When we take a test, we have some idea of how well we’re going to do. When we start a task, we can predict how long it’ll take us to finish it. When we field a question, we can judge whether we need to consult the oracle of Google. We can do all of this because of a skill called metacognition—the ability to reflect upon our own minds, to monitor their degree of certainty, to have knowledge about our knowledge.

Now, a new study from Louise Goupil, Margaux Romand-Monnier, and Sid Kouider at Paris Sciences et Lettres Research University suggests that we have this ability from a very early age. Even twenty-month-old infants have some sense of how well they remember the location of a hidden toy, and they’ll ask for help if they’re uncertain.

That might not be surprising to you but many previous studies have concluded that preschoolers have terrible metacognition. They’ll claim to know what’s hidden inside a box whose contents are a total mystery. Or they’ll say they’re certain about which of two toys is hidden, when they only have a 50:50 chance of guessing the right answer. At that age, they suck up knowledge at a breakneck pace, but they suck at reflecting on that knowledge. They’re metaignorant, apparently.

Not so, says Goupil. She thinks that these early results said more about the experiments than the infants. Perhaps the tasks were just too hard and the children “couldn’t report if they were sure or unsure because they were just confused all the time,” she says. “And maybe they’re bad at talking about their own mental states, but can reflect on their own competencies and knowledge if you ask them to do so non-verbally.”

So she designed a simple wordless experiment. Her team invited 80 Parisian parents to bring their 20-month-old infants to a lab, where Romand-Monnier played with them for a few minutes. She then hid a toy in one of two boxes, either in full view of the infant or behind an obscuring curtain. After a pause of 3 to 12 seconds, she asked the baby to point to the box that the toy was in. Half the babies did so unaided. The other half had the option of silently asking for help from their parents, by turning to them and making prolonged eye contact.

The team found that the infants asked for help more often when the task was harder—when the memorization gap was longer (12 seconds versus 3 seconds), or when the toy was being hidden behind a curtain. And because of that, they got better results: The two groups of infants made the same number of right guesses, but those who could ask for help made far fewer wrong guesses. By correctly gauging their own uncertainty, they avoided errors by asking for help, and so improved their performance.

“They didn’t just constantly ask for help, but only when the task was difficult,” says Uta Frith from University College London, who was not involved in the study. “You can infer from this that they were ‘aware’ that: a) they couldn’t do the task; b) that someone else could do it; c) that that person would be able to tell them how to do it. So it’s a whole lot of useful information that these babies compute.”

Scientists have used similar tests to investigate if animals like monkeys, rats, and honeybees also have metacognition. But the results of these studies are controversial and fiercely debated. The animals in question are often extensively trained to perform the tasks in the experiments, so skeptics suggest that they have simply learned to, say, seek help on difficult trials.

That can’t be the case for Goupil’s infants. “It’s unlikely they could learn precise contingencies,” she says, since each of them only experienced two training trials and ten actual ones. They also didn’t get any better with time, which suggests that they weren’t simply learning how to do better in the study. Instead, the team writes, they can “communicate metacognitive information to others,”  which suggests that “they consciously experience their own uncertainty.”

But here’s the thing, says Frith: The infants only showed metacognition when they were told they could ask for help. “I would not have predicted that a clue would be necessary,” she says. Perhaps that’s how the skill develops. At 20 months, infants have it but it’s latent, and they need a nudge to use it. Later, they wield metacognition more readily. And after four years, as per earlier studies, they understand what they’re doing enough to talk about it.

“Our idea is that you have two components of metacognition,” Goupil explains. There’s an implicit component that allows organisms to reflect upon their own mental state, and is also found in other animals. Then there’s an explicit component, which allows us to communicate about our mental states, and that develops over childhood. “The explicit aspects emerge slowly and thoughtfully with verbal instruction, but the implicit aspects are built-in,” says Goupil.

Her team has already tested even younger infants to work out how early the skill develops. And she suspects that, as with walking and talking, babies climb the metacognitive ladder at different paces. After all, in their latest study, they found that fourteen infants never asked for help, even when they could. They might be showing a kiddie version of the Dunning-Kruger effect, or they might just have a weaker bond with their parents.

“That’s a really interesting question for future research,” says Goupil. Do children start reflecting on their own knowledge at different points? Does that affect their ability to learn, or their performance at school? Whatever the answers, it’s clear that metacognition, for such an everyday skill, is one that we still have much to learn about. Maybe we could all do with more metametacognition.