Childhood Is for Exploring Widely Before Doing Anything Useful
A Berkeley psychologist argues evolution gave humans a long, protected early life to solve the same tradeoff AI designers face.
The Brainwaves Video Anthology · Alison Gopnik - The Amazing Minds of Very Young ChildrenAlison Gopnik, a psychologist at Berkeley who studies how very young children learn, had been puzzling over an expensive human habit. Chimpanzees are mature by age 7, producing as much food as they consume, while humans do not reach that break-even point until at least age 15. Why would evolution leave us helpless and unproductive for so long?
Her answer grew out of a collaboration with computer scientists trying to build a machine that learns as well as a toddler does. In artificial intelligence, she said, designers face a tradeoff between searching for new information and putting what is known to work. A system has to poke around, run tests, and learn how things operate, and it also has to use what it has learned to make swift, efficient, and effective decisions.
"You could think about childhood as being a kind of evolutionary way of implementing that strategy," Gopnik said.
The strategy she meant is familiar in computing: start by exploring widely, and only later narrow the range and exploit what you found. Childhood, in her argument, does exactly that. The long, sheltered early years let youngsters undertake wide-ranging exploration and learning when they do not have to make decisions or do anything urgent. The payoff comes later, when grown-ups draw on that early sampling to handle the jobs of adult life.
Gopnik suggests we could think of the long childhood as evolution's way of building in a long stretch for sampling the world before settling on answers. Gopnik called the long protected childhood "one of the great secrets of human success."
Questions people ask
What do computer scientists mean by exploration versus exploitation?
In reinforcement learning, an AI agent has to choose between trying new actions to learn about uncertain rewards and repeating the best-known action to earn rewards now. Too much exploration wastes time, too much exploitation can lock in a mediocre choice, so designers often program systems to explore a lot early and exploit more over time.
Why do humans have such a long childhood compared with other animals?
Partly, big brains take time to wire up, and human skills like language, tool use, and social cooperation require years of practice and cultural transmission. Gopnik's point is that this delay is not only a constraint but a feature, because a protected period lets children sample widely before they must perform.
Are children really better learners than adults at anything?
In her lab work, she found that four-year-olds were more likely than college undergraduates to find an unusual or unlikely answer to a causal puzzle. She did not offer a mechanism for the difference.