Artificial Intimacy: An Exploration of the Personal and Intimate in Natural Language Processing Models

Mirabelle Jones, Nastasia Griffioen, Irina Shklovski, Obaida Hanteer

Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review

64 Downloads (Pure)

Abstract

Artificial Intimacy is an AI art installation that explores what natural language processing (NLP) models in our everyday lives would feel like if they were to be personalized to match our own personalities and values. We explored the possibility of fine-tuning NLP models using personal social media data. Our selected data sources - Leslie Foster and Gorjeoux Moon - have offered their own social media data to fine-tune the models. We present a video capturing their conversations with their social media selves. The interactive portion of the installation invites the audience to engage with Foster's and Moon's chatbots and explore interactions with NLP models that are personalized in this way.

Original languageEnglish
Title of host publicationParticipative Computing for Sustainable Futures - Adjunct Proceedings of the 12th Nordic Conference on Human-Computer Interaction, NordiCHI 2022
Number of pages2
PublisherAssociation for Computing Machinery, Inc.
Publication date2022
Article number23
ISBN (Electronic)9781450394482
DOIs
Publication statusPublished - 2022
Event12th Nordic Conference on Human-Computer Interaction: Participative Computing for Sustainable Futures, NordiCHI 2022 - Aarhus, Denmark
Duration: 8 Oct 202212 Oct 2022

Conference

Conference12th Nordic Conference on Human-Computer Interaction: Participative Computing for Sustainable Futures, NordiCHI 2022
Country/TerritoryDenmark
CityAarhus
Period08/10/202212/10/2022
SponsorDepartment of Computer Science, Aarhus University, DIREC, IT-vest, Stibo Fond
SeriesACM International Conference Proceeding Series

Bibliographical note

Publisher Copyright:
© 2022 Owner/Author.

Keywords

  • Artificial Intelligence
  • Model finetuning
  • NLP
  • Personalization
  • Values

Cite this