AI analysis of babies' movements reveals important insights into early stages of development, highlighting the importance of foot movements in learning. Recent advances in computing and artificial intelligence and new insights into infant learning suggest that machine and deep learning techniques can be used to study how infants transition from random exploratory movements to purposeful action. To date, most studies have focused on infants' spontaneous movements and distinguished between fidgety and non-fidgety behaviors.
Although an infant's early movements may seem disjointed, they reveal meaningful patterns in the infant's interaction with the environment. However, we still lack an understanding of how infants consciously interact with their surroundings and the principles that guide their goal-directed actions.
To explore how babies begin to act with purpose, researchers at Florida Atlantic University and their collaborators conducted a baby cell phone experiment, a developmental research technique that has been used since the late 1960s. In this experiment, a colorful mobile phone was gently tethered to a baby's foot and moved when the baby kicked, thus linking the baby's behavior to what they were seeing. This setup helps researchers understand how babies control their movements and discover their ability to influence their surroundings.
In the study, researchers tested whether artificial intelligence tools could capture complex changes in infants' movement patterns. Infant movements tracked using the Vicon3D motion capture system are classified into different types - from spontaneous movements to reactions while moving. By applying various artificial intelligence techniques, the researchers studied which method best captured the subtle behaviors of babies in different situations and how movements evolved over time.
The findings, published in Scientific Reports, highlight artificial intelligence as an important tool for understanding early infant development and interactions. Both machine learning and deep learning methods accurately classified five seconds of three-dimensional clips of infant movements into different stages of the experiment. Among these methods, the deep learning model 2D-CapsNet performs best. Importantly, foot movement had the highest accuracy of all tested methods, meaning that the movement patterns of the feet changed most significantly during the various stages of the experiment compared to other parts of the body.
"This finding is significant because the AI system was not told anything about the experiment and did not know which part of the baby's body was connected to the phone," said study co-authors Glenwood Krich and Martha Krichko of Florida Atlantic University's Center for Complex Systems and Brain Sciences. Dr. Scott Kelso, Distinguished Scholar in Science, said: "This suggests that the feet - as end effectors - are most affected by interaction with the phone. In other words, the way the baby connects to the environment is most affected at the point of contact with the world. Here, it's 'feet first'."
The 2D-CapsNet model achieved an accuracy of 86% when analyzing foot movements and was able to capture the detailed relationships between different body parts during movement. Foot movements consistently had the highest accuracy of all tested methods, approximately 20% more accurate than hand, knee, or whole-body movements.
"We found that infants explored more after being disconnected from their phones than before they had the opportunity to control their phones. It seems that losing the ability to control their phones made them more eager to interact with the world to find ways to reconnect," said co-author Aliza Sloan, Ph.D., a postdoctoral research scientist at the Center for Complex Systems and Brain Sciences at Florida Atlantic University. "However, some infants showed movement patterns during the disconnection phase that contained cues from their previous interactions with the phone. This suggests that only some infants understand their relationship with the phone well enough to maintain these movement patterns in the expectation that they will still generate responses from the phone even after disconnection."
If the accuracy of the baby's movements remains high during disconnection, it may indicate that the baby learned something during the previous interaction, the researchers said. However, different types of movements may mean that babies discover different things.
"It's important to note that studying infants is more challenging than studying adults because infants cannot communicate verbally," said co-author Dr. Nancy Aaron Jones, a member of the Center for Brain Science. "Adults can follow instructions and explain their actions, but infants cannot. That's what people do This is where AI can help. AI can help researchers analyze subtle changes in babies' movements, even their resting states, to give us insights into how they think and learn, even before they can talk, and it can help us understand the huge individual differences that occur as babies grow."
Observing how the AI's classification accuracy changes for each infant gives researchers a new way to understand when and how babies begin to engage with the world.
"Past AI methods have primarily focused on classifying spontaneous movements that correlate with clinical outcomes, and combining theory-based experiments with AI will help us better assess infant behavior in relation to the infant's specific environment," Kelso said. "This could improve how we identify risk, diagnose and treat disease."
Compiled from/SciTechDaily