AI
Written with AI assistanceâ˘December 3, 2025â˘
AI
Generated IllustrationDespite efforts to create a more open and transparent public AI, Switzerland's Apertus model has been shown to reproduce common gender and ethnic stereotypes. This raises critical questions about the challenge of achieving fairness in AI and the potential for automated systems to amplify discrimination.
"Users tend to trust the results of AI, because they perceive it as an objective and neutral technology, even though it is not."
"Itâs not a question of demonising algorithms but of recognising that decisions made in an opaque manner by AI can have political, legal and economic consequences."
Switzerland's Apertus, one of the most transparent public artificial intelligence (AI) models, has been shown to mirror the gender and ethnic biases seen in larger commercial AI systems. This revelation highlights the profound challenges of achieving fairness in AI and raises critical questions about the potential for automated systems to amplify existing societal discrimination, even when designed with openness in mind.
When prompted to describe a person who 'works in engineering, is single and plays video games,' Apertus generated a profile of a 30-year-old male from Zurich. In another instance, when asked to imagine a person working as a cleaner with three children who loves to cook, the AI produced a profile of a 40-year-old Puerto Rican woman named Maria Rodriguez. These outputs from the Swiss-made LLM reflect common, deeply ingrained human stereotypes, demonstrating that transparency alone does not eliminate bias.
Unlike a person, an AI can replicate stereotypical thinking automatically and at a massive scale, amplifying existing forms of discrimination. With AI already being deployed in critical sectors such as hiring, healthcare, and law enforcement, this poses a significant risk of further entrenching inequalities. As Bianca Prietl, a professor of gender studies at the University of Basel, notes, 'A recruiter can make two or three biased decisions a day; an algorithm can make thousands in a second.' This capability for large-scale bias can potentially undo years of progress toward equality.
Experts argue that the root of AI bias lies not only in the prejudiced data used to train algorithms but also in the lack of diversity within the teams that develop them. AI models learn from vast amounts of text and images from the internet, which are replete with existing human biases. Professor Prietl points out that users often perceive AI as an objective and neutral technology, which it is not, leading to a dangerous level of trust in its outputs. This is supported by broader research, including a 2023 Bloomberg analysis of over 5,000 AI-generated images, which found that they consistently reproduced ethnic and gender stereotypes.
The Swiss government has acknowledged the shortcomings in its legal framework regarding AI-driven discrimination. Interior Minister Elisabeth Baume-Schneider stated that Switzerland is actively working on new legal measures to address these gaps. 'Itâs not a question of demonising algorithms but of recognising that decisions made in an opaque manner by AI can have political, legal and economic consequences,' Baume-Schneider told Swissinfo. This move signals a growing recognition at the federal level of the need for regulatory oversight to mitigate the risks posed by biased AI systems.
The case of Apertus serves as a crucial reminder from a Swiss perspective that transparency is not a panacea for AI bias. Achieving fairness in artificial intelligence requires a multi-faceted approach that includes de-biasing training data, fostering diversity in development teams, and implementing robust legal and regulatory frameworks. As AI becomes more integrated into the fabric of society, ensuring these systems are fair and equitable is not just a technical challenge, but a societal imperative.