Google’s conversational AI assistant Gemini faced backlash earlier this year when its image generation feature produced historically inaccurate and racially biased pictures. In response, Google suspended this ability of Gemini to create images. Now, Google DeepMind CEO Demis Hassabis has promised that Gemini will be back in operation within a few weeks with improvements.
When Google launched Gemini as an upgraded version of its previous AI chatbot Bard, it introduced new functionalities like image generation. However, users quickly noticed issues with the types of images being created in response to certain prompts. For example, when asked to generate pictures of historical figures like popes or English kings, Gemini sometimes depicted them with darker skin tones inaccurately. There were also complaints that it had difficulty showing “white people” appropriately.
This led to accusations that Gemini was being “too woke” and was racially biased against depicting white individuals. Many felt the AI was getting basic historical facts wrong. In response to the backlash and user reports of inaccurate and inappropriate images, Google paused Gemini’s image generation ability to address the issues. The company acknowledged that this feature “missed the mark” and needed improvements.
Now, Google DeepMind CEO Demis Hassabis has provided an update on the status of Gemini. In a statement to Reuters, Hassabis said that with Gemini, things did not go as intended by the company. He explained they are working to fix the shortcomings that were reported. Hassabis added that in a few weeks time, the suspended image generation feature of Gemini will return after the necessary changes have been made.
According to Hassabis, Google and DeepMind “care about historical accuracy”. The CEO mentioned they took the image creation ability offline temporarily to resolve the inaccuracies, and hope to restore this functionality in the next “couple of weeks, a few weeks”. He made it clear the tool was “not working the way we intended” previously.
In a blog post earlier, Google search head Prabhakar Raghavan had also acknowledged Gemini “missed the mark” for image generation. He thanked users for their feedback and apologized for the issues. Raghavan explained they aims to ensure diversity but some cases were inappropriate. The AI also became overly cautious over time, refusing prompts wrongly deemed sensitive.
With DeepMind CEO Hassabis now providing a timeline, users can expect to see Gemini’s image generation feature restored within the next few weeks once Google has improved it. The company is working to enhance Gemini’s ability to accurately portray people of all backgrounds in a historically appropriate and unbiased manner through additional training of its systems. This will hopefully address the concerns raised about the AI tool when it was previously rolled out to the public.
