The Biology Phenomenon Categorizer: A Human Computation Framework in Support of Biologically Inspired Design

Locating relevant biological analogies is a challenge that lies at the heart of practicing biologically inspired design. Current computer-assisted biologically inspired design tools require human-in-the-loop synthesis of biology knowledge. Either a biology expert must synthesize information into a standard form, or a designer must interpret and assess biological strategies. These approaches limit knowledge breadth and tool usefulness, respectively. The work presented in this paper applies the technique of human computation, a historically successful approach for information retrieval problems where both breadth and accuracy are required, to address a similar problem in biologically inspired design. The broad goals of this work are to distribute the knowledge synthesis step to a large number of nonexpert humans, and to capture that synthesized knowledge in a format that can support analogical reasoning between designed systems and biological systems. To that end, this paper presents a novel human computation game and accompanying information model for collecting computable descriptions of biological strategies, an assessment of the quality of these descriptions gathered from experimental data, and a brief evaluation of the game's entertainment value. Two successive prototypes of the biology phenomenon categorizer (BioP-C); a cooperative, asymmetric, online game; were each deployed in a small engineering graduate class in order to collect assertions about the biological phenomenon of cell division. Through the act of playing, students formed assertions describing key concepts within textual passages. These assertions are assessed for their correctness, and these assessments are used to identify directly measurable correctness indicators. The results show that the number of hints in a game session is negatively correlated with assertion correctness. Further, BioP-C assertions are rated as significantly more correct than randomly generated assertions in both prototype tests, demonstrating the method's potential for gathering accurate information. Tests on these two different BioP-C prototypes produce average assertion correctness assessments of 3.19 and 2.98 on a five-point Likert scale. Filtering assertions on the optimal number of game session hints within each prototype test increases these mean values to 3.64 and 3.36. The median assertion correctness scores are similarly increased from 3.00 and 3.00 in both datasets to 4.08 and 3.50. Players of the game expressed that the fundamental anonymous interactions were enjoyable, but the difficulty of the game can harm the experience. These results indicate that a human computation approach has the potential to solve the problem of low information breadth currently faced by biologically inspired design databases.