Summary: When it comes to detecting what drives a person’s actions, infants outperform current artificial intelligence algorithms. The results highlight fundamental differences between computation and human cognition, highlighting shortcomings in current machine learning and identifying areas where improvements are needed for AI to fully replicate human behavior.
Source: NYU
Infants outperform artificial intelligence in detecting what motivates the actions of others, according to a new study by a team of psychology and data science researchers.
His findings, which highlight fundamental differences between cognition and computation, highlight gaps in current technologies and areas where improvements are needed for AI to more fully replicate human behavior.
“Adults and even infants can easily make reliable inferences about what motivates the actions of others,” says Moira Dillon, an assistant professor in New York University’s psychology department and lead author of the paper, which appears in the journal. Cognition. “Current AI finds these inferences difficult to make.”
“The innovative idea of putting infants and AI head-to-head on the same tasks allows researchers to better describe infants’ intuitive knowledge about others and suggest ways to integrate this knowledge into AI,” she adds.
“If AI aims to train flexible, common-sense thinkers like human adults become, then machines should rely on the same basic abilities infants possess to sense goals and preferences,” says Brenden. Lake, assistant professor in the Center for Data Science and Department of Psychology at NYU. and one of the authors of the article.
It is well established that infants are fascinated by others, as evidenced by the length of time they stare at others to observe their actions and engage socially with them. Moreover, previous studies focusing on infants’ “common sense psychology” – their understanding of the intentions, goals, preferences, and rationality underlying the actions of others – have indicated that infants are capable of assign goals to others and expect others to pursue their goals rationally and rationally. effectively. The ability to make these predictions is fundamental to human social intelligence.
Conversely, “common sense AI” – driven by machine learning algorithms – directly predicts actions. That’s why, for example, an ad featuring San Francisco as a travel destination pops up on your computer screen after reading a news story about a newly elected city official. However, what AI lacks is flexibility in recognizing the different contexts and situations that guide human behavior.
To develop a fundamental understanding of the differences between human and AI capabilities, the researchers conducted a series of experiments with 11-month-old infants and compared their responses to those provided by a neural network based on the state-of-the-art learning. models.
To do this, they deployed the previously established “Baby Intuitions Benchmark” (BIB) – six tasks probing the psychology of common sense. BIB was designed to enable testing of both infant and machine intelligence, allowing for comparison of performance between infants and machines and, significantly, providing an empirical basis for building AI like human.
Specifically, babies on Zoom watched a series of videos of simple animated shapes moving across the screen, like in a video game. The shapes’ actions simulated human behavior and decision-making through retrieving objects on the screen and other movements.
Similarly, researchers built and trained learning-driven neural network models — AI tools that help computers recognize patterns and simulate human intelligence — and tested the models’ responses to the same videos.
Their results showed that infants recognize human-like motivations even in the simplified actions of animated shapes. Infants predict that these actions are motivated by hidden but consistent goals, for example, retrieving the same object from the screen regardless of its location, and efficiently moving that shape even when the environment changes.
Infants demonstrate such predictions by looking longer at such events that violate their predictions – a common and decades-old measure for assessing the nature of infant knowledge.
Adopting this “surprise paradigm” to study artificial intelligence allows direct comparisons between an algorithm’s quantitative measure of surprise and a well-established human psychological measure of surprise: infant gaze time.
The models showed no evidence of understanding the motivations underlying such actions, revealing that they lack fundamental fundamentals of common-sense psychology that infants possess.
“A human child’s foundational knowledge is limited, abstract, and reflects our evolutionary heritage, but it can adapt to any context or culture that child might live and learn in,” observes Dillon.
See also
The other authors of the article are Gala Stojnić, an NYU postdoctoral fellow at the time of the study, Kanishk Gandhi, an NYU research assistant at the time of the study, and Shannon Yasuda, an NYU doctoral student. .
About this artificial intelligence research news
Author: Press office
Source: NYU
Contact: Press Office – NYU
Picture: Image is in public domain
Original research: Free access.
“Common Sense Psychology in Human Infants and Machines” by Gala Stojnić et al. Cognition
Abstract
Common Sense Psychology in Human Infants and Machines
Human babies are fascinated by others. They bring to this fascination a constellation of rich and flexible expectations about the intentions motivating people’s actions.
Here, we test 11-month-old infants and state-of-the-art learning-driven neural network models on the “Baby Intuitions Benchmark (BIB)”, a suite of tasks involving both infants and machines the challenge of making high-level predictions. on the underlying causes of the agents’ actions.
Infants expected agent actions to be directed toward objects, not places, and infants demonstrated default expectations of rationally efficient agent actions toward goals. Neural network models have failed to capture infant knowledge.
Our work provides a comprehensive framework in which to characterize infant commonsense psychology and is the first step in testing whether human cognition and human-like artificial intelligence can be constructed from the foundations postulated by cognitive and developmental theories.