A recent NYU study compares the ability of babies and machines to understand human behavior.
Image from the office. (Courtesy of Graylin Lucas)
Children can know a lot more than we think. A recent NYU study shows that compared to artificial intelligence, babies are better at identifying the motivation behind human decisions.
Moira Dillon, senior author of the study and associate professor of psychology at NYU, said the goal of the study is to give artificial intelligence an example of basic human knowledge so that it can be used as a tool for comprehension-based tasks. She noted that the researchers wanted to develop a program that replicated infants’ knowledge of human behavior.
“An AI inspired by human intelligence will have in its repertoire the same abilities to understand other people that we humans have, and may be able to help us,” Dillon said. “It might be able to, for example, identify the needs of a person who is trying to act but can’t achieve their goal.”
Researchers studied 11-month-old babies to determine how different their ability to recognize human motivation might be from AI. To measure the difference between the two, they created the Baby Intuitions Benchmark, a six-task system designed to measure baby and machine intelligence.
Babies in the study responded to animated sequences of moving figures that mimic human behavior, which were recorded by the researchers and compared to AI responses to the same videos. The researchers also monitored human and AI responses when the animation deviated from predictable human behavior.
The researchers found that the AI models lacked the common-sense psychological instincts that babies had because the computer models could not predict the motivation behind the sequences.
Grace Lindsay, a professor of psychology and data science at New York University who was not involved in the research, said that despite the study’s findings, it is possible that artificial intelligence will one day be able to match the innate human ability to recognize motivation.
“The basic science of understanding human intelligence and translating it into machines in a way that really works is also difficult,” Lindsay said. “In this study, they built this task explicitly to be used by humans and machines, and that at least greases the wheels of that transmission.”
Dylan is not concerned about AI becoming “too human” as her findings suggest that AI is not good at basic reasoning and inference.
“The concern that artificial intelligence will be able to understand rich, multi-layered and multi-agent scenarios is still out of reach,” Dillon said. “I don’t think it’s impossible for artificial intelligence at some point to make these conclusions and reason, but even modern artificial intelligence has problems that we’ve exposed here.”
Contact Graylin Lucas at [email protected]
