| Image understanding is a general challenge in Artificial Intelligence (AI) because of its complexity. It is considered an AI-complete problem. We focus on the specific, important, and difficult case of microfossil identification, which is currently done manually. Microfossil identification has applications in paleoenvironmental research and oil exploration. We use evolutionary prototyping to engineer a complex system that employs crowdsourcing, mainly human-based computation. Our latest prototype, called the Microfossil Quest, combines human intelligence, including expert and citizen science, with computer intelligence, including unsupervised and supervised learning. A front-end website was developed to accommodate human interaction. It integrates easy-to-use interfaces for search and identification, detailed and interactive digital representations, and information for educational and motivational purposes. Computer intelligence is used in the back-end to synthesize and leverage human intelligence. To ensure a high quantity of high quality identifications are obtained quickly, the dynamic hierarchical identification algorithm was created to cluster specimens, propagate knowledge, and prioritize input. In this fashion, we provide not only a clear and strong approach to the specific problem of microfossil identification but also a significant case study for image understanding in general. |
