Researchers at New York University say infants are better at detecting the motivation behind an individual’s actions than artificial intelligence models.
In a paper published in the Cognition journal, researchers noted that infants outperform AI at inferring the reason behind people’s actions. This study shows the significant difference between computation and cognition.
The paper’s senior author, Moira Dillon, said:
“Adults and even infants can easily make reliable inferences about what drives other people’s actions. Current AI finds these inferences challenging to make.”
Dillon, an assistant professor at NYU’s Department of Psychology, says the research aims to understand infants’ intuitive knowledge better and find ways to integrate that for AI.
One of the authors, Brenden Lake, added that the ultimate goal of AI is to develop into flexible, common sense thinkers like human adults. Therefore, “the machines should draw upon the same core abilities infants possess in detecting goals and preferences.”
AI models lack intuition
Researchers deployed the “Baby Intuitions Benchmark” (BIB) for the comparison. This test measures the common sense psychology of 11-month-old infants and AI models by making both watch videos of animated shapes. In the videos, these shapes simulated human behavior and decision-making by making several movements and retrieving objects.
Infants’ responses to the videos differed significantly from the AI models. According to the study, infants recognize human-like motivations and predict there’s a reason behind the movements of the animated shapes.
They demonstrated this by spending more time looking at events that violated their predictions, a phenomenon known as the “surprise paradigm.”
But AI models did not exhibit the same response and showed no evidence they understood the motivation behind the actions.
Long way to go for AI
AI tools have been all the rage in recent months. This is understandable due to the abilities of tools like ChatGPT that can hold conversations, write programming codes, pen lyrics, and compose poems, among other things.
While the hype around the tech is justifiable, events show there is still plenty of room for growth. In recent weeks, AI tools such as Google’s Bard and Microsoft’s ChatGPT-powered Bing have had public fumbles.
With this study, it is also clear that AI tools aren’t close to humans in terms of cognitive ability. At least not yet. Previous studies have established “common sense psychology” in infants that enables them to observe other people and understand their intentions and the reason behind their actions. This is a crucial ability that AI is not capable of replicating yet.
Common sense AI can directly predict actions but lacks the flexibility to understand the underlying motivations behind such human behaviors. This flexibility is fundamental to building AI tools that are more human.