Google has decided to temporarily halt Gemini’s image generation of people to enhance its accuracy. Earlier this month, Google introduced a new image generation feature for the Gemini conversational app (formerly known as Bard), allowing users to create images of people.
However, some generated images were inaccurate or offensive, prompting Google to acknowledge the issue and express gratitude for user feedback.
Acknowledging Mistakes and Explanation
Google admitted the flaws in the image generation feature, which was based on the AI model Imagen 2. They explained that while tuning the feature, they aimed to prevent issues seen in previous image generation technologies, such as creating inappropriate images or biased depictions.
Google emphasized their commitment to making the feature inclusive for users worldwide. However, despite their efforts, the feature encountered two major problems.
- Firstly, the tuning failed to address scenarios where a range of people was unnecessary.
- Secondly, the model became overly cautious over time, refusing to respond to certain prompts, even benign ones, leading to embarrassing and inaccurate outcomes.
Improved Version Gemini’s image generation – Coming Soon
Prabhakar Raghavan, Google’s Senior Vice President, stated that the intention was never to discriminate or provide inaccurate representations.
To rectify the issue, Google has temporarily disabled the image generation feature for people and plans to improve it extensively before reactivation, including rigorous testing.
Additionally, Raghavan emphasized that Gemini, as a creativity and productivity tool, may not always be reliable, especially concerning current events or contentious topics. He acknowledged the challenge of AI inaccuracies, highlighting ongoing efforts to enhance reliability.
Raghavan recommended using Google Search for factual information, as separate systems curate fresh and reliable content from various sources.
Commenting on the announcement, Prabhakar Raghavan, Senior Vice President at Google, said:
I can’t guarantee that Gemini won’t occasionally produce embarrassing, inaccurate, or offensive outcomes — but I can assure you that we will consistently address any problems we encounter. AI is an emerging technology with vast potential for positive impact, and we’re committed to deploying it safely and responsibly.