• News

First ComAI Diversity Lecture: AI, Discrimination, and Stereotypes

On 21 May 2026, the ComAI research group’s Working Group on Diversity & Gender Equity launched its new internal Lecture Series. The inaugural session featured an input talk by Katharina Mosene from the Leibniz Institute for Media Research | Hans Bredow Institute (HBI) and the Humboldt Institute for Internet and Society (HIIG), on AI, discrimination, and stereotypes.

Drawing on an intersectional feminist perspective, Mosene showed how historical training datasets embed hegemonic worldviews, dominated by content from the Global North and predominantly English-language sources. She employed the (also criticized) concept of “Stochastic Parrots” (Timnit Gebru et al.) as an analytical frame, describing LLMs as systems that statistically reproduce linguistic patterns, perpetuating rather than challenging societal stereotypes.

Through concrete examples, Mosene illustrated how image-generating AI systems skew occupational representations: roles with high social status are disproportionately rendered as male and white, while care and domestic work appears female-coded. Translation systems exhibit similar biases. From a feminist standpoint, she argued, the EU AI Act falls short by failing to address structural power questions.

A closing example from journalism proved particularly striking: multiple LLMs were queried about accessing abortion services in Germany, yielding starkly different responses — measured and bounded in one case, increasingly surfacing anti-abortion content in another. This highlighted the critical question of when AI models set communicative limits, especially for vulnerable users who increasingly turn to AI first with sensitive issues.

The following discussion addressed methodological challenges in systematically measuring bias, participatory prompting as a research approach, echo chamber effects in personalised AI systems, and the decolonial dimension of AI infrastructure.

The next Lecture Series event will take place on 3 September 2026 at 4 pm, featuring Lauren Klein (Emory University) on Data Feminism. A new internal workshop series launches on 18 June 2026.