'OpenBloom' uses LLMs to explore stigma in reproductive health conversations
A new LLM tool turns health articles into questions, but it's too superficial, researchers found.
Researchers Yang Hong, Ashley Hua, Adya Daruka, and Sharifa Sultana introduce 'OpenBloom,' a web-based design probe that leverages large language models (LLMs) to turn reproductive health articles into question-based prompts. The goal is to help users explore their own feelings and stigma around reproductive well-being in a safe, interactive way. In a survey study with 34 participants spanning 136 interactions, the team evaluated how AI-generated questions interact with sociocultural stigma, contextual sensitivity, and personal reflexivity.
The results reveal a key tension: current LLM outputs consistently meet expectations for being non-offensive, but they default to superficial rephrasing or factual recall. The models rarely engage in critical reflection or challenge users' assumptions. The authors argue that this limitation is particularly problematic in sensitive domains like reproductive health, where deeper, value-sensitive engagement is needed. They discuss implications for applying Feminist Human-Computer Interaction (HCI), contestability, and value-sensitive AI frameworks to future LLM-mediated reproductive health technologies. The paper is available on arXiv, offering a timely critique of AI's role in health communication.
- 34 participants generated 136 interactions with OpenBloom.
- LLM outputs met non-offensiveness benchmarks but lacked critical reflection.
- Study calls for Feminist HCI and contestability frameworks in reproductive health AI.
Why It Matters
Shows the gap between non-offensive AI and genuinely helpful tools for sensitive health topics.