Babies outperform AI at detecting what motivates other people’s actions, according to a new study by a team of psychologists and data scientists. Highlighting the fundamental differences between cognition and computation, their findings point to shortcomings in current technology and where AI needs to be improved to more precisely replicate human behavior.

“Adults and even infants can easily make reliable inferences about what drives other people’s actions,” explains Moira Dillon, assistant professor of psychology at New York University and senior author of a paper published Feb. Information. “Modern AI is difficult to draw these conclusions from.”

“The new idea of ​​pitting babies and AI against the same tasks allows researchers to better identify babies’ intuitive knowledge of other people and suggest ways to integrate this knowledge into AI,” he adds.

“If artificial intelligence is to create the resilient, discreet thinkers that adults become, then machines need to rely on the core abilities babies have to set goals and preferences,” says Brenden Lake, an associate professor at the Center for Data Science. and Department of Psychology, New York University. and one of the authors of the article.

Babies are well known to be attracted to other people, as evidenced by the time they spend looking, observing their movements, and interacting with others. Additionally, previous research focusing on infants’ “common sense psychology” – their understanding of the intentions, goals, preferences, and rationality that underlie others’ actions – has shown that infants can attribute goals to others and expect others to pursue those goals. rationally and rationally. The ability to effectively make such predictions is the foundation of human social intelligence.

Conversely, “healthy artificial intelligence” driven by machine learning algorithms directly predicts actions. So, for example, after you read the news about a newly elected city official, an ad appears on your computer screen that San Francisco is a tourist destination. But what AI lacks is the flexibility to recognize different contexts and situations that drive human behavior.

To develop a basic understanding of the differences between human and artificial intelligence, the researchers conducted a series of experiments with 11-month-old babies and compared their responses with those obtained using cutting-edge learning-driven neural network models.

To do this, they used the previously established “Comparison of Children’s Intuitions” (BIB), six tasks that examine common sense psychology. The BIB is designed to allow comparisons between infant and machine performance, allowing testing of both infant and machine intelligence, and more importantly, providing an empirical basis for building human AI.

Specifically, the babies on Zoom watched a series of videos with simple animated figures moving around the screen, similar to a video game. The figures’ actions simulated human behavior and decision making by searching for objects and other movements on the screen. Similarly, the researchers created and trained learning-based neural network models (artificial intelligence tools that help computers recognize patterns and simulate human intelligence) and tested the models’ responses to the same videos.