Development & Learning
Compared to other animals, humans are flexible generalists. We are born “half baked”, with relatively little in the way of initial skills or “core knowledge”. We spend decades growing and maturing and even once mature can continue to be able to learn new things. This session will explore a zoomed out perspective on how the human mind achieves this impressive flexibility. This touches on how the ability to learn and change shifts over our lifespans and the benefits and costs of childlike cognition.
Primary Readings
Everyone should read these and be prepared to discuss:
| Spelke and Kinzler (2006) | Core knowledge. Developmental Science, 1, 10. |
| Gopnik, A., Griffiths, T. L., & Lucas, C. G. (2015). | When younger learners can be better (or at least more open-minded) than older ones. Current Directions in Psychological Science, 24(2), 87-92. |
Secondary Readings
The presenter should read and incorporate these:
| Bramley, N. R., Dayan, P., Griffiths, T. L., & Lagnado, D. A. (2017). | Formalizing Neurath’s ship: Approximate algorithms for online causal learning. Psychological Review, 124(3), 301.This article discusses the use of approximate algorithms for online causal learning, using Neurath’s Ship as a formalization framework. The study proposes an algorithmic-level model of causal structure learning and examines the computational challenges and the need for approximations in causal learning. They use the concept of Neurath’s Ship to describe the incremental growth and evolution of beliefs about causal structure. Overall, the article presents various models and approaches to understanding human causal judgments and active learning. It discusses the use of approximations and the need for a general-purpose algorithm that can explain human success in learning complex causal models. The article also highlights the use of methods from machine learning to construct such an algorithm. |
| Ullman, Tomer D. & Tenenbaum, Joshua B. (2020). | Bayesian Models of Conceptual Development: Learning as Building Models of the World. Annual Review of Developmental Psychology, 2 (1).The use of Bayesian inference in model-building, including hierarchical Bayesian models and probabilistic generative programs, is discussed. The role of core knowledge in child learning and the relationship between intuitive theories and core knowledge are explored. Learning is described as a stochastic search process, with theory learning being a search for programs. The text also discusses the theory theory and the child as scientist theory, as well as the use of Bayesian models and probabilistic programs to capture the structure of intuitive theories. The tensions between different methods of scientific experimentation are explored, and the same cognitive processes are noted to have different outcomes depending on the circumstances. The author suggests that a single program learning mechanism can be used to explain data but that its effectiveness depends on relevant data representations, input analyzers, and primitives available. The text acknowledges the challenges of model-building in science and suggests that a comprehensive account of how humans develop models of the world would need to consider various factors such as individual differences, cultural influences, and computational constraints. The passage concludes by discussing ongoing work towards creating a computational model of cognitive development. |
Questions under discussion
- In what sense is childhood like simulated annealing?
- What is the “Child as scientist” theory of development?
- What might order effects in learning tasks reveal about about learning?
