Information – Data – Knowledge, or: How do Human and Artificial Intelligence Differ? (in german)
Chair: Dennis Möbus (Hagen)
10:00–10:40 am
Almut Leh (Hagen): “The Answer is 42“ – When Algorithms Take Over the Digital Memory. Experiences with Artificial Intelligence in the Archive „Deutsches Gedächtnis“
Almut Leh will present the digitization work of the first German oral history records. The Deutsches Gedächtnis archive contains about 3000 audio and video interviews. With “Life and Social Culture in the Ruhr Area” (LUSIR), the pioneering work of German oral history was created from 1980 onwards with about 350 contemporary witness interviews covering a period from the German Empire to the 1980s. For some years now, the Deutsches Gedächtnis archive has been working on a digital archive with partners from e-learning and IT.
The online platform – developed by the Center for Digital Systems at Freie Universität Berlin (CeDiS) – will be presented, as well as the state-of-the-art speech recognition system from Fraunhofer IAIS, which is used to automatically transcribe interview tapes.
0:40–11:20 am
Felix Engel (Hagen): Digital Methods for Analyzing Historical Narrative Interviews
Prospective developments of the online archive aim in particular at indexing technologies to break down the content of the unstructured interviews and make them more accessible for research. A central method is Named Entity Recognition to automatically filter out persons, places, and events from the transcripts. AI methods – synonymous here with machine learning – are used for this purpose.
Felix Engel is planning to set up corresponding algorithms on the collections of the Deutsches Gedächtnis archive and will report on the possibilities and limits of Named Entity Recognition for these data holdings. Special attention will be given to the discussion of machine “learning”: What does learning mean, what is the meaning of computational learning, and where does it differ from human learning?
11:20 am –12:00 pm
Tobias Hodel (Bern): Understanding the Machine? Approaches to the Black Box in the Humanities
Based on the results presented and the questions raised on this panel, Tobias Hodel will present epistemological and methodological considerations for discussion from the perspective of a historian and digital humanist. While errors and inconsistencies in the results may point to weaknesses in the methods used, an extended critique of sources and methods should focus on the algorithms in order to gain insights into the apparent black box.
Using examples of Automated Handwriting Recognition, Named Entity Recognition, and Topic Modeling applications, the potential but also the problems of deep learning can be demonstrated. This leads us back to epistemological questions of cultural studies, concluding the panel by elaborating possibilities for understanding and making sense of the technology.