Carney Institute (CI): Let's go back a bit. What led you to your current area of research?

Daphna Buchsbaum (DB): I took a very circuitous route. I actually went to Brown as an undergraduate. I always had an interest in animals so when I came here, I thought I was going to be a veterinarian. I started taking some biology courses and one class on animal behavior in the Department of Ecology and Evolutionary Biology really inspired me. We went out and observed animals on campus and tried to learn to make close observations of their behavior. I loved the course so much that I later TAed it twice. That was something that really started pushing my interest beyond just biology and toward intelligent behavior. Later in my studies, I also became interested in computer science, starting with the intro course series CS 17/18 (CSCI0170 is still taught by the same professor, John “Spike” Hughes). Initially, I tried Computer Science just for fun, but I got really into it and took a bunch of courses for the last two years of my degree. So, there was already this kind of parallel of studying computer behavior and animal behavior. 

From there, I did a master's degree at MIT in the Media Lab in a group that did artificial intelligence for animated characters and robots, that were specifically inspired by animal behavior and cognitive science. The idea was that the behavior of the animals that we were animating was generative; that it was coming from the programming to give them certain kinds of animal-like abilities or inference behaviors, not from somebody animating the whole thing. In many ways, this research felt like it had been created especially for me.

Following my master’s degree, after a few years working in industry I decided that I wanted to go back to academia. I had a big debate in terms of what I wanted to focus upon: Did I want to go to a computer science graduate program, or more of a cognitive science or psychology graduate program? I actually applied to both kinds of programs, and it was a bit of happenstance that I ended up more on the psychology side. I ended up at UC Berkeley and began to see ways of bringing these two aspects together. I had two advisors for my PhD, one of whom was doing computational cognitive science work, and one of whom was doing cognitive development research. I was primarily working with kids, studying children’s thinking and learning, but I still had this interest in animals that was in the back of my mind the entire time.

CI: So, was there a “sea change” moment that brought you back around to working with animals?

DB: While I was doing my Ph.D., I got a dog, and I started training as a K9 search and rescue team with her. That really started to make me think about some of the things I was at that time researching in children with respect to how the dog was conceptualizing tasks that we were doing together.

At the time, I was doing research with kids about how they learn cause and effect relationships, reason socially, and learn causal information from social interactions. To give a specific example, we were investigating how a child watching an adult complete a task decides which part of the action to imitate to complete that same task. It's a very sophisticated and challenging activity but I was curious to learn to what extent animals can approach a task or learn in a similar way, or what other learning strategies they might use instead.

In search and rescue dogs, you see what looks like a lot of flexibility in the dog’s understanding that, at least to a human, appears to go beyond the rote task that you taught them. For example, when a dog is learning to locate an individual, you train them in a very rigid sequence of events. First, they have to go and find the person, and then they find them from farther and farther away, and then you train them to come back to you. Then you train them to show you where the person is. This could all be pure association or reinforcement learning; the dog associating returning to their trainer with getting a reward. But often what appears to be coming out of that — and I'm being careful to say “appears” to the human handler— is that in some cases the dog seems to get the bigger picture of the game. Rather than just learning "Run to the person. Run back to the handler. Run to the person," the game perhaps becomes "Find this person and tell their trainer about them." I was very interested in understanding whether this was pure anthropomorphism on the part of the human handlers, or whether at least some dogs really are capable of learning a more abstract understanding of the task.  

CI: Is canine cognition a growing area of neuroscience? 

DB: While most of us know about Pavlov's famous experiment, canine cognition was not a very popular area of research for a long time, perhaps because dogs were thought to be rather boring from a research perspective. In the past 20 years, however, interest in canine domestication and its effects on dogs’ social cognitive abilities has really exploded. While dogs are not that closely related to us phylogenetically, there’s a hypothesis that argues that through domestication we've been putting selection pressure on them to develop social learning and reasoning abilities that could tell us something interesting. 

