- Computer scientists developing artificial intelligence want their technology to be more like a child’s brain.
- Children’s brains are great at collecting information and learning from cues in the world around around them, something that AI systems struggle with.
- A cognitive development expert, along with AI specialists from Alphabet’s DeepMind and Stanford discussed how a better understanding of the child’s brain could provide the blueprint for the next generation of AI.
STANFORD, California — At first take, it might not be clear why an expert in children’s cognitive development was a featured speaker at a conference on artificial intelligence.
It turns out that kids — and those who understand how they learn — may have a lot to teach the experts in artificial intelligence about how to improve their systems.
AI systems have gotten good and swallowing gobs of data and using it to make predictions based on all that information, said Alison Gopnik on Monday during a panel discussion at an event here sponsored by the Stanford Institute for Human-Centered Artificial Intelligence. But they’re not very good at generalizing from small amounts of data, said Gopnik, a professor who studies cognitive development at the University of California, Berkeley. Nor are they good at collecting data on their own to make those generalizations or at learning about the world from the cues given by other intelligent entities around them, she added.
But babies and young children excel at all those things, she said.
“So those three things — model building, exploration, social learning — are some clues to how children can learn so much, and those are things that are just at the beginning in terms of what AI can do,” said Gopnik, the author of “The Scientist in the Crib: What Early Learning Tells Us About the Mind.”
Read this: AI could soon be all around us — here’s how that could upend 8 different industries
AI researchers are already using what psychology experts such as Gopnik have discovered about the way children learn and applying them to their field. Gopnik herself is working with researchers at her university to develop AI systems that are meant to be curious, like kids. They’re designed to go out and collect data on their own, she said.
One of the key insights that’s helped AI research advance so rapidly in the last five to 10 years has been the realization that they needed to design an actual curriculum, a teaching program for their systems, said Demis Hassabis, cofounder of Alphabet-owned AI lab DeepMind, who sat on the same panel as Gopnik. They couldn’t expect their systems to master tasks immediately, but had to allow the systems to build up to those tasks incrementally by mastering steps along the way, Hassabis said.
“You can’t just go from zero to one,” he said. “You actually need to do easier versions of the task and build up in the way that we teach children.”
One of the promises of patterning AI after the way children’s brains develop is that the technology could be much more efficient. Instead of relying on huge data sets and lots of computing power to make sense of the world, such child-like AI systems could potentially rely on much less data and power, said Chris Manning, a professor of linguistics and computer science at Stanford’s Artificial Intelligence Laboratory.
“You can get this orders of magnitude more efficient learning,” Manning said.
Got a tip? Contact this reporter via email at firstname.lastname@example.org, message him on Twitter @troywolv, or send him a secure message through Signal at 415.515.5594. You can also contact Business Insider securely via SecureDrop.
SEE ALSO: AI is great at recognizing nipples, Mark Zuckerberg says
Join the conversation about this story »
NOW WATCH: Why Amazon canceled HQ2 in New York City and what it means for everyone involved