Brains, Bricks, Ducks and AI
Researchers from Digital Creativity Labs and the Department of Computer Science at the University of York together with the Interacting Minds Centre at Aarhus University, have presented a novel method for assessing and comparing human and computational creativity.
The study was presented at the online Eleventh International Conference on Computational Creativity, ICCC’20, organised by the Association for Computational Creativity.
How to evaluate creativity is a major and ongoing concern in research, the creative industries, education and many other areas.
Most standard methods for measuring creativity are either scalable but provide low quality evaluations, or require a large amount of human expert evaluators, which is quite laborious and prone to human biases.
Researchers examined a LEGO duck construction task to evaluate if a deep learning model could provide a more reliable and efficient way of assessing human creativity – and a way of comparing human and computational measures and capacity.
The project assembled a large quantity of human-made LEGO ducks and used a deep learning model to project photo images of the ducks into a representation space. Analysis of this representation space found groupings based on human recognisable features and hence could be used as a basis for computational novelty metrics.
The research team concluded that under certain conditions, their approach could provide fast and frugal ways of automatically assessing large quantities of creative works.
Researchers cite that exploring the generalisability of their approach to more diverse corpora is an interesting area for future work. Additionally, in order to further validate the findings, a comparison with human expert evaluators should be conducted.
Click here to access the research paper.