Researchers began recognizing that dogs are really good, arguably uniquely good, at following human social communicative cues — at least better at it than a lot of other species, including other non-human primates. So there's various reasons to think they've been especially selected to cooperate with humans and to learn about our social cues or even that they may have some more human-like social cognitive abilities because of domestication. That made people much more interested in how dogs are able to so successfully live in human social environments and work and communicate with us. Then there's also an argument on the flip-side of that, that domestication didn’t make dogs better or smarter across the board — it made them different. For instance, some researchers believe that, because dogs came to rely upon us to solve certain kinds of problems for them, they became better at social reasoning and worse at physical and causal reasoning.

What's also exciting about studying dogs is that wolves and dingoes are still in existence. Dogs were the very first species we domesticated, at least 30 or 40,000 years ago. Then, around 4,000 to 8,000 years ago, early domesticated dogs were brought to Australia where they re-wilded to become the dingo. As a result, dingoes are arguably this midpoint in the domestication process compared to our current dogs. You can do similar kinds of studies on dogs, dingoes and wolves and that might tell you something about their common ancestors; about what behaviors are due to domestication; and what behaviors are shared across the species, and might be due to other selective pressures, like cooperative hunting. 

CI: I know you work with children, the human kind, in the Computational Cognitive Development Lab. What are you investigating?

DB: Some of the things we're studying now are questions related to how children's understanding of categories and concepts change over time.

One of these areas is the concept of color. As adults, for the most part speakers of the same language categorize colors consistently. In English, for example, speakers agree on which shades and hues are blue and which are green. However, not all languages divide up the perceptual space of color in the same way — languages can have different color terms and even different numbers of color terms. A relatively well known example is that Russian has two primary shades of blue. So, as children learn the color terms of their language, they have to figure out which parts of the perceptual space correspond to which color word. However, we actually know relatively little about how children use early color terms and how this changes as they acquire more color words. What do red and green mean to a child who knows only those two color words? Are these colors the same as “adult red and green?” Does the use of color terms differ from child to child early in development? So, in our research, we elicit color judgments from children, asking them to name or sort colors for us to see how this develops over time. Right now, we're just doing this research in English but we're very interested in comparing the languages that use different color words.

We're also really interested in children's learning of animal categories. There's an idea that children understand these relationships from a more perceptual understanding; that animals that look similar and share features are indeed similar. A child might think that sharks and dolphins go together or that butterflies go with birds. Conversely, they may think that penguins are not birds. We understand this from snapshots at different ages but we have much less of an idea of how this differs from individual to individual.

Aside from this recent work on category development, my lab also conducts a lot of research studying how children learn cause and effect relationships, particularly how they learn from observing others’ actions. So, when a child sees someone bringing about an interesting effect–making a toy play music, getting a sticker out of a puzzle box, how do they decide which of those actions to perform themselves when it’s their turn? Do they faithfully imitate all of the actions that they observe, or are they more selective, particularly if some of the actions seem implausible or unnecessary? This is a set of questions that is particularly relevant to cultural transmission and cultural learning, how children quickly come to be experts in all of the tools and artifacts used in their society.  

CI: Your team runs the Brown Dog Lab. The big question many people may be asking themselves is “Can my dog be part of the lab’s studies?”

DB: The answer is yes! The only requirement is that the dog have their rabies vaccination and not be aggressive to humans. If your dog is feeling a bit anxious, we also do some Zoom studies for some of our cohort. 

One exciting thing that we've been doing is head-mounted eye-tracking. So this is a bit more of a time investment, but for dogs who fit the study criteria, we help the owners train their dog to be comfortable wearing training goggles. Once they're comfortable wearing them, we have them do different kinds of tasks or go on a walk while wearing the eyetracking goggles and we can actually see what the dog is visually attending to in their environment. We're collaborating with Thomas Serre and some students in his lab to help us automatically identify the objects in the dog's field of view, so that we don't have to manually code what it is that's available in every frame of the video. We can actually use some computer vision techniques to help us process this data.

If you’re interested in finding out more, or signing up to participate with your dog, you can do that here.

Of course, we also have both in-person and Zoom studies for children, and your children are also more than welcome to participate! You can find out more here.