Increasing Transfer of Complex Systems Principles from Computer Simulations with Analogical Comparison

• In Event: 2-084 - Relational Learning in Real-World Contexts: Cognitive, Educational, and Socio-Cultural Factors

Integrative Statement

Learning many concepts in science (e.g., evolution, brain development, climate change) requires not just learning particular facts, but engaging in a new form of causal reasoning (Chi et al., 2012; Jacobson et al., 2010). Typical causal reasoning breaks events into a series of subparts, where A first causes B, which then causes C etc. This reasoning attributes causal power to individual agents that bring about change in a linear, step by step manner. Some scientific concepts can be reasoned about in roughly this way, (e.g., photosynthesis), but many others cannot. These phenomena are often better understood by the causal principles of complex systems, which explain how macro-level phenomena (e.g., columnar organization of visual cortex) emerge from a large number of interacting agents (e.g., neurons), governed by a small set of rules (e.g., excitatory and inhibitory connections). Critically, the patterns in the interacting agents on the micro-level do not in any clearly divisible and linear manner cause the macro-level pattern.
Prior research has shown that students’ use of agent-based model simulations supports understanding how the interactions at the micro-level are distinct from the observable patterns at the macro-level (Samon & Levy, 2017; Lai, Jacobson, & Goldwater, 2018). In the current research, students learned about climate change with agent-based models (see Figure 1). These models allow students to run simulations of some system, (e.g. the climate) and infer the causal relations between key variables (e.g., the relationship between CO2 in the atmosphere and global temperature) because students can change various parameter settings (e.g., CO2 levels) before running the simulation, which displays how the system changes over time (e.g., increasing global temperatures) based on those particular settings.
Ninety-seven 9th grade students in Sydney, Australia participated in a four-lesson unit on the climate as a complex system to teach them about how climate change occurs, and about complex systems more generally. Key principles about complex systems concerned the nature of emergent patterns; how negative feedback cycles produce dynamic equilibrium; and how positive feedback cycles disrupt equilibrium producing non-linear accelerated growth (such as with current global temperatures).
Half of the students learned just with models of the climate, while the other half had additional models to compare with the climate models that highlighted these key principles (e.g., a wolf-sheep predator-prey model displaying negative feedback cycles). Critically, time with the models overall was equivalent, such that it would be plausible to predict disadvantages for the analogical comparison condition because they had more material to consider in the same amount of time. However, these were not the results.
For declarative knowledge of the climate and complex systems, learning across conditions was equivalent. For a far transfer problem wherein the solution required spontaneous application of complex systems principles to a novel domain (a population of simple robots foraging on an alien planet), the analogical comparison condition significantly outperformed the climate-model-only condition, f(1,92) = 10.403, p < .01, η2p = .102. These results are consistent with theories of analogical comparison that emphasizes its critical role in knowledge transfer.

Author

• Micah B Goldwater, The University of Sydney
  Presenting Author