In light of discovered historical inaccuracies related to racial depictions, Google has temporary suspended its image generator in the Gemini chatbot.
The decision to fine-tune the Gemini image generator was taken less than 24 hours after the first reports of malfunctions were received. When users requested images of historical groups or individuals, the AI mostly presented non-Caucasian figures. These anomalies led to online speculations that Google intentionally avoids portraying Caucasians.
For instance, when Gemini was asked to render a 19th-century American senator, it generated images of Black and Native American women. Given the first female senator in the US, who was white, started office only in 1922, the Gemini image generator inadvertently erased traces of racial and gender discrimination.
When Gemini is currently asked to generate a human image, it responds: “We are working on improving the ability to generate images of people. We expect this feature to return soon, and we will inform about the update release when it happens.” The image generator feature was incorporated into the Gemini chatbot (formerly known as Bard) earlier this month in line with similar offerings by OpenAI and Microsoft’s Copilot, generating images based on textual prompts.
This post was last modified on 02/22/2024