Agent-based dynamic support for learning from collaborative brainstorming in scientific inquiry

Abstract

This paper seeks to contribute new insight to the process of learning during idea generation (i.e., brainstorming) by proposing and evaluating two alternative operationalizations for learning, which we refer to as connection-based learning and multi-perspective learning, during a carefully designed idea-generation task in the earth-sciences domain. Specifically, this paper presents two controlled experiments. In the first study we manipulate two independent factors, first whether students work individually or in pairs, and second whether students work with the VIBRANT agent or not. The second study includes one additional hybrid agent condition motivated by results from the first study as well as other enhancements to the VIBRANT agent’s discussion-analysis technology. Our finding is that while brainstorming in pairs leads to short-term process losses in terms of idea-generation productivity, with a corresponding reduction in connection-based learning, it produces a gain in multi-perspective learning. Furthermore, automatically generated feedback from VIBRANT improves connection-based learning. In the second study, support from an enhanced version of VIBRANT showed evidence of mitigating the process losses that were associated with reduced learning in the pairs condition of the first study.