Jag fick idag glädjande besked av RJ om att projektansökan Moderna tider 1936 beviljats. Tillsammans med mina kollegor Maria Eriksson, Fredrik Norén och Emil Stjernholm kommer vi att undersöka vad samtida algoritmer för mönsterigenkänning ser och hör när de appliceras på mediehistoriskt källmaterial. All historieskrivning kräver förstås tolkningsarbete – men vilka algoritmiska uttolkningar av det förflutna åstadkommer mjukvara? En webbsida om projektet kommer att lanseras vad det lider – men anslaget till vår engelska ansökan ger en vink om vad vi tänker oss:
Media historians usually argue that the past is only available to us through media—be they antique graffiti, disintegrating newspapers, sepia-toned photographs, or last year’s YouTube clip. All humanistic infrastructures (such as libraries) can hence be seen as media archives, where media-specific conditions regulate what is discursively stored. Today, however, cultural heritage institutions are not simply storage facilities but dynamic repositories of digitized content that can be explored with computational methods in new and fascinating ways.
In 1991, Manuel de Landa forecasted a coming age of robots dedicated to understanding their historical origins; he even envisioned “specialized robot historians” committed to trace their genesis, writing “a different kind of history” than humans. Similarly, Hannes Alfvén’s 1966 Swedish science fiction novel, translated as The tale of the big computer conjured up a future where computers not only ruled the world but also had the power to write its past. All likely a computer (“en data”) narrated Alfvén’s story: “When a historian has reached his own time, he ought perhaps to lay down his pen … But how do computers view the problem of man?”
Today, we find ourselves in a situation where machines can be assigned the task of seeing and modeling the human past. What once had a sci-fi character is nowadays a scholarly reality. The project proposal Modern Times 1936— acronym MODERN-36—departs from the fact that the past is not only mediated but increasingly numerically stored and hence prone to computational analysis in ways once envisioned by Alfvén and de Landa. Yet—and this is the overarching research question—what exactly is it that software sees, hears, and perceives when technologies for pattern recognition are applied to sonic and visual media historical sources? MODERN-36 will examine machinic ways of interpreting expressions of modernity in media archival materials from 1936, intentionally zooming in on a comparatively non-spectacular year in Swedish history. Coincidentally, this was also the year when Charlie Chaplin’s film Modern Times premiered, with the little tramp struggling in an increasingly industrialized world—a film that has an iconic status as a critical comment on modernity.
MODERN-36 will focus on one year in order to study quotidian signs of the modern condition. 1936 was a fairly peaceful and politically stable year in Sweden where modernity is often said to have arrived late. At the same time, this was an era of dramatic transformation: several Swedish urban renewal projects were underway in the mid 1930s; technological developments reshaped manufacturing and consumption practices; and a series of new media (such as radio, sound film, and illustrated press) were becoming increasingly popular as sources of both entertainment and education, calling forth new forms of collective experiences and cultural encounters.
MODERN-36 will explore how artificial intelligence (AI) and machine learning (ML) methods can foster new and stimulating knowledge about the history of Swedish modernity—while at the same time developing, and critically scrutinizing, methodological toolboxes for the study of the past. The research is aligned with the recent turn towards large-scale computational analysis of sound and visual content in historical research, and focuses on sonic, photographic, and audiovisual collections from 1936: some 15,000 digitized photographs from DigitaltMuseum, all preserved radio programs from Swedish Radio (Radiotjänst), and all newsreels and short films produced by Svensk Filmindustri. Approaching these collections as datasets, MODERN-36 will develop, and critically explore state-of-the-art digital humanities methods such as speech recognition, object detection and probabilistic topic modeling in three ways: (1.) to examine how software can assist historians in discerning new historical knowledge, (2.) to construct midsize and curated datasets to increase the research capacity of media historical sources and ways of studying these, and (3.) to interrogate algorithmic detection by evaluating what machines can—and cannot—notice in the selected data.