30.10.2025
©Terzo Algeri/Fotoatelier M/ TUM
Language Shapes Gender Bias in AI Images
TUM News
Alexander Fraser, MCML PI, and his team discovered that AI image generators reproduce gender stereotypes differently across languages. In their study of nine languages, they found that generic prompts like “accountant” mostly produced male images, while explicitly feminine or neutral prompts reduced bias but sometimes affected image quality.
The study highlights that AI is not language‑agnostic and careful wording can influence outcomes, underlining the need for fairness and multilingual awareness in AI systems.
Related
19.01.2026
MCML at AAAI 2026
MCML researchers are represented with 9 papers at AAAI 2026 (6 Main, and 3 Workshops).
19.01.2026
MCML PI Frauke Kreuter Featured on ARD Alpha on AI
MCML PI Frauke Kreuter featured on ARD alpha discussing AI in daily life, workplace applications, and responsible, future-ready use.
15.01.2026
Blind Matching – Aligning Images and Text Without Training or Labels
CVPR 2025 research from Daniel Cremers’ group shows how images and text can be aligned without training data, labels, or paired examples.
12.01.2026
MCML PIs Featured in Süddeutsche Zeitung
MCML PIs Xiaoxiang Zhu and Felix Dietrich featured in Süddeutsche Zeitung for TU Munich’s 2025 breakthroughs in AI and data science.
