Infants outshine artificial intelligence in determining what drives people's motivations, according to a new study published in the journal Cognition. The study, led by a team of psychology and data science researchers from New York University, highlights fundamental differences between cognition and computation, and points to the shortcomings in today's technologies. The study also suggests what improvements could be made by artificial intelligence to more fully replicate human behaviour. 


In a statement released by New York University, Moira Dillon, the senior author on the paper, said adults and even infants can easily make reliable inferences about what drives other people's actions. She added that current artificial intelligence finds these inferences challenging to make. 


Dillon also said that the novel idea of putting infants and artificial intelligence head-to-head on the same tasks is allowing researchers to better describe infants' intuitive knowledge about other people and suggest ways of integrating that knowledge into artificial intelligence. 


Brenden Lake, one of the authors on the paper, said if artificial intelligence 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. 


What is commonsense psychology?


Infants often look at others for long periods of time to observe their actions and engage with them socially, indicating how fascinated they are by other people. 


Studies have been conducted on infants' commonsense psychology, which refers to their understanding of the intentions, goals, preferences and rationality underlying others' actions. The studies have found that infants are able to attribute goals to others and expect others to pursue goals rationally and efficiently. This ability to make predictions forms the foundation of human social intelligence.


What is commonsense artificial intelligence? 


On the other hand, commonsense artificial intelligence, which is driven by machine-learning algorithms, predicts actions directly. This is the reason why a person who just read a news story on a newly elected city official of San Francisco might get an advertisement showing the same place as a travel destination. Or if a person searches for certain types of clothes online, they start getting advertisements for similar apparel on social media. 


The drawback of artificial intelligence is that it lacks flexibility in recognising different contexts and situations that guide human behaviour. 


How the study was conducted


The researchers conducted a series of experiments with 11-month-old infants and compared their responses to those yielded by state-of-the-art learning-driven neural-network models, in order to develop a foundational understanding of the differences in the abilities of humans and artificial intelligence. 


The researchers deployed the previously established "Baby Intuitions Benchmark", or BIB, to achieve this. As part of BIB, six tasks probing commonsense psychology are performed. BIB was designed in a way such that it could allow for testing both infant and machine intelligence. This allows for a comparison of performance between infants and machines, and significantly provides an empirical foundation for building human-like artificial intelligence. 


Infants on Zoom were asked to watch a series of videos of simple animated shapes moving around the screen, similar to a video game. The actions of the shapes simulated human behaviour and decision-making through the retrieval of objects on the screen and other movements. 


The researchers also built and trained learning-driven neural-network models, which are artificial intelligence tools that help computers recognise patterns and simulate human intelligence. The researchers then tested the responses of the neural-network models to the same videos the infants were made to watch.


Infants outperformed artificial intelligence


According to the study, the infants recognised human-like motivations even in the simplified actions of animated shapes. Infants were able to predict that these actions are driven by hidden but consistent goals. For instance, the infants were able to predict the on-screen retrieval of the same object no matter what location it was in and the movement of that shape efficiently even when the surrounding environment changed. Infants are able to demonstrate such predictions by observing events for long periods of time. 


The researchers observed that the neural-network models did not show any evidence of understanding the motivations underlying such actions. This revealed that artificial intelligence is lacking the key foundational principles of commonsense psychology that infants possess. 


Dillon said a human infant's foundational knowledge is limited, abstract, and reflects evolutionary inheritance, yet, it can accommodate any context or culture in which the infant might live and learn.