HCII 2022 | 26 June-1 July 2022 | VirtualPanel: "AI-in-the-loop" - Reconfiguring HCI for AI development | HCI International 2022

24. Juni 2022

Foto: HCII 2021

D-WISE invites to their panel and presentation at the upcoming 24th International Conference on Human-Computer Interaction virtually taking place 26 June - 1 July 2022

"AI-in-the-loop" - Reconfiguring HCI for AI development

An important paradigm in the development of interactive AI systems is “humans-in-the-loop” as trainers in machine learning (see https://humansintheloop.org/) or as evaluators of results or predictions in unsupervised machine learning, primarily optimizing the ML model. Even though dominant, this paradigm however does not adequately represent all approaches made in AI. What is not yet prominent in the debate are alternative HCI paradigms in AI development, e.g. the option to put AI in the loop for primarily supporting human analyses, which might still include mechanisms to enhancing AI performance; this we are currently exploring in the D-WISE project (https://www.dwise.uni-hamburg.de/en.html). In our project, qualitative discourse analyses, in the tradition of the sociology of knowledge, get support by automated analyses of multimodal materials for “filtering” (e.g. theoretical sampling, annotating, development of categories, simultaneous data expansion) relevant materials: AI is in the loop of human analysis and supports human decision making. At the same time, in this mode of human-computer interaction, we are not only motivated to improve human analyses; we aim to constantly improve the representations in the AI systems in the loop of our system, creating a win-win situation for both human understanding and training of the AI system.

“Humans-in-the-loop” and “AI-in-the-loop” are paradigms of human and computer interaction with rather contradicting ideas of how AI and human analyses relate to each other. Thus far, AI-in-the-loop is mostly referred to as an idea but rarely developed as a paradigm in AI research. This panel is interested in bringing together approaches that set AI-in-the-loop of human analytics, rather than operating the other way round. We wish to learn about different ways of putting AI-in-the-loop for human analyses and the different modes of how human-computer interaction is arranged.

Convenors of the panel
Prof. Dr. Gertraud Koch, Institute of Anthropological Studies in Culture and History, University of Hamburg
Prof. Dr. Chris Biemann, Language Technology Group, University of Hamburg
with collaboration of: Dr. Teresa Stumpf, Alejandra Tijerina García, Isabel Eiser, Tim Fischer, Florian Schneider