AI’s remarkable computational capacities are well known, but less recognised is the human labour required to train AI systems. Crowdsourcing platforms and start-ups employ thousands of workers to label the datasets AI systems use to recognise text, images, video, etc. As accounts of algorithmic “bias” have shown, a more equitable AI depends on the recognition and deeper understanding of this labour. This project aimed to: build a network of researchers in the UK and India to study the hidden labour of data labellers; identify wider challenges in the production of training data, recognising and rewarding labellers’ crucial contribution to the sector; engage with industry partners and policy makers in India to establish a framework for fair and equitable practices.