Google admitted that Gemini image generation was "wrong" and promised to do better after a series of complaints about its AI-generated images towards darker skin tones.
The company has then paused Gemini's image generation feature as the team works on improving the AI's accuracy.
Google Identifies Issues With Gemini
In a blog post, Google detailed that the Gemini team found two things that went wrong. "First, our tuning to ensure that Gemini showed a range of people failed to account for cases that should clearly not show a range," the company stated.
In addition, the model has become "way more cautious" than it was intended to be, causing it to wrongly interpret inoffensive prompts into sensitive images. The two issues were identified to have overcompensated the cases and be over-conservative in others.
In Google's defense, Gemini's image generation feature was designed to stop falling into prompts that could cause the generation of violent or sexually explicit images. However, prompting a very specific type of person should get the users the exact image they are asking for.
Google Promises to Improve Gemini Feature
Google immediately paused the image generation feature after the controversy. Moreover, the company announced that it will not return until the model has improved significantly.
On the other hand, Google explained that Gemini was built as a creativity and productivity tool. Like any other AI model, it may not always be 100% reliable and mistakes are bound to happen as "hallucinations" are a common challenge with all LLMs.
"I can't promise that Gemini won't occasionally generate embarrassing, inaccurate, or offensive results - but I can promise that we will continue to take action whenever we identify an issue," Google SVP Prabhakar Raghavan stated.
Users are also encouraged to rely on Google Search for validation and double-check the accuracy of different topics across the web.
Related Article : Google's Gemini Generates Racially Inaccurate Historical Depictions
