People often wonder if computers make children smarter.
Scientists at the University
of California, Berkeley (UC
Berkeley) are asking the reverse question: Can children make computers smarter?
And the answer appears to be ‘yes.’
UC Berkeley researchers are tapping the cognitive smarts of
babies, toddlers, and preschoolers to program computers to think more like
humans.
If replicated in machines, the computational models based on
baby brainpower could give a major boost to artificial intelligence, which
historically has had difficulty handling nuances and uncertainty, researchers
said.
“Children are the greatest learning machines in the
universe. Imagine if computers could learn as much and as quickly as they do,”
said Alison Gopnik a developmental psychologist at UC Berkeley and author of “The Scientist in the Crib” and “The Philosophical Baby.”
In a wide range of experiments involving lollipops, flashing
and spinning toys, and music makers, among other props, UC Berkeley researchers
are finding that children—at younger and younger ages—are testing hypotheses,
detecting statistical patterns, and drawing conclusions while constantly
adapting to changes.
“Young children are capable of solving problems that still
pose a challenge for computers, such as learning languages and figuring out
causal relationships,” said Tom Griffiths, director of UC Berkeley’s
Computational Cognitive Science Laboratory. “We are hoping to make computers
smarter by making them a little more like children.”
For example, researchers said, computers programmed with
kids’ cognitive smarts could interact more intelligently and responsively with
humans in applications such as computer tutoring programs and phone-answering
robots.
And that’s not all.
“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,” said Griffiths. He is applying
a statistical method known as Bayesian probability theory to translate the
calculations that children make during learning tasks into computational
models.
This spring, to consolidate their growing body of work on
infant, toddler, and preschooler cognition, Gopnik, Griffiths,
and other UC Berkeley psychologists, computer scientists, and philosophers will
launch a multidisciplinary center at the campus’s Institute of Human
Development to pursue this line of research.
Exploration key to developing young brains
A growing body of child cognition research at UC Berkeley suggests
that parents and educators put aside the flash cards, electronic learning games,
and rote-memory tasks and set kids free to discover and investigate.
“Spontaneous and ‘pretend play’ is just as important as
reading and writing drills,” Gopnik said.
Of all the primates, Gopnik said, humans have the longest
childhoods, and this extended period of nurturing, learning, and exploration is
key to human survival. The healthy newborn brain contains a lifetime’s supply
of some 100 billion neurons which, as the baby matures, grow a vast network of
synapses or neural connections—about 15,000 by the age of 2 or 3—that enable
children to learn languages, become socialized and figure out how to survive
and thrive in their environment.
Adults, meanwhile, stop using their powers of imagination
and hypothetical reasoning as they focus on what is most relevant to their
goals, Gopnik said. The combination of goal-minded adults and open-minded
children is ideal for teaching computers new tricks.
“We need both blue-sky speculation and hard-nosed planning,”
Gopnik said. Researchers aim to achieve this symbiosis by tracking and making
computational models of the cognitive steps that children take to solve
problems in the following and other experiments.
Calculating the lollipop odds
In UC Berkeley psychologist Fei Xu’s Infant Cognition and Language Laboratory,
pre-verbal babies are tested to see if they can figure out the odds of getting
the color of lollipop they want based on the proportions of black and pink
lollipops they can see in two separate jars. One jar holds more pink lollipops
than black ones, and the other holds more black than pink.
After the baby sees the ratio of pink to black lollipops in
each jar, a lollipop from each jar is covered, so the color is hidden, then
removed and placed in a covered canister next to the jar. The baby is invited
to take a lollipop and, in most cases, crawls towards the canister closest to
the jar that held more pink lollipops.
“We think babies are making calculations in their heads
about which side to crawl to, to get the lollipop that they want,” Xu said.
The importance of pretend play
Gopnik is studying the “golden age of pretending,” which typically
happens between ages 2 and 5, when children create and inhabit alternate
universes. In one of her experiments, preschoolers sing “Happy Birthday”
whenever a toy monkey appears and a music player is switched on. When the music
player is suddenly removed, preschoolers swiftly adapt to the change by using a
wooden block to replace the music player so the fun game can continue.
Earlier experiments by Gopnik—including one in which she
makes facial expressions while tasting different kinds of foods to see if
toddlers can pick up on her preferences—challenge common assumptions that young
children are self-centered and lack empathy, said Gopnik, and indicate that, at
an early age, they can place themselves in other people’s shoes.
Preschoolers take new evidence into account
UC Berkeley psychologists Tania Lombrozo and Elizabeth Bonawitz are
finding that preschoolers don’t necessarily go with the simplest explanation,
especially when presented with new evidence. In an experiment conducted at
Berkeley and the Massachusetts Institute of Technology, preschoolers were shown
a toy that lit up and spun around. They were told that a red block made the toy
light up, a green one made it spin and a blue one could do both.
It would have been easiest to assume the blue block was
activating the toy when it simultaneously spun and lit up. But when the
preschoolers saw there were very few blue blocks compared to red and green
ones, many of them calculated the odds and decided that a combination of red
and green blocks was causing the toy to spin and light up at the same time,
which is an appropriate answer.
“In other words, children went with simplicity when there
wasn’t strong evidence for an alternative, but as evidence accumulated, they
followed its lead,” Lombrozo said. Like the children in the study, computers
would also benefit from looking at new possibilities for cause and effect based
on changing odds.
Overall, the UC Berkeley researchers say they will apply
what they have learned from the exploratory and “probabilistic” reasoning
demonstrated by the youngsters in these and other experiments to make computers
smarter, more adaptable—and more human.