NEW YORK – Scientists at New York University have embarked on an ambitious project to understand how children learn language by using the footage from a headcam worn by a baby named Sam. The baby’s weekly sessions were recorded and then fed into an artificial intelligence (AI) program, posing the question of whether a simple AI can learn language from a child’s experiences.
Published in the journal “Science,” the researchers reported that the AI, after being exposed to a small fraction of Sam’s experiences, was able to match words with their corresponding images. This research is part of a bigger quest to eventually create an AI that mimics a baby’s mind, potentially leading to advancements in our understanding of human development and the creation of more intuitive AI systems.
Unlike “chatbots” or “large language models” which are trained on vast amounts of text, babies learn language from sensory input and play in their environment. The researchers wanted to find out if the AI could pick up basic language elements from a child’s experiences without any prior knowledge of grammar or other social abilities.
The study utilized footage from the SAYCam database, which contains audiovisual recordings from individual babies and toddlers. Families contributed to the database, providing a unique window into the cognitive development of children between 6 and 32 months old. In Sam’s case, 600,000 video frames and 37,500 transcribed words of training data were used to train the AI.
It was found that the relatively simple neural network was able to match basic nouns and images with similar accuracy to AI trained on 400 million images from the web. This raises questions about the existing theories of how humans learn language and whether the AI’s learning abilities are comparable to those of a small child.
One of the lead researchers, Brenden Lake, acknowledged that the AI’s learning deviates from that of children, as it excelled at learning certain words but struggled with others. Nonetheless, the project marks an important step in understanding the cognitive skills humans need to learn language and lays the groundwork for further research in this field.