.png)
2 hours ago
1
Google DeepMind claims it has made a “historic” artificial intelligence breakthrough akin to the Deep Blue computer defeating Garry Kasparov at chess in 1997 and an AI beating a human Go champion in 2016.
A version of the company’s Gemini 2.5 AI model solved a complex real-world problem that stumped human computer programmers to become the first AI model to win a gold medal at an international programming competition held earlier this month in Azerbaijan.
In a performance that the tech company called a “profound leap in abstract problem-solving”, it took less than half an hour to work out how to weigh up an infinite number of possibilities in order to send a liquid through a network of ducts to a set of interconnected reservoirs. The goal was to distribute it as quickly as possible.
None of the human teams, including the top performers from universities in Russia, China and Japan, got it right.
It failed two of the 12 tasks it was set, but its overall performance ranked it in second place out of 139 of the world’s strongest college-level computer programmers. Google said it was a “historic moment, towards AGI [artificial general intelligence]”, which is widely considered human-level intelligence at a wide range of tasks.
“For me it’s a moment that is equivalent to Deep Blue for Chess and AlphaGo for Go,” said Quoc Le, Google DeepMind’s vice-president. “Even bigger, it is reasoning more towards the real world, not just a constrained environment [like Chess and Go] … Because of that I think this advance has the potential to transform many scientific and engineering disciplines.” He cited drug and chip design.
The model is a general purpose AI but was specially trained to solve very hard coding, maths and reasoning problems. It performed “as well as a top 20 coder in the world”, Google said.
“Solving complex tasks at these competitions requires deep abstract reasoning, creativity, the ability to synthesise novel solutions to problems never seen before and a genuine spark of ingenuity,” the company said.
Speaking before the details were made public, Stuart Russell, a professor of computer science at the University of California at Berkeley, said the “claims of epochal significance seem overblown”. He said AI systems had been doing well on programming tasks for a while and the Deep Blue chess breakthrough had “essentially no impact on the real world of applied AI”.
However, he said “to get an ICPC question right, the code actually has to work correctly (at least on a finite number of test cases), so this performance may show progress towards making AI-based coding systems sufficiently accurate for producing high-quality code”.
He added: “The pressure on AI companies to keep claiming breakthroughs is enormous”.
Michael Wooldridge, Ashall professor of the foundations of artificial intelligence at the University of Oxford, said it sounded like an impressive achievement and “being able to solve problems at this level is exciting”. But he questioned how much computing power was needed. Google declined to say, apart from confirming it was more than available to an average subscriber to its $250-a-month Google AI Ultra service using the lightweight version of Gemini 2.5 Deep Think in the Gemini App.
Dr Bill Poucher, executive director of the ICPC, said: “Gemini successfully joining this arena, and achieving gold-level results, marks a key moment in defining the AI tools and academic standards needed for the next generation.”
Four machine intelligence breakthroughs
1957 The Perceptron
after newsletter promotion
Frank Rosenblatt, an academic at Cornell University,worked out that it should be possible to create a “perceiving and recognising automaton”. He named it the Perceptron and said an electronic system would be able to learn to recognised patterns in optical, electrical or tonal information “in a manner which may be closely analogous to the perceptual process of a biological brain”. The following year he built the device, which was the size of a small room. It 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 reigning world chess champion in a match under standard tournament controls. It beat Garry Kasparov in what became an inflection point in computing power, but the contest was close. Kasparov won the first game, Deep Blue the second followed by three draws. Deep Blue won game 6 to secure the win. It showed how brute force computing power could create a system to defeat a human, albeit at a narrow task. “The computer is far stronger than anybody expected,” said Kasparov, conceding defeat.
2016 AlphaGo
Go is one of the most complex games ever devised, and one of the world’s master players was Lee Sedol, a South Korean professional. In 2016, DeepMind, the UK AI company set up by Demis Hassabis, took him on with its computer AlphaGo. It won 4-1 and some of its moves seemed to display truly original thinking. Move 37 in particular went down in lore. Hassibis said: “It might be the first glimpse of a bright and bold future where humanity harnesses AI as a powerful new tool, helping us discover new knowledge that can solve some of our most pressing scientific problems.”
2020 AlphaFold
Another breakthrough by Hassibis and DeepMind was an AI program that can predict how proteins fold into 3D shapes, a highly complex process fundamental to understanding life’s biological machinery. The Royal Society, the 360-year old London scientific institution, called it “a stunning advance”.
When researchers know how a protein folds up, they can start to uncover mysteries such as how insulin controls sugar levels in the blood or how antibodies fight viruses. After further iterations, the system helped Hassibis and his colleague John Jumper share a Nobel prize for chemistry in 2024.
