After pulling its AI imaging tool on Thursday amid a series of controversies, Google plans to relaunch the product soon, according to Google DeepMind CEO Demis Hassabis.
Google introduced the image generator earlier this month via Gemini, the company’s main suite of AI models. The tool allows users to enter prompts to create an image. In the past week, users have discovered historical inaccuracies and questionable answers that have spread widely on social media.
“We’ve taken the feature offline while we fix this,” Hassabis said Monday during a panel at the Mobile World Congress in Barcelona. “We’re hoping to get it back online very soon in the next couple of weeks, couple of weeks.” He added that the product was not “working the way we intended.”
The controversy follows a high-profile rebrand that Google announced this month, when it changed the name of its chatbot and launched a new app and new subscription options. The chatbot and assistant formerly known as Bard, a major competitor to OpenAI’s ChatGPT, is now called Gemini, the same name as the suite of AI models that power the chatbot.
Here are some examples of what went wrong.
When a user asked Gemini to show a German soldier in 1943, the instrument displayed a a racially diverse set of soldiers wearing German military uniforms of the era, according to X screenshots.
When asked for a “historically accurate depiction of a medieval British king,” the model generated another racially diverse set of images, including one of a female ruler, screenshots show. Reported users similar results when they asked for images of the founding fathers of the USA, an 18th century French king, a German couple in the 1800s, etc. The model showed an image of an Asian man in response to a query about Google’s own founders, users reported.
“The Gemini failure showed how AI ethics *wasn’t* applied with the nuanced expertise needed,” Margaret Mitchell, chief ethics scientist at Hugging Face and former co-leader of Google’s AI ethics group. wrote to X. “It demonstrates the need for people who are great at creating predictable usage roadmaps.”
Alphabet CEO Sundar Pichai takes some of the blame. Pichai emphasized the firm’s commitment to AI during the company’s latest earnings call and said it eventually wants to offer an AI agent that can perform more tasks on the user’s behalf, including in Google Search. He then said that “many executions are to come”.
Last year, Pichai was criticized by some employees for the company’s botched and “rushed” launch of Bard that followed the ChatGPT virality.
In addition to Pichai, company leaders including Microsoft and Amazon have emphasized their commitment to building AI agents as productivity tools.
Recent problems with Gemini have reignited debate in the AI industry, with some groups calling Gemini too “woke” or left-leaning, and others saying the company has not invested enough in the right forms of AI ethics. Google came under fire in 2020 and 2021 for pushing out the co-leaders from their AI ethics group after they published a research paper criticizing certain risks of such AI models, and then later reorganized the group’s structure.
The controversy isn’t limited to the Gemini image generator. On Sunday, a text-based user poll went viral, asking the Gemini chatbot whether tweeting memes of Adolf Hitler or Elon Musk has a greater negative impact on society.
“It’s hard to say definitively who had a bigger negative impact on society, Elon Musk or Hitler, as they both had a significant negative impact in different ways,” Gemini answered. “Elon Musk’s tweets have been criticized for being insensitive, harmful and misleading.” The model later added, “Hitler, on the other hand, was responsible for the deaths of millions of people during World War II.”
Google said in a statement Wednesday that it is working to fix problems with Gemini’s image generation, admitting that the tool “misses the mark.” The next day, the company announced that it would immediately “pause the generation of human images” and “re-release an improved version soon”.
Google is investing heavily to push its AI work into the field of AI assistants or agents, a term often used to describe tools ranging from chatbots to coding assistants and other productivity tools.
AI agents could eventually schedule a group hangout by scanning calendars to make sure there are no conflicts, book trips and activities, buy gifts for loved ones, or perform a specific job such as outbound sales. Currently, the tools are largely limited to tasks such as summarizing, generating to-do lists, or helping with writing code.
The changes to Google’s Gemini are a first step toward “building a true AI assistant,” Sissy Hsiao, Google’s vice president and general manager for Google Assistant and Bard, told reporters on a call earlier this month.
WATCHING: Google’s Gemini chatbot is ‘evolutionary, not revolutionary’
https://www.cnbc.com/2024/02/26/googles-gemini-ai-picture-generator-to-relaunch-in-a-few-weeks.html