Artificial Intelligence Research Draws From the Mind of Babies

Abhinav Raj, Writer
@uxconnections

Is our artificial intelligence (AI) technology any better at understanding the world than an infant? Researchers at Trinity College Dublin debate in newly published research.

As humans, we are natural-born explorers of the world around us—right out of the womb, and from within. 

“We enter the world ready to ‘read the perfect cues out of the environment”, described Kathy Hirsh-Pasek, professor of psychology specializing in early child development at Temple University, in a 2018 interview with the Guardian.  

Babies actively learn from the environment around them. From the moment they are born, they perceive the world around them through their sense of hearing, smell, sight and touch in unsupervised settings. Research suggests that at just six months of age, infants can comprehend cause and effect relationships—such as understanding that pushing a button on a toy makes it play music. 

Modern unsupervised machine learning (ML) algorithms suffer from critical limitations that inhibit them from extrapolating meaningful information and learning from unlabelled data in their surroundings—a task that an average baby actively excels at. Conceptually, an infant’s learning process is similar to how artificial unsupervised learning occurs—by interpreting vast amounts of unlabelled data from the surroundings. 

Most ML algorithms require structured and labelled datasets to learn. (Image: Canva)

Experts are examining the ability of artificial intelligence (AI) to learn in unsupervised learning environments by drawing parallels to infant learning. 

In their recently published research ‘Lessons from infant learning for unsupervised machine learning’, Trinity College Dublin research fellows Lorijn Zaadnoordijk and Tarek R. Besold describe how the new and improved generation of artificial intelligence algorithms can benefit from richer datasets that allow for a deeper insight into our world that can afford our AI tools a steeper developmental trajectory while overcoming the dependence on labelled datasets. 

“As AI researchers we often draw metaphorical parallels between our systems and the mental development of human babies and children”, argues Besold. 

“It is high time to take these analogies more seriously and look at the rich knowledge of infant development from psychology and neuroscience, which may help us overcome the most pressing limitations of machine learning.”

The research appears in the machine learning and artificial intelligence journal, Nature Machine Intelligence.  

Modern unsupervised learning algorithms lack neuroplasticity—that is, the brain’s ability to modify, adapt or change structurally or functionally in response to the environment through experiences. 

Researchers suggest that an AI model inspired by infant learning may be able to benefit from perceiving the world through data sets that present a fuller, more comprehensive view of the world, potentially allowing machines to discover how the world looks, sounds, and ‘feels’ like. 

But until then, babies will have an edge over AI.