AI can accurately identify baby's age and gender based on "temperament"

It can be difficult to tell whether a newborn is a boy or a girl based on appearance alone, as well as behavioral traits such as whether the baby is afraid, smiling, or laughing. But once the baby is about a year old, things start to change and the different "temperaments" of different newborns begin to emerge.

A new study in PLOS ONE used machine learning to analyze "temperament" data from 4,438 infants, attempting to categorize the infants by gender and age.

The results showed that it was much easier for a computer algorithm to determine a baby's age based on temperament data during the first 48 weeks of life than to decipher the baby's sex. However, once babies were older than 48 weeks, gender differences in infancy became more prominent during this time, given that the gender classification algorithms improved.

"This at least suggests a picture where, around one year of age, temperament starts to differ by sex in more robust ways," said Maria Gartstein, a professor of psychology at Washington State University and lead author of the study.

Previous studies have investigated age- and sex-based differences in temperament in infants, but few have looked at the two variables together. Gartstein says this is largely due to the difficulty for individual labs to collect enough infant behavior data to make their findings statistically reliable and relevant to a broad enough population.

To overcome this challenge, Gartstein and her colleagues reached out to more scientists and pooled data from the Infant Behavior Questionnaire collected between 2006 and 2019. The questionnaire is a parent-reported measure of temperament that asks parents to record the frequency of 191 different behaviors their children exhibit from 3 to 12 months of age. The data can then be used to score infants on 14 different temperament dimensions, such as smiling, activity level, anger/frustration, and fearfulness.

Although many methods have been developed for measuring child temperament, including various observational procedures and physiological techniques, overall parent report remains the most widely used.

Rothbart's psychobiological model is generally considered to be the most widely accepted theory or framework of temperament. This approach views temperament as individual differences in reactivity and self-regulation based on constitution, which refers to the relatively enduring biological makeup of an individual, influenced by genetics, development, and experience; reactivity, which refers to the arousability of emotional, motor, and attentional responses, assessed by threshold, latency, intensity, time to peak intensity, and recovery time of the response; and self-regulation, which embodies processes that can be used to regulate reactivity, such as soothing and inhibitory control.

Ultimately, Gartstein and his colleagues collected data on 2,298 boys and 2,093 girls. For the analysis, co-author Erich Seamon of the University of Idaho used a machine learning algorithm to rate infants aged 0-24 weeks, 24-48 weeks, and older than 48 weeks as male or female on 14 temperament dimensions. Accuracy increased with age, ranging from 38% in age group 1 to 57% in age group 3.

“This was a cool opportunity to do a demonstration study using these machine learning techniques, which require very large data sets and are not common in research on social-emotional development,” Gartstein said. “It gave us the first opportunity to really consider the extent to which sex differences are influenced by the age of the infant.”

The results of the researchers' analysis showed that fearfulness was the most important characteristic that differentiated boys and girls in the youngest and middle age groups. As infants got older, responsiveness, or the ability to recover quickly from high-stress situations and showing more active interactions and willingness to engage with people and objects, became more influential.

For infants older than 48 weeks, low-intensity play or enjoying familiar calming activities (e.g., playing peekaboo with a parent) was the most influential variable differentiating boys from girls. Overall, girls were higher in fear, decreased reactivity, and low-intensity pleasure, while boys were higher in external contact.

Interestingly, certain temperamental traits reduced the accuracy of the machine learning algorithm in classifying babies by gender, specifically cuteness, vocal responsiveness, smiling, and laughing in the youngest age group, and smiling, laughing, perceptual sensitivity (e.g. noticing very subtle changes), and activity in the oldest age group.

While many factors could have influenced the researchers’ pattern of results, their work is consistent with previous research showing that the effects of socialization do begin to take effect around the first year of life.

"Mothers take different approaches to socializing their sons and daughters, and these differences can lead to different trajectories in temperament over time," Gartstein said. "Specifically, parents may prioritize relationship orientation in daughters and competence and autonomy in sons."

Looking ahead, Gartstein said she and her collaborators will next use the machine learning methods developed in the current study to investigate other hard-to-answer questions about infants’ social-emotional development.

"What I'm really interested in now is seeing if I can predict differences in the quality of care based on the children's brain activity," Gartstein said. "The analytical approach we developed for this study is particularly powerful when answering questions that rely on multiple input variables to solve classification problems."

References:
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0266026
https://techxplore.com/news/2022-05-ai-infant-age-gender-based.html