Overview and details of the sessions of this conference. Please select a date or room to show only sessions at that day or location. Please select a single session for detailed view (with abstracts and downloads if available).
AI Models and Their Worlds: Investigating Data-Driven, AI/ML Ecosystems Through a Work Practices Lens
C. T. Wolf
IBM Research - Almaden, United States of America
When we invoke the “future of work,” to whose work do we refer? This pa-per considers various everyday work practices through which contemporary artificial intelligence (AI) and machine learning (ML) ecosystems are made possible. The “future of work” is often talked about in relation to the antic-ipated domains settings where AI/ML systems might be implemented and the labor conditions such implementations might re-configure (foreseeing or noting changes in medical/health, legal, or manufacturing work, for ex-ample). This paper turns our attention to the various forms of labor that must be undertaken to conceive of, train/test, deploy, and ongoingly main-tain AI/ML systems in practice. In particular, this paper draws on an ongo-ing ethnographic endeavor in a large, global technology and consulting cor-poration and leverages a work practices lens to examine three themes: curating datasets (everyday work practices of data pre-processing); tend-ing models (everyday work practices of training, deploying, and maintain-ing predictive models); and configuring compute (everyday work practices of back-end infrastructuring, commonly called “DevOps”). This paper con-siders the value of a work practices lens in studying contemporary soci-otechnical labor ecosystems. By locating the work practices through which AI/ML systems emerge, this paper shows that these technologies indeed re-quire considerable human labor, at the same time they are often talked about as drivers of automation and displacers of work. This extends discourses around the “future of work,” giving light to the various standpoints and ex-periences of labor such imaginaries implicate and ongoingly re-configure.
Identifying Historical Travelogues in Large Text Corpora Using Machine Learning
J. Rörden1, D. Gruber2, M. Krickl3, B. Haslhofer1
1AIT Austrian Institute of Technology, Vienna, Austria; 2Austrian Academy of Sciences, Vienna, Austria; 3Austrian National Library, Vienna, Austria
Travelogues represent an important an intensively studied source for scholars in the humanities, as they provide insights into people, cultures and places of the past.
However, existing studies rarely exceed a dozen primary sources, as the human capacities of working with a large quantity of historical sources are naturally limited.
In this paper, we define the notion of travelogue and report on an interdisciplinary method that, using machine learning as well as domain knowledge, can effectively identify German travelogues in the digitized inventory of the Austrian National Library with F1 scores between 0.94 and 1.00.
We applied our method on a corpus of 161,522 German volumes and identified 345 travelogues that could not be identified using traditional search methods, resulting in the most extensive collection of early modern German travelogues ever created. To our knowledge, this is the first time such a method was implemented for the bibliographic indexing of a text corpus on this scale, improving and extending the traditional methods in the humanities. Overall, we consider our method as being an important first step in a broader effort of developing a novel mixed-method approach for the large-scale serial analysis of travelogues.
Bridging DH & Humanistic HCI: Distant Reading as Transdisciplinary Method
J. S. Seberger
University of California, Irvine, United States of America
Bowker’s Age of Potential Memory describes a new era characterized by a culture of knowledge production that fosters and stifles certain forms of statements depending on the logics that subtend them. Through processes of ubiquitous data collection, analysis, and feedback, individuals are increasingly reduced to users; users are re-created as data doubles or data doppelgangers, post hoc, through the aggregation and analysis of their data traces. This discursive transformation of the human that will arise in relation to living alongside and through these doubles or doppelgangers is difficult to understand within the framework of extant disciplinary silos. And yet methods that connect disciplines are emerging. To realize these connections, translational work is required. This paper explores the complementarity of digital humanities (DH) and humanistic human-computer interaction (hHCI) through the lens of distant reading. I focus on distant reading—topic modelling in particular—because of its methodological popularity and relation to discourse. I argue that distant reading comprises a useful connection between these two young domains: a pivot that allows for the inter- or transdisciplinary study of the future human through the analysis of its potential sociotechnical, discursive compositions.