People and machines: co creating with heritage collections

Machine-learning technologies such as Handwritten Text Recognition (HTR) are beginning to be used to open and transform cultural heritage material, enabling search, discovery and the creation of linked data. Whereas in 2014, Ridge described cultural heritage crowdsourcing projects as asking the public to undertake tasks that cannot be done automatically,1 today, some projects are asking volunteers to undertake tasks which could be completed by a machine with a high level of speed and accuracy. A deeper exploration of the opportunities and challenges presented by increased incorporation of machine-learning technologies into crowdsourcing projects has yet to be undertaken. This workshop was designed to start these discussions. The ‘People and Machines’ interdisciplinary workshop explored the best routes to fuse digital innovations with the dedication and enthusiasm of volunteers through discussion of what makes an effective crowdsourcing task, how to maintain volunteer motivation, and methods of supporting volunteers to produce useful data, both as part of traditional crowdsourcing, and as part of workflows incorporating machine learning. This report uses the term ‘machine learning’ to refer to a range of technical approaches which use statistical models and algorithms to analyse and draw inferences from patterns in data. Machine learning models and algorithms ‘learn’ from the data they are given, and autonomously adapt and improve their accuracy in response. Machine learning, which includes supervised learning, unsupervised learning, and deep learning, is a subfield of Artificial Intelligence