Infants outsmart artificial intelligence in commonsense psychology


Paul Inkles

Paul Inkles | Flickr

Melani Bonilla, Multimedia Editor

Artificial intelligence has been recently trending and recognized as a movement among the public. With creations such as ChatGPT, many people are exposed to artificial intelligence and its abilities more frequently.

To measure AI abilities, scientists and researchers are conducting experiments to test AI features in comparison to humans.

A study conducted by New York University found that infants are more successful than artificial intelligence at inferring people’s motivations. Researchers conducted a series of experiments with 11-month-old infants and compared the infants’ responses to that of AI.

These researchers decided to use infants because they are highly observational. Although infants are still developing and trying to understand the world around them, they still hold a level of common sense that increases with age.

This level of understanding in infants can be referred to as “commonsense psychology.” This is an understanding of goals, intentions and motivations in connection to actions. This ability in infants sets the foundation for their emotional intelligence as they grow.

To conduct the experiment, researchers used the Baby Intuitions Benchmark. Using research on infant cognition, “BIB challenges machines to achieve generalizable, common-sense reasoning about other agents like human infants do,” according to a research paper.

BIB allows scientists to directly compare infants and AI. By using video sequences, researchers observed whether artificial intelligence can accurately comprehend and predict human behavior and compared it to the comprehension of that of infants.

Researchers at NYU showed infants on Zoom animated shapes that stimulated human behavior. When shown to the artificial intelligence models, they did not display the same cognitive understanding of behavior and motivations.

Although the artificial intelligence lacked the understanding of human motive and behavior, there was still “commonsense AI.” This type of commonsense in artificial intelligence refers to its ability to make connections between things like someone’s search history and someone’s interests. This is a direct prediction of actions, but it does not recognize the motive behind someone’s actions.

“If AI aims to build flexible, commonsense thinkers like human adults become, then machines should draw upon the same core abilities infants possess in detecting goals and preferences,” said Brenden Lake, an assistant professor in NYU’s Center for Data Science and Department of Psychology.

Artificial intelligence companies can draw on this study and suggestions from researchers such as Lake to improve the abilities of their models and technology. It is a question of whether these companies can bring human complexity to the machines they create.

AI-based machines are fast, more accurate, and consistently rational, but they aren’t intuitive, emotional, or culturally sensitive,” commented Harvard Business Review. “And, it’s exactly these abilities that humans possess and which make us effective.”