Beyond citizen science: Hybrid intelligences

The impressive advances within artificial intelligence (AI) and machine learning (ML) are largely due to two distinct novelties: increasing computational power and the increasing availability of massive labeled datasets. Despite their success in specialized applications, they also clearly demonstrate that future AI systems will continue to rely on human insight and intuition, or even interoperate with human intelligence. Humans sometimes exhibit the ability to solve high dimensional, complex problems relying on the ability to extrapolate from sparse data, by applying domain-specific heuristics in the form of ‘intuition’. Recognizing this, we explore the potential of citizen science to better understand human cognitive processes and design human-AI systems to tackle complex challenges.

Our unique approach is to exploit three novel uses of the citizen science: We will setup a novel infrastructure allowing for simple and flexible initialization of online, large-scale social science experiments (social science supercollider). We will create a suite of simple online games, which can be used as an orthogonal basis from which individual player cognitive characteristics can be extracted. We will utilize the massive amounts of human player data generated in natural science research games to train machine learning algorithms to tackle various problems such as search and optimization.