Results from NIST’s GenAI Text-to-Text(T2T) Discriminator Challenge

The rapid advancement of Generative AI (GenAI) technologies offers significant benefits and presents substantial risks, particularly in the housing sector. The increasing sophistication of Large Language Models (LLMs) makes it challenging to distinguish between AI-generated and human-generated text, potentially leading to discriminatory practices like racial steering, misleading property listings, and restricted access to housing resources. The National Fair Housing Alliance (NFHA) participated in the NIST GenAI Text-to-Text (T2T) Discriminator Challenge to develop building blocks that can be used to address these concerns. Our findings from the Challenge indicate that certain machine learning models, such as support vector machine (SVM) and extreme gradient boosting (XGBoost), show promise in differentiating between AI-generated and human generated text. However, we encountered persistent difficulties in tracing the origin of AI generated content, raising concerns about accountability and transparency. The implications of these findings extend to platform regulation, transparency measures, auditing and detection tools, literacy and awareness, and high-risk use cases. NFHA’s ongoing participation in this initiative and related research aims to further explore these implications and contribute to the development of responsible GenAI practices in the housing sector.