Q&A with Brenden Lake
NYU Center for Data Science’s Brenden Lake talks cognitive science, data, & intelligence
Brenden Lake been at the NYU Center for Data Science (CDS) as a Moore-Sloan Data Science Fellow since 2014, and is now an Assistant Professor of Psychology and Data Science. He researches computational cognitive science, structured probabilistic models, one-shot learning, and human-inspired machine learning.
1. What drew you to cognitive science?
I first learned about cognitive science in high school, and I have been hooked ever since.
Cognitive science asks a lot of big questions: Is the mind a computer? If so, what kind of computer is it? What is intelligence? Can we build machines that are intelligent in the same way that we are?
Cognitive science is inherently interdisciplinary, sitting at the interface of data science, AI, psychology, neuroscience, linguistics, anthropology, and philosophy.
To truly understand intelligence, and to build machines that learn and think like people, we will need contributions from each of these disciplines. The more you learn about the mind, the more fascinating (and sometimes mysterious) it becomes. It’s a very exciting area to work in!
2. What are some of the research projects that you have been working on while you have been at CDS?
I study computational problems that are easier for people than they are for machines. Although there has been exciting recent progress in artificial intelligence, natural intelligence is still by far the best example of intelligence.
We can accelerate progress by studying problems that people excel at, with the aim of reverse engineering the mind’s solutions.
With this guiding principle, my work has touched on questions such as: How do people learn a new concept from just one or a few examples? How do people act creatively when designing new concepts? How do people learn by asking questions?
Typically, I study these questions through a combination of behavioral experiments and computational modeling. My most successful projects have revealed key cognitive ingredients that people use to solve these problems, but are missing from contemporary machine learning and data science. By including these ingredients, we can build smarter and more human-like learning algorithms.
3. It looks like most of your research involves helping machines learn in a more human-like way. But, although human learning processes are effective, they are also flawed. Why, then, do you think it is important for machines to learn like we do?
It’s true that human learning and decision-making aren’t perfect (nobody is!).
But compared to the best machines, even a four year old is a much more impressive learner in many ways. Children are continually learning new concepts, asking questions, making explanations, and running experiments — often based on significantly less data than what we provide to machine learning algorithms.
We still have a lot to learn about intelligence from studying people, especially children, who are the best learners on the planet.
For more, see Brenden Lake’s “Building Machines That Learn and Think Like People.”
Interview by Cherrie Kwok