As AI continues to develop rapidly, there is a growing curiosity about whether its capabilities will surpass those of humans. Many are wondering if there is already an equation between the human mind and AI machines. However, the reality is that AI capabilities remain limited.
There is one particular example to demonstrate it. According to a recent study conducted by a psychology and data science team at NYU, infants perform better than artificial intelligence in identifying the motivations behind other people's actions.
To gain insight into the differences between human and AI capabilities, a group of researchers conducted a series of experiments comparing the responses of 11-month-old infants to those of state-of-the-art neural network models driven by machine learning. The researchers used the "Baby Intuitions Benchmark" (BIB), a set of six tasks designed to probe commonsense psychology, to test infant and machine intelligence. This approach allowed for comparing performance between the two and, importantly, provided an empirical foundation for building AI that mimics human intelligence.
During the experiments, the infants watched videos of simple animated shapes moving around the screen in ways that simulated human decision-making and behavior. The same videos were also presented to the neural network models built and trained by the researchers. The results showed that even in the simplified actions of animated shapes, infants could recognize human-like motivations and predict that consistent goals drove these actions. They demonstrated this ability by looking longer at events that violated their predictions, a well-established method for measuring infants' knowledge. However, the neural network models showed no evidence of understanding the motivations underlying such actions, indicating that they lack some of the foundational principles of commonsense psychology that infants possess.
The fascination of infants with other individuals is well-known, as evidenced by their extended observation of others' behavior and their desire to interact socially. Research conducted on infants' comprehension of "commonsense psychology," which involves understanding the intentions, goals, preferences, and rationality behind the actions of others, has shown that infants can attribute goals to others and anticipate their pursuit of goals logically and effectively. These predictive skills are crucial to human social intelligence.
Lead author Sara Dillon commented that the human infant's foundational knowledge is limited, abstract, and reflects our evolutionary inheritance. Yet, it can accommodate any context or culture in which that infant might live and learn. The paper's co-authors were Gala Stojnić, Kanishk Gandhi, and Shannon Yasuda, all affiliated with NYU during the study.
Moira Dillon, an assistant professor in New York University's Department of Psychology and the paper's senior author, which appears in the journal Cognition, explains: "Adults and even infants can easily make reliable inferences about what drives other people's actions. Current AI finds these inferences challenging to make. The novel idea of putting infants and AI head-to-head on the same tasks is allowing researchers to better to describe infants' intuitive knowledge about other people and suggest ways of integrating that knowledge into AI".
The study's findings underscore the fundamental distinctions between Cognition and computation and highlight the inadequacies in today's technologies, indicating areas where AI needs to be improved to achieve a more comprehensive replication of human behavior. The current AI models have limitations in replicating human behavior comprehensively because human behavior is complex and varies depending on the situation, context, and individual perspectives.