Google Deepmind claims it made a “historic” artificial intelligence breakthrough similar to the deep blue computer that defeated Garry Kasparov in chess in 1997 and the AI-like one that defeated AI in 2016.
The company’s version of Gemini 2.5 AI model became the first AI model to win a gold medal at the International Programming Competition held in Azerbaijan earlier this month, solving complex real-world problems that baffle human computer programmers.
In what tech companies call “a deep leap in abstract problem solving,” it took less than 30 minutes to solve a way to weigh the possibilities of an infinite number of possibilities to send liquids through a network of ducts into interconnected reservoirs. The goal was to distribute it as soon as possible.
Human teams, including top performers from universities in Russia, China and Japan, did not do that right.
AI failed two of the 12 tasks configured, but its overall performance was ranked second out of 139 of the world’s strongest university-level computer programmers. Google said it was “towards a historic moment, AGI (artificial general information).”
“For me, it’s the equivalent of the deep blue of chess and deep blue of alphago,” said Quoc Le, vice-president of Google Deepmind.
“There’s a reason we’re heading towards the real world, not just a larger, more constrained environment (such as chess and go) – so I think this advancement could change many scientific and engineering fields,” he cited the design of drugs and chips.
This model is a general purpose AI, but was specially trained to solve very hard coding, mathematical and reasoning problems. According to Google, they performed “The World’s Top 20 Coders.”
“Resolving complex tasks in these competitions requires deep abstract reasoning, creativity, the ability to integrate new solutions to problems you’ve never seen before, and a real spark of ingenuity,” the company said.
Speaking before details are made public, Stuart Russell, a professor of computer science at the University of California, Berkeley, said “the groundbreaking importance claims seem exaggerated.”
He said that AI systems have been doing well in programming tasks for some time, and the deep blue chess breakthrough “basically does not affect the real world of applied AI.”
However, he said, “Because code actually needs to function correctly (at least in a finite number of test cases) in order to get the ICPC (International University Programming Contest) questions correctly, this performance could indicate advances towards making AI-based coding systems accurate enough to generate high-quality code.”
He added: “The pressure for AI companies to continue to advocate for breakthroughs is enormous.”
Michael Woldridge, a basic professor of artificial intelligence at Oxford University, said impressive results and “it’s exciting to be able to solve problems at this level.”
However, he questioned how much computing power it would require. Google refused to say it except that it would use the lightweight version of the Gemini 2.5 Deep Think in the Gemini app to make sure that the average subscriber is more than it is available for the $250 Google AI Ultra service.
“We are pleased to announce that we are committed to providing a wide range of services,” said Dr. Bill Poucher, executive director of ICPC. “Gemini has successfully made its arena, marking the key moments of achieving gold-level results and defining the AI tools and academic standards needed for the next generation.”
Dive weekly into the way technology shapes our lives
Privacy Notice: Newsletters may contain information about charities, online advertising, and content funded by external parties. If you do not have an account, create a guest account at TheGuardian.com to send this newsletter. You can complete your full registration at any time. For more information on how you use your data, please see our Privacy Policy. We use Google Recaptcha to protect our website and the application of Google Privacy Policy and Terms of Use.
After the newsletter promotion
Four Machine Intelligence Breakthroughs
1957 Perceptron
Cornell academic Frank Rosenblatt thought it should be possible to create “automata perception and recognition.” He named it Perceptron and stated that electronic systems can learn to recognize patterns of optical, electrical, or tonal information “in a way that may closely resemble the perceptual processes of the biological brain.”
The following year he built a device that was small room sized. This was considered one of the early breakthroughs in artificial intelligence based on neural networks.
1997 Big Blue
In May 1997, IBM’s Big Blue became the first computer system to defeat a real world chess champion in a standard tournament control match. It became an inflection point for computing power, beating Garry Kasparov, but the contest was close.
Kasparov won the first game. Deep Blue won Game 6 and secured the victory. It showed how brute-force computing power can create a system that defeats humans, albeit a narrow task. “The computers are much stronger than everyone expected,” Kasparov admitted defeat.
2016 Alphago
Go is one of the most complicated games ever devised, with one of the world’s master players being Lee Sedol, a Korean expert. In 2016, Deepmind, a British AI company founded by Demis Hassabis, took the computer Alphago.
It won 4-1 and some of the moves seemed to show genuinely original thinking. Movement 37 in particular has been reduced due to legend. Hassavis said: “It gives us a glimpse into a bright and bold future that will help humanity use AI as a powerful new tool and discover new knowledge that can solve some of the most pressing scientific problems.”
2020 Alphafold
Another breakthrough by Hassabis and Deepmind was an AI program that could predict how proteins fold into 3D shapes. This is a fundamental and extremely complicated process for understanding the biological machinery of life. The Royal Society, a 360-year-old London scientific institution, calls it “a spectacular progress.”
When researchers learn how proteins fold, they can begin to uncover the mysteries of how insulin controls sugar levels in the blood, and how antibodies fight the virus, among other things. After further iterations, the system helped Hassavis and his colleague John Jumper share the Nobel Prize in Chemistry in 2024.
