Google temporarily suspended its Gemini AI tool’s ability to generate human images after receiving complaints about inaccuracies. Users reported instances where Gemini placed individuals of color in historical contexts where their presence would be improbable, such as depicting an African American as a founding father.
Initially, Google attempted to resolve the issue, but it proved more complex than anticipated. Consequently, the company disabled the feature of generating human images altogether. Prabhakar Raghavan, Senior Vice President of Google, provided further insight into the situation in a blog post.
Raghavan clarified that while the AI was trained to ensure diversity, its training failed to consider historical contexts where diversity would be inappropriate. Additionally, Gemini misinterpreted certain innocent prompts as sensitive, leading to excessive caution.
This resulted in images that Raghavan characterized as “embarrassing and incorrect.” Google is now planning substantial changes to Gemini, including extensive testing to guarantee the AI’s ability to generate historically accurate images. The image generation feature will remain disabled until these improvements are implemented.
