Eric Schulz’ research combines cognitive science with machine learning to understand the fundamental principles of intelligence. To this end, he combines complex and interactive experiments with heavy computational modeling to gain insights into how people learn from little data, curiously explore their environments, and find and adapt simple solutions to complex problems. Since these abilities emerge rapidly over the early years of children’s development, he wants to study children’s behavior to help building a science of learning that can help us build more powerful algorithms, detect when children’s development goes astray, and design tools that teach children how to explore and generalize appropriately.
My plans for the fellowship period
I want to make progress on the following three research branches: 1. Purposeful Exploration, 2. Adaptive Curiosity, and 3. Intelligent Generalization. In the first branch, I want to build a computational model that can reliably track how children and adults generalize from past observations and explore previously unknown options and apply this model in clinical populations as well as in young children. The goal of the second branch is to build a computational model for classifying the exploration type of a child. This can be measured by letting children play problem solving games such as “Mastermind” and then estimating––based on their behavior––parameters that tell us how they generate and test hypotheses. Finally, in the last branch, I want to study how children manage to generalize from experience. For this, I want to use methods from program induction to see how children and adults learn programs from observations. Knowing about the computational primitives that drive people’s generalizations and how they emerge will not only improve our current AI systems, but we can also use this knowledge to teach children how to program and solve problems more efficiently.
How will my work change children’s and youth’s lives?
My ultimate research goal is to create a true “science of learning”, i.e. a computational understanding of how children learn and explore. This understanding then can be used to improve children’s and youth’s lives in three ways. First, using the computational models I will develop, we will be able to detect the mechanisms that make some children explore less and provide them with the help they require, for example by custom-tailoring learning games to their computational phenotype. Second, we will use computational models to track the development of psychiatric disorders such as depression and monitor changes brought about by cognitive behavioral therapy better than by using questionnaires alone. Finally, we will use games and computational modeling to teach children key skills and capacities for a successful life in modern society. These include both specific skills such as programming and designing experiments, but also general capacities such as problem solving and asking intelligent questions.