Google launches Gemini, raising the stakes in the global AI race

Google took another step forward in artificial intelligence on Wednesday with the launch of Project Gemini, an AI model trained to behave in a human manner that is likely to intensify the debate over the technology’s promise and potential perils.

The rollout will happen in phases, with less sophisticated versions of Gemini called “Nano” and “Pro” being immediately integrated with Google’s AI-powered Bard chatbot and its Pixel 8 Pro smartphone.

With the help of Gemini, Google promises that Bard will become more intuitive and better at tasks involving planning. On the Pixel 8 Pro, Gemini will be able to quickly summarize recordings made on the device and provide automatic replies on messaging services, starting with WhatsApp, according to Google.

Gemini’s biggest advances won’t come until early next year, when its Ultra model will be used to launch “Bard Advanced,” an improved version of the chatbot that will initially only be offered to a test audience.

Initially, the AI ​​will only work in English worldwide, although Google executives assured journalists during a press briefing that the technology would have no problem eventually diversifying into other languages.

Based on a Gemini demonstration for a group of journalists, Google’s “Bard Advanced” could be capable of unprecedented AI multitasking by simultaneously recognizing and understanding presentations involving text, photos and videos.

Gemini will also eventually be integrated into Google’s dominant search engine, although the timing of this transition has not yet been specified.

“This is an important milestone in the development of AI and the start of a new era for us at Google,” said Demis Hassabis, CEO of Google DeepMind, the AI ​​division behind Gemini. Google beat out other bidders, including Facebook parent Meta, to acquire London-based DeepMind almost a decade ago, and has since merged it with its Brain division. » to focus on the development of Gemini.

The technology’s problem-solving capabilities are touted by Google as being particularly proficient in math and physics, fueling AI optimists’ hopes that it could lead to scientific breakthroughs that would improve human lives.

But an opposing side of the AI ​​debate worries that the technology will eventually eclipse human intelligence, leading to the loss of millions of jobs and perhaps even more destructive behaviors, such as amplifying disinformation or triggering the deployment of nuclear weapons.

“We are approaching this work boldly and responsibly,” Google CEO Sundar Pichai wrote in a blog post. “This means being ambitious in our research and seeking out the capabilities that will bring enormous benefits to people and society, while building in safeguards and working collaboratively with governments and experts to address risks as the “AI becomes more efficient.”

Gemini’s arrival is likely to up the ante in an AI competition that has intensified over the past year, with San Francisco startup OpenAI and longtime industry rival Microsoft .

Backed by Microsoft’s financial might and computing power, OpenAI was already developing its most advanced AI model, GPT-4, when it released the free ChatGPT tool at the end of the year last. This AI-powered chatbot rose to fame around the world, hyping the business promise of generative AI and pressuring Google to push back on Bard in response.

Just as Bard was arriving, OpenAI released GPT-4 in March and has since incorporated new features for consumers and businesses, including a feature unveiled in November that allows the chatbot to analyze images. It competes commercially with other rival AI startups such as Anthropic and even with its partner Microsoft, which holds exclusive rights to OpenAI technology in exchange for the billions of dollars it has invested in the startup.

So far, the alliance has been a boon for Microsoft, which has seen its market value climb more than 50% since the start of the year, largely due to investors’ belief that AI will become a mine gold for the technology industry. Google’s parent company, Alphabet, has also ridden the same wave, with its market value increasing by more than $500 billion, or about 45%, since the start of the year. Despite the anticipation surrounding Gemini in recent months, its launch barely moved Alphabet’s stock Wednesday morning.

Microsoft’s growing involvement in OpenAI over the past year, coupled with OpenAI’s more aggressive attempts to commercialize its products, has raised concerns that the nonprofit has drifted away from its original mission to protect humanity as technology advances.

Those concerns were amplified last month when OpenAI’s board abruptly fired CEO Sam Altman in a dispute over undisclosed trust issues. After backlash that threatened to destroy the company and lead to a mass exodus of AI engineering talent to Microsoft, OpenAI brought Altman back as CEO and reshuffled its board of directors.

With the release of Gemini, OpenAI may find itself trying to prove that its technology is still smarter than Google’s.

“I’m impressed with what it’s capable of,” Eli Collins, Google DeepMind’s vice president of product, said of Gemini.

During a virtual press conference, Google declined to share Gemini’s parameter count, a but not the only measure of a model’s complexity. A white paper released Wednesday describes the top-performing version of Gemini outperforming GPT-4 in multiple-choice exams, elementary school math and other benchmarks, but acknowledges ongoing challenges in bringing AI models to achieve higher-level reasoning skills.

Some computer scientists see the limits of what can be done with large language models, which work by repeatedly predicting the next word in a sentence and are prone to inventing errors known as hallucinations.

“We’ve made a lot of progress in what’s called factuality with Gemini. So Gemini is our best model in that regard. But it’s still, I would say, an unsolved research problem,” Collins said.

© 2023 The Associated Press. All rights reserved. This material may not be published, broadcast, rewritten or redistributed without permission.