Organizing humans to teach machines: How service design helps put the intelligence in artificial intelligence (English—live translation available)

For some, artificial intelligence appears magical in its provenance. For others, it appears as a hyper-technical field, driven by highly specialized scientists tinkering away in labs. Both perspectives ignore what really happens behind the curtain as AI teams build models that work in the wild. While the front stage shows an apparently self-teaching machine, the back stage is animated by human actors. Amongst these are the people who manually annotate data, tilling the ground for machine learning. Data annotation involves complex interaction between 1) non-technical labellers (the so-called blue collar workers of the AI era), 2) technical teams that ingest human-labelled data, use it to improve models, and provide return feedback on quality, and 3) project managers who configure and monitor labelling projects. This talk considers how the principles of service design can help us build more effective data annotation processes and, ultimately, better and more impactful AI models.

 
 
 
jason.png

Jason Stanley, Element AI

Jason Stanley is Design Research Lead at Element AI, where he helps the company's product teams identify opportunities for innovation and learn from iterative attempts to address them. He has previously worked as a product researcher and data scientist for several software companies, as an advisor on labour market policy for the Government of Canada, and as a researcher investigating the use and impacts of technology around the world, including in Asia, South America, Europe, and North America. He holds a Ph.D. in Sociology from New York University and graduate degrees from Oxford University.