Today I’m here to talk to you about babies and their understanding of the world. Specifically, their understanding of dangerous situations. But first, in case you haven’t interacted with a baby recently, here are some clips of what we call typical one-year-old behavior. Well! (Laughter) From this, you might think that babies don’t understand anything about dangerous situations, or maybe they don’t understand anything about anything. But today I’m going to tell you five surprising things that I learned while I did this work. The first surprise is that babies are really willing to do dangerous things. So these are videos captured in the lab, and the finding from this work is that one-year-old babies are perfectly willing to walk off the edge of these steep drop-offs without even thinking twice. And in fact, they need months of experience learning to walk before they start to show any signs of fear in these situations. So, that was work completed before I started my PhD. And so I looked at these findings and I thought, “Really? Is it really true that babies are oblivious to danger? Or can we figure out another way to study this question?” The history of developmental psychology tells us that if we study babies by measuring complex behaviors, we might miss out on hidden truths about their minds. So what should we do instead? Well, another way is we can ask them what they think when other people are put into the same situation. From the first months of life, babies have expectations about the world, that objects are solid and don’t float in mid-air, that people have intentions and goals, and their expectations are often clearest when they themselves are looking at the events rather than participating in them. So that's going to be surprise number two. If you test babies at the same age as I just showed you, in the second way, you see a completely different pattern of results. So, here is an example of one of our stimuli. And babies are going to see someone else, this red guy in the middle, face a choice: to jump left or jump right. Intuitively, you can recognize that one of the choices is more dangerous than the other. Falling in the ditch on the left would be a lot worse. So all else being equal, maybe this red guy is going to go right. And the question for babies was, can they make that distinction and do they have that same expectation? So here we’re measuring how long babies look at each of these outcomes. And the simple logic of this measure is that babies, and indeed people of all ages, tend to pay more attention to what’s surprising. In this case, when the red guy does the more dangerous thing. So here’s what we found. I’m plotting looking time on seconds on the vertical axis, and in red is the average looking when babies saw the more dangerous choice in green and the safer choice in pink. So indeed, they look longer when someone chooses a dangerous thing over a safer thing. But in contrast, the very same babies responded completely differently when they saw these very, very similar control videos with the same obstacles, except this time there are no characters in the scene, so there’s no danger at all. And this suggests that they’re not responding to the obstacles themselves, and they’re really responding to the danger imposed by the obstacles. So that’s the second surprise. Babies do understand something about dangerous situations. But remember, based on their own choices, this is shocking! We wouldn’t expect a one-year-old to care about the distinction between a deep versus a shallow drop at all. But measured in this way, babies are making exactly that distinction when someone else faces the same choice. So surprise number three is that this finding generalizes, regardless of whether babies are tested in sterile lab environments like this, an empty room with nothing else but our videos to look at and cluttered home environments. In the middle of this work, Covid hit, and developmental psychologists scrambled to figure out whether data collection was even possible. If you asked me, I would have predicted definitely not. There’s a reason that we conduct studies in the lab. Our movies can be projected to these huge screens. We can minimize distraction. We can get highly precise, high-definition videos of where the babies are looking. But we figured we had to try. So after a long process of figuring out exactly how to do this, we ended up video-calling families at home, and they played the same movies for babies on their laptop screens, and we measured their looking behaviors just like before. So let's just pause and take stock of all the differences between these two setups. In many ways, this was doomed to fail. I certainly thought it was going to fail, but that's not what happened. So on the left are the results I just showed you with babies tested in the lab. And on the right is the same experiment we repeated in a new group of babies tested at home. And as you can see, the main results look remarkably similar. So that was surprise number three. This finding is similar, regardless of whether babies are tested in the lab or at home. So I just want to pause and state how non-trivial and actually really groundbreaking this was for my field. My prediction that we cannot study babies using these measures at home wasn’t true at all! During this process, we learned not only that these studies can work at all, but that they can actually work really well. And at the end of the day, this process was just easier and faster for everyone. So we ended up being able to collect video data faster than ever before, which actually required us to engineer new tools for processing the data. So here I'm showing you a tool we built that can find a baby's face, given a webcam of their study, and detect where they’re looking as good as a human can, that is robust even for these very fuzzy, low-quality webcam videos. So, this whole experience was amazing for our team. And for me, this experience is its own achievement. It took families and scientists working together to figure out how to study baby minds in a changing world. So that was surprise number four. The pandemic, rather than delaying and maybe ending the work in my field, actually accelerated it. I never would have tried online research had Covid not happened, but now that we've tried it, it's vastly expanded the reach of our work. These tools enable us to study babies anywhere in the world with an internet connection. It lets us work with families of varying demographics, and it lets us more easily access clinical populations that can be harder to reach. So the last surprise, surprise number five. All along I’ve been telling you that babies look longer when someone chooses a dangerous action or are willing to do really dangerous things, when I probably should have said: on average. On average, the babies we tested look longer at X. On average, babies are willing to do Y. So to illustrate this point, here are the same graphs I showed you before, but hidden in these graphs and in all the data sets from my field are vast individual differences. So now I've added back all of the raw data that you didn't see before. Now, each pair of points is a baby, and as you can see, there is striking variability between babies that average out to an effect. Now, to be clear, I don't think this undermines the conclusion of the work. The key prediction is that at home and in the lab, babies are sensitive to something about these dangerous situations. But one remaining puzzle from my field is what these individual differences mean. So, one answer is: that’s just measurement noise. We’re studying babies, they can’t follow directions, they have short attention spans, they’re free to basically look as long as they want. So maybe these differences don’t mean anything. It’s just random noise. A second answer is that part of these differences is driven by external factors. So maybe for our online sample, the babies who showed the biggest effects had the biggest screens to look at or maybe the clearest Internet connections. And the third answer, maybe the one that’s the most interesting, is that these differences reflect something that’s true about the actual differences between these babies. So for example, suppose that we put babies from this looking time study in the physical situations I showed you from the beginning of the talk. Remember, there are individual differences in those experiments too. So a question could be: would the babies who show the strongest effects here be the ones who also are more likely to be afraid and avoid doing the dangerous things themselves? And similarly, are the babies who do not show an effect here, the ones that are willing to boldly plunge into the abyss? So to be clear, I don’t know which one of these answers is right, but I think it's important to show this variability in our papers and to study them. So, I think that these challenges are actually tractable in part because of the tools that were delivered by the pandemic. Stay tuned and maybe I’ll have an answer for you next time we see each other. So that’s the last surprise, certainly a surprise for me who started out looking at bar plots of average behaviors and assuming that most babies behave the same. Instead, there’s variability in human development on just about any measure we can make, on just about any topic we can study, from physical understanding to language learning. And making sense of these individual differences, telling apart the signal from the noise, continues to be a challenge and an opportunity for my field. So the next time you look into the eyes of a baby, note that there’s more going on in their minds and brains than you might think. Zooming out to the big picture, let’s think about what these findings tell us about the origins of the mind. So to me, these findings and many other findings from our field tell us that babies have abstract knowledge about the social and the physical world. They can perceive and reason about events they’ve never seen before, like these stimuli. They can represent not only what happened but what could happen next, or maybe what could have happened but didn’t. And early in development, babies are learning to solve these two really, really hard problems. The first problem of connecting other people’s actions to their physical situations and hidden mental states. And also the second hard problem of figuring out how to act in the world, given what they know. So of course, that’s both an interpretation and a hypothesis. We have a lot to learn about these profoundly intelligent minds. And actually, the magic of that process reminds me a lot of a one-year-old. Falling over. Having a long look at the obstacles before us and then taking a big leap. Thank you. (Applause)
