Even a toddler is smarter than the smartest AI

Illustration for article titled Even a toddler is smarter than the smartest AI

Sure, when it comes to making calculations and accumulating knowledge, computers have long since left most of us in the dust. But for tasks involving even a little nuance and uncertainty, computers can't even compete with your average preschooler.

Indeed, when it comes to raw learning ability, young children arguably have no equal. Their ability to learn languages is basically incomprehensible to adults - even if we all once had the same capabilities - and young children possess a remarkable capacity to navigate and grasp even the basics of a world that is essentially alien to them. UC Berkeley developmental psychologist Alison Gopnik calls them "the greatest learning machines in the universe", and that's why she and her fellow researchers are hoping to share some of these abilities with computers.


This potential application grows out of a number of experiments - which, like any exploration of artificial intelligence worth its salt, have prominently involved "lollipops, flashing and spinning toys, and music makers", among others - that have shown that even babies possess the rudimentary capacity to test hypotheses, notice patterns, and use this information to draw conclusions.

In particular, young children's ability to adapt quickly to new information - such as being presented with an unfamiliar word - could help computers improve their human interaction skills in everyday AI situations like robotic answering services and computerized tutoring programs. Tom Griffiths, the director of UC Berkeley's Computational Cognitive Science Lab, says he is working on developing a statistical model that can translate the sorts of calculations and assumptions children make during learning into something that can be applied to computers. He explains the particular areas where these little tykes can help out computers:

"Young children are capable of solving problems that still pose a challenge for computers, such as learning languages and figuring out causal relationships. We are hoping to make computers smarter by making them a little more like children. Your computer could be able to discover causal relationships, ranging from simple cases such as recognizing that you work more slowly when you haven't had coffee, to complex ones such as identifying which genes cause greater susceptibility to diseases.

In a sense, this research is meant to give computers access to all the different ways humans approach solving problems throughout their lives. While most artificial intelligence research has focused on approximating the goal-oriented, logical reasoning typical of adults, not much attention has been paid to the far more open-minded approach taken by children, which often involves heavy use of imagination and pretend play to help make sense of the world.

Considering the remarkable feats of learning demonstrated by your average toddler, there's definitely a lot to be said for incorporating their cognitive style into our artificial intelligence research. Now it's just a question of seeing how far this new line of inquiry can take us. Personally, I'm hoping a byproduct of all this is more adorable, toddler-like computers...which may be a big reason why no scientists are looking to base their AI models on my brain.


Check out UC Berkeley for more. Image by 26kot, via Shutterstock.

Share This Story

Get our newsletter


I'm still not sure why we WANT computers to be able to learn. I want to be able to program a mower-zoomba to mow my yard (I hate grass) and that's it, thank you. Maybe I've had too many Terminator and Mass Effect Geth nightmares, but I don't want a toaster or Toyota asking me if it has a soul. Keep it simple, keep it stupid, and keep it away from learning, please.