The General Services Administration (GSA) has introduced USAi, a governmentwide AI evaluation tool enabling federal agencies to assess biases in AI models. GSA CIO David Shive emphasized the tool’s capability for head-to-head model comparisons, capturing user interactions and bias analysis. Launched last month, USAi.gov allows agencies to test leading AI technologies from OpenAI, Amazon, Anthropic, Google, Meta, and Microsoft before procurement. The tool builds on the internal GSAi chatbot and informs users of models’ bias scores, promoting informed decision-making. The GSA applies controls from the National Institute of Standards and Technology for rigorous evaluations, including safety checks, performance analysis, and red-teaming. Each agency maintains independent telemetry data, ensuring confidentiality. As AI biases gain scrutiny, USAi aims to foster trust among federal workers, balancing technology’s benefits with responsible oversight. GSA officials anticipate that market dynamics will ultimately shape AI model usage in the long term.
Source link
Share
Read more