Can Babies Teach Us to Build the Next Generation of AI?

At Trinity College Dublin, a team of neuroscientists has just published new guiding principles for artificial intelligence, claiming that infants can help unlock the next generation of AI.

The research, appearing on the 26th in Nature Machine Intelligence , examines the neuroscience and psychology of infant learning and distills three principles to guide future AI development, aiming to overcome the most pressing limitations of current machine learning.

Dr. Lorijn Zaadnoordijk, a Marie Sklodowska‑Curie Fellow at Trinity, explains that while AI has made remarkable strides—delivering smart speakers, autonomous systems, intelligent apps, and enhanced medical diagnostics—its progress is now hampered because machine‑learning models rely on large, human‑curated datasets. Infants, by contrast, learn efficiently through experience, often mastering concepts after a single exposure.

In the paper “Training Unsupervised Machine Learning on Infant Learning,” Dr. Zaadnoordijk, Professor Rhodri Cusack, and Dr. Tarek R. Besold argue for better methods to learn from unstructured data. They identify specific insights from infant learning that can be applied to machine learning and provide concrete recommendations.

The authors assert that machines need built‑in preferences to shape their learning, richer datasets that capture visual, auditory, olfactory, gustatory, and tactile information, and a developmental trajectory that evolves as the system “grows,” mirroring how infants interact with their environment.

Dr. Besold adds that AI researchers often draw metaphorical parallels between their systems and infant or child development, and now is the time to study the psychological and neuroscientific knowledge of infant development to overcome AI’s most urgent constraints.

Professor Thomas Mitchell, director of the Trinity Institute for Neuroscience, notes that artificial neural networks are already brain‑inspired, but current implementations differ from how humans and animals learn. Interdisciplinary research, he says, can enable infants to help unlock the next generation of AI.