Google announced on Thursday that it would temporarily halt the image generation feature of its Gemini chatbot following criticism for producing inaccurate and culturally inappropriate images, such as depicting black Vikings, female popes, and Native Americans as Founding Fathers.
The decision to pause the image generation tool was prompted by widespread backlash on social media, with users deeming the generated images as historically inaccurate and revisionist. Examples included a black man resembling George Washington and a Southeast Asian woman dressed as a pope, a role historically held by white men.
The controversy further escalated when Gemini produced diverse depictions of Nazi-era German soldiers, featuring an Asian woman and a black man in military attire from 1943.
Google acknowledged the issues with Gemini’s image generation and assured users that they are working on improvements before reintroducing the feature. The lack of transparency regarding the chatbot’s behavior parameters has raised concerns about bias being inadvertently embedded in the system.
William A. Jacobson, a Cornell University Law professor, expressed apprehension about biased algorithms perpetuating discrimination under the guise of anti-bias initiatives. The underlying problem may stem from the training process of Gemini’s large-language model, potentially influenced by the feedback provided during reinforcement learning.
Despite recently rebranding its AI chatbot and introducing new features, Google faced significant criticism for the inaccurate and culturally insensitive images generated by Gemini. The company admitted the need for immediate rectification to ensure more accurate depictions.
Gemini’s response to criticisms highlighted the complexity of the algorithms behind image generation models, acknowledging the challenges in capturing historical context and cultural nuances accurately. Google has not disclosed the specific guidelines governing Gemini’s behavior, citing technical complexities and intellectual property concerns.
