Center for Uncertainty Studies Blog

On May 12, 2025, the Center for Uncertainty Studies (CeUS) at Bielefeld University held its fifth Uncertainty Research Afternoon, continuing its tradition of interdisciplinary scholarly exchange. The event, available both in-person and online, brought together researchers to examine uncertainty across different academic fields. The afternoon featured three presentations that explored uncertainty through educational science, mathematics, and historical analysis.

Uncertainty in Educational Contexts: Visibility and Social Practices

Saskia Bender, professor for Educational Science at Bielefeld University, began with her presentation, "The (In)Visibility of (Un)Certainty in Educational Contexts." Drawing from research on inclusive classrooms, Bender argued that societies often work to keep certainty invisible, particularly within democratic structures.

Her research showed that despite goals for inclusive education, practices that reinforce social status continue through what she called "arenas of particularization." Bender observed that cooperative practices in inclusive classrooms often don't achieve their intended outcomes, with students needing help sometimes rejecting assistance.

Bender presented the concept of "covering as a social practice," drawing from Laclau and Mouffe's theory of articulation. She proposed that democratic societies use covering to manage conflicts, hiding structural stability. This creates tension as societies promote inclusion while being structurally unable to fully achieve it.

In the discussion, participants questioned what causes exclusion when inclusion should happen. Bender noted that despite changing expectations, teaching strategies remain largely unchanged. When asked about signs that certainties are becoming more visible, she pointed to the increased transparency of covering practices, though teaching methods have evolved little. The discussion also explored how uncertainty appears differently between peer relationships and teacher-student dynamics.

The Paradox of Uncertainty in Machine Learning

Benjamin Gess, professor for mathematics at TU Berlin, shifted the focus to computational uncertainty with his presentation, "Taming Uncertainty and Profiting from Randomness in Machine Learning." Gess took on the challenge of explaining complex mathematical concepts without using formulas, making machine learning accessible (and entertaining) to a diverse audience.

Gess explained supervised learning by describing how neural networks process and categorize data. He outlined how machines learn through "gradient descent," a process that surprisingly works despite theoretical predictions suggesting it shouldn't. The presentation highlighted the counterintuitive phenomenon of "overparameterization" – the finding that increasing network size (to billions of parameters as in models like GPT-3) can improve performance rather than causing overfitting.

Particularly interesting was Gess's examination of randomness in machine learning. He explained how randomness enters these systems through initially random weights, randomly deactivated neurons, and stochastic gradient descent. Rather than hindering performance, this randomness can actually improve it by helping algorithms find "flatter" minima and cross tipping points in training loss landscapes.

The discussion revealed knowledge gaps in the field. When asked about the importance of hidden layers for model performance, Gess acknowledged that "some architectures just work better" without complete theoretical explanations. Most notably, when questioned about optimal forms of stochastic gradient descent, Gess candidly admitted that such questions "cannot really be answered mathematically," emphasizing that the field largely operates through empirical knowledge and experience rather than complete theoretical understanding. His acknowledgment that "it's interesting that this is not explainable and maybe not understandable at all" highlights a profound insight about uncertainty in a field designed to manage it.

Historical Dimensions of Uncertainty: Rumors in Premodern Communication

The colloquium concluded with Jan Siegemund's (postdoc, Bielefeld University) historical analysis, "Rumors in Flux: Handling Uncertain News in Changing Communication Systems (ca. 1400-1800)." Siegemund's ongoing project examines how premodern merchants dealt with uncertain information during three major transformations: the media revolution, the communications revolution, and European expansion.

Siegemund redefined rumors as "communication of marked uncertainty," using both quantitative content analysis and close reading to identify markers of uncertainty in historical texts. His findings showed that wars, plagues, information about deaths and whereabouts of people as well as everyday life generated substantial rumors. Of particular interest was the relationship between rumors and emerging markets, where the need for current information increased alongside growing economic interdependence through e.g. credit systems.

Siegemund's approach differs from conventional rumor research that tends to focus on effects or views rumors as "improvised news" during crises. Instead, he analyzes rumors through linguistic markers of evidentiality (information source) and epistemicity (credibility assessment).

In the discussion, participants explored the historical evolution of rumor communication, with Siegemund noting that first attempts at textual analysis show more uncertainty markers in 18th-century letters than in 16th-century documents. When asked about different types of rumors across contexts like warfare and trade, Siegemund acknowledged the dominance of military contexts in existing rumor research while advocating for more attention to everyday rumor communications.

The discussion also addressed the social function of rumors, drawing on Norbert Elias's theory to consider rumors as mechanisms for controlling social inclusion and exclusion. Participants identified historical counterparts to today's "water cooler effect," with wells and wash houses serving as potential premodern spaces for informal information sharing.

Connecting Disciplines Through Uncertainty

The fifth Uncertainty Research Afternoon demonstrated how uncertainty functions as a connecting concept across disciplines. From educational practices that try to hide uncertainty, to machine learning systems that benefit from randomness, to historical communication patterns that mark uncertain information, the presentations revealed uncertainty not just as a problem to solve but as a fundamental part of knowledge itself. As moderator Silke Schwandt noted during the discussion, democratic systems allow uncertainty to a certain extent – suggesting that uncertainty might even serve as "a normatively positive thing in democracies." This observation opens new avenues for research into the relationship between political structures and knowledge frameworks.

The Center for Uncertainty Studies continues to facilitate cross-disciplinary dialogue, providing a space for researchers to develop integrated approaches to uncertainty as both theoretical concept and practical reality. Stay tuned for more here on the CeUS Blog.