Abstract:
The recent advancements in language modeling and conversational AI have been accompanied by the promise of a dramatic impact on the adoption of language technologies. Small and medium businesses are not immune to this hype, but often (1) lack the in-house expertise to develop user-centered design and (2) do not have enough (representative) data for training their machine learning modules, nor resources to collect it.
The NLP community has cared about data “before it was cool”. However, in the broader and more variegated community of ML/AI practitioners, the most relevant aspects of data are size of training data and the popularity of benchmark datasets (“the Kaggle game", Manning 2015). Users are also not typically part of the picture, and even less so are crowdworkers, whose humanity is hidden inside a metaphorical “mechanical Turk”. A scarce attention to data and users thus also has ethical consequences on how data is obtained and what data is used for.
I will present some recent work in conversational AI which is largely motivated by the need for a more user- and data-centered perspective. I argue that this perspective is a missing link when transferring technologies outside academia and into industrial use cases.
Location: Attend in person at J577 or via Zoom, https://gu-se.zoom.us/j/66299274809?pwd=Yjc2ejc2VVhraXVJMmhWeWtOQ2NuUT09
Time: 13:15-15