Google CEO Sundar Pichai denounced “totally unacceptable” errors in its artificial intelligence application Gemini in a letter to his teams, while the generation of historically inaccurate images led to the suspension on Thursday of creating images of people.
The controversy emerged weeks after Google announced the name change of Bard, its generative AI, to “Gemini”, a competitor to Microsoft-backed OpenAI’s ChatGPT.
On social networks, users pointed out certain inaccurate creations, particularly in terms of gender and origin, appearing to underrepresent white people. A query on a German soldier from 1943, for example, resulted in images of Asian or black-skinned soldiers.
“I would like to discuss the latest difficulties linked to the generation of [contenus] problematic in the Gemini application,” Mr. Pichai wrote on Tuesday, in a letter published by the information site Semafor, authenticated by a Google spokesperson to AFP.
“I know some of his responses were biased and shocked our users. And, to be clear, this is completely unacceptable and we were wrong. »
Sundar Pichai said Google teams were working “around the clock” to resolve the issues, but did not say when people image generation would be available again.
“No AI is perfect, especially at the emerging stage this industry is at, but we know the bar is high and we will continue working as long as it takes,” he wrote.
Since the end of 2022 and the success of ChatGPT, generative AI, capable of producing texts, sounds, images or videos on a simple request in everyday language, has aroused massive enthusiasm and the tech giants have engaged in a race to deploy tools for organizations and individuals.
But AI models have long been criticized for perpetuating racial and gender bias in their results.
Google said last week that Gemini’s problematic generations were due to its efforts to try to eliminate these biases.
Since the appearance of ChatGPT, which revealed the potential of AI to the general public, many experts and governments have warned of certain risks it entails, such as the dissemination of false photographs larger than life, warning against to the danger of manipulation of opinion.