creativity test, children and adults scored similarly for their age groups, with average percentile rankings not differing significantly. During the ChatGPT conversation, adults produced slightly more diverse and wide-ranging narratives, reflected in a marginally higher semantic integration score (0.82) than the first graders (0.81).
Brain analyses revealed notable age-related differences. Within individual networks, adults showed markedly stronger internal synchronization in the frontoparietal control network (0.201) than children (0.113); similar advantages appeared in the salience (0.289 versus 0.198) and dorsal attention networks (0.211 versus 0.162). Only the frontoparietal difference remained after correcting for multiple comparisons.
Internal connectivity within the default mode network did not differ significantly between groups (0.187 in adults, 0.181 in children). Cross-network communication was also greater in adults, with the strongest difference seen in the connectivity between the frontoparietal control network and the dorsal attention network (0.126 in adults versus 0.039 in children) after adjustments for multiple tests.
Before such corrections, adults also showed stronger links connecting memory retrieval regions to cognitive control and attention systems, as well as stronger coupling between the default mode network and memory retrieval. The researchers then asked whether baseline creativity scores predicted how strongly brain networks communicated during ChatGPT dialogue; in adults, higher creativity was associated with stronger frontoparietal–salience connectivity, a pattern not observed in first graders.
Horowitz-Kraus described the findings as surprising in two respects. “First, children and adults looked similar in creative ability on the standard test, yet the neural route they took during the conversation was quite different; the difference lay in how creativity is implemented in the brain, not in baseline capacity. Second, the default mode network, which is central to most models of creativity, did not differ between the groups. One possible explanation is that conversing with ChatGPT in a scanner was novel for everyone, but that needs to be tested directly.”
The study’s authors conclude that even when young children show age-appropriate creative potential, their control, attention and memory networks engage in a less integrated way during conversations with a generative AI than adults do. They suggest that adults may use the salience and frontoparietal networks to filter ChatGPT’s rapid responses and steer brainstorming toward a goal—roles that six- and seven-year-olds may not yet perform.
“Talking to a chatbot looks effortless, but it draws on planning, monitoring, attention and memory systems that are still under construction in a six-year-old, which is why adult guidance matters when young children use these tools,” Horowitz-Kraus said.
She emphasised the practical message for families and educators: “The practical message is that young children should not be left to use generative AI on their own. For parents and teachers, this means sitting with the child, asking questions about the answers, and treating the AI as a shared activity instead of an independent tutor. For educators, pediatricians and policy makers, it means that the age at which children start using these tools deserves the same careful attention we already give to screen exposure. This study cannot set a specific ‘safe age’ since it focused on two very distinct age groups, but it is a clear signal that tools designed around adult abilities should not simply be handed to young children.”
The researchers cautioned that the study’s small sample size and design limit the ability to draw broad conclusions. The children and adults were not tested against a human partner or a non-AI task in the scanner, making it hard to separate general developmental differences from AI-specific effects. The sex distribution was uneven, with 11 boys and four men participating, and several other potential confounds—such as vocabulary size, speech rate, turn-taking, and prior experience with the AI—were not controlled.
Looking ahead, the team plans larger samples, direct comparisons with human interaction, and longitudinal follow-up to determine when control networks mature toward adult-like engagement and how that relates to creativity. They also aim to assess children’s familiarity with AI and how the AI’s perceived authority shapes engagement, with the broader aim of guiding developmentally appropriate use of generative AI in education and healthcare settings.
“Technology can support learning, but it cannot lead it; children learn best alongside an adult who guides and challenges them,” Horowitz-Kraus added.
The study, “Lower engagement of cognitive control in children on ChatGPT,” is by Tzipi Horowitz-Kraus, Yoed N. Kenett, Dafna Link-Sourani, Emeel Ashqar and Rola Farah.
