Working with computer scientists and cognitive researchers, psychologists have constructed a framework for training AI to learn like human infants.
The project may inform theories of human neurodevelopment as dynamically as it advances computer and data science.
Already it’s shown the human mind to be light years ahead of machine learning when it comes to building crucial aspects of social intelligence, known in some contexts as “commonsense psychology.”
This complex process largely consists of being able to deduce individuals’ motivations, goals and preferences so as to predict what action they intend to take next.
Despite the yawning gap between human and machine learning—or perhaps thanks to it—the present work promises to help flesh out “the origins and development of human common sense and provide an avenue for building the future of human-like AI.”
So state Moira Dillon, PhD, and colleagues at New York University in a study published online Feb. 16 in the journal Cognition.
Banking on baby’s intuition
Using a neuropsychological battery of tests called the Baby Intuitions Benchmark, or “BIB,” the team evaluated sophisticated algorithms alongside infants under 1 year old.
The BIB methodology administers a set of six tasks that, together, gauge the development of commonsense psychology.
In their study discussion, Dillon and co-authors summarize a key finding from the BIB exercise that deserves revisiting as commonsense AI advances in coming years:
Our comparison reveals that the state-of-the-art ‘machine theory of mind’ … is indeed missing key principles of commonsense psychology that infants possess.”
Restating this finding for NYU’s news office, Dillon underscores that adults and even infants “can easily make reliable inferences about what drives other people’s actions. Current AI finds these inferences challenging to make.”
Born to figure things out
The study further showed babies’ ability to adapt to changes in test settings.
For example, the young participants were adept at predicting next moves of simple animated shapes that barely resembled humans. And this held when changes were made to the shapes’ location and surrounding environment.
“A human infant’s foundational knowledge is limited, abstract and reflects our evolutionary inheritance,” Dillon comments, “yet it can accommodate any context or culture in which that infant might live and learn.”
Study co-author Brenden Lake, PhD, points out:
“If AI 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.”
Click here to read the full study and here for the NYU news item.
Dave P. has worked in journalism, marketing and public relations for more than 30 years, frequently concentrating on hospitals, healthcare technology and Catholic communications. He has also specialized in fundraising communications, ghostwriting for CEOs of local, national and global charities, nonprofits and foundations.