#Riseandgrind: Lessons From a Biased AI

Research output: Contribution to conferencePaperpeer-review

Abstract

#RiseandGrind is a research-based artwork that, through a process of active engagement with the machine-learning tools of what is known as artificial intelligence, sought to make visible the complex relationship between the origins and context of training data and the results that are produced through the training process. The project using textual data extracted from Twitter hashtags that exhibit clear bias to train a recurrent neural network (RNN) to generate text for a Twitter bot, with the process of training and text generation represented in a series of gallery installations. The process demonstrated how original bias is consolidated, amplified, and ultimately codified through this machine learning process. It is suggested that this is not only reproductive of the original bias but also constitutive, in that black-box machine learning models shape the output but not in ways that are readily apparent or understood. This paper discusses the process of creating and exhibiting the work and reflects on its outcomes.
Original languageEnglish
DOIs
Publication statusPublished - 2019
EventRadical Immersions: Digital Research in the Humanities and Arts - London, United Kingdom
Duration: 8 Sep 201910 Sep 2019

Conference

ConferenceRadical Immersions: Digital Research in the Humanities and Arts
Country/TerritoryUnited Kingdom
CityLondon
Period8/09/1910/09/19

Keywords

  • machine learning
  • artificial intelligence
  • training data
  • bias
  • Twitter
  • recurrent neural network
  • text generation
  • gallery installations
  • black-box models

Fingerprint

Dive into the research topics of '#Riseandgrind: Lessons From a Biased AI'. Together they form a unique fingerprint.

Cite this