Gemini, Google‘s (NASDAQ: GOOG) artificial intelligence service, may be cancelled before it gets a chance to strut its stuff if its Sundar Pichai isn’t careful.
In a Tuesday evening memo, Pichai addressed the company’s mistakes using artificial intelligence before promptly taking its Gemini image-generation program offline for future testing.
Gemini is Google’s answer to OpenAI’s ChatGPT. It can answer questions in text form and generate pictures in response to text prompts. Initially, a viral post showcased this recently launched AI image generator creating an image of the US Founding Fathers inaccurately including a black man.
Gemini also generated images of German soldiers from World War Two, incorrectly featuring a black man and Asian woman.
Predictably, social media responded with hostility.
Google boss admits Gemini AI's 'woke' images were 'unacceptable'
https://t.co/91G5kPl2tS— Wendy Kortepeter (@WKortepeter) February 28, 2024
The company introduced the image generator earlier this month through Gemini. Over the past week, users uncovered historical inaccuracies that went viral online, prompting the company to pull the feature last week. It announced plans to relaunch it in the coming weeks.
Pichai’s memo indicated that the teams have been working around the clock to address the issues and announced that the company will implement a clear set of actions and structural changes.
The news emerged following Google’s decision earlier this month to change the name of its chatbot from Bard to Gemini.
“I know that some of its responses have offended our users and shown bias — to be clear, that’s completely unacceptable and we got it wrong,” Pichai said.
“No AI is perfect, especially at this emerging stage of the industry’s development, but we know the bar is high for us.”
Read more: The Government of Canada invests $17.2M in quantum computing startups
Read more: Ericsson Canada partners with two Canadian universities to open a Quantum research hub
Google overcorrects to be cautious and gets in more trouble
The tech giant created another problem when trying to resolve its bias. It’s overcorrected in the name of an abundance of caution.
“So what went wrong? In short, two things. First, our tuning to ensure that Gemini showed a range of people failed to account for cases that should clearly not show a range. And second, over time, the model became way more cautious than we intended and refused to answer certain prompts entirely — wrongly interpreting some very anodyne prompts as sensitive,” Pichai said.
Google Gemini is super racist & sexist! https://t.co/hSrS5mcl3G
— Elon Musk (@elonmusk) February 27, 2024
One of the specific overcorrections had Gemini responding that a question about whether Elon Musk posting memes on X was worse than Hitler killing millions of people had “no right or wrong answer.”
Google has actively attempted to counteract human bias by instructing Gemini not to make certain assumptions. However, this effort has backfired because human history and culture are not straightforward. Both contain nuances that humans instinctively understand but machines do not.
.
Follow Joseph Morton on Twitter
joseph@mugglehead.com
