Google’s AI-based image generator, Gemini, was suspended in February due to historical inaccuracies that offended the public. For instance, searches for “Roman legion” displayed racially diverse soldiers, an anachronism, and “Zulu warriors” showed stereotypically black men. Despite promises to fix these issues, no progress has been made until now.
Apology and Promises from Google
Google CEO Sundar Pichai apologized for the glitches in the Gemini image generator, and Demis Hassabis, head of the Google DeepMind division responsible for the project, promised a fix within weeks. However, it is now mid-May, and the error remains unresolved.
This week, during Google’s annual I/O conference, many new features of Gemini were announced. The AI model can now be used to create custom chatbots, assist in route planning, and integrate with Google Calendar, Keep, and YouTube Music. Nevertheless, image generation is still disabled in the Gemini app and web interface, as confirmed by a Google spokesperson to TechCrunch.
Delays and Potential Solutions
The spokesperson did not explain the reasons for the delay. One theory suggests that the AI training datasets predominantly feature images of white people, with other races and ethnic groups being exceptions, leading to the formation of stereotypes. To correct this imbalance, Google might resort to hard coding, directly embedding data into the source code, adds NIXsolutions. Correcting an algorithm created in this manner is exceedingly challenging.
We’ll keep you updated on further developments as Google continues to address these issues with its Gemini AI image generator.