AI with Nobel Prices: Double victory sparked discussion about scientific disciplines

AI with Nobel Prices: Double victory sparked discussion about scientific disciplines

The Nobel Committees have recognized the transformative power of artificial intelligence (KI) in two of this year's prices-they honored Pioneers of the neural networks in the physics price and the Developer of Calculation tools for examining and designing proteins in the Chemical Prize. But not all researchers are satisfied.

Only a few moments after the announcement of the winners of this year's Nobel Physics Prize by the Royal Swedish Academy of Sciences, the social media world experienced a flash of discussions. Several physicists argued that the science underlying mechanical learning research, which was celebrated in the awards for Geoffrey Hinton and John Hopfield, was not actually physics.

"I am speechless. I appreciate machine learning and artificial neural networks as well as any other, but it is difficult to see that this is a physical discovery," wrote Jonathan Pritchard, an astrophysicist at Imperial College London, on x . "I guess the Nobel Prize was hit by the AI ​​hype."

The research of Hinton at the University of Toronto in Canada and Hopfield at Princeton University in New Jersey "belongs to the area of ​​computer science," says Sabine Hossenfelder, a physicist at the Munich Center for Mathematical Philosophy in Germany. "The annual Nobel Prize is a rare opportunity for physics - and the physicists - to step into the spotlight. It is the day when friends and family remember that they know a physicist and maybe ask what this last Nobel Prize was about. But not this year."

unite some perspectives

Not everyone was worried: many physicists welcomed the news. "The research of Hopfield and Hinton was interdisciplinary and brought physics, mathematics, computer science and neurosciences together," says Matt Strassler, a theoretical physicist at Harvard University in Cambridge, Massachusetts. "In this sense, it belongs to all of these specialist areas."

Anil Ananthaswamy, a science journalist from Berkeley, California and author of the book "Why Machine Learn", notes that the research cited by the Nobel Committee is not theoretical physics in the purest sense, but is rooted in techniques and concepts from physics, such as energy. The "Boltzmann networks" invented by Hinton and the Hopfield networks "are both energy-driven models," he says.

The connection to physics became weaker in the later developments in machine learning, adds Ananthaswamy, especially with the "feedforward" techniques that made neural networks easier to train. Nevertheless, physical ideas return and help to understand researchers why the increasingly complex deep learning systems do what they do. "We need the way of thinking of physics to study machine learning," says Lenka Zdeborová, which researches the statistical physics of the calculation of the Swiss Federal Institute for Technology in Lausanne (EPFL).

"I think that the Nobel Prize for Physics should continue to penetrate more and more areas of physical knowledge," says Giorgio Parisi, a physicist at the Sapienza University Rome, the shared the Nobel Prize 2021 . "Physics is becoming increasingly wider and includes many areas of knowledge that have not existed in the past or were not part of physics."

not just ki

The computer science seemed to take over the Nobel Prize the day after the physics award was announced, as Demis Hassabis and John Jumper, co-founder of Ki-Tools for protein structure forecast Alphafold at Google Deepmind in London, half of the Chemistry Nobel Prize. (The other half was awarded to David Baker from the University of Washington in Seattle for works on protein design that do not use machine learning).

The prize was recognition of the disruptive power of the AI, but also the constant increase in knowledge in structural and computer -aided biology, says David Jones, a bioinformatist at the University of College London, who worked with Deepmind on the first version of Alphall. "I don't think Alphafold represents a radical change in the underlying science that was not already available," he says. "It's just about how everything was put together and designed so that Alphafold could reach these heights."

A keyinput that Alphafold uses are the sequences of related proteins from various organisms that can identify amino acid pairs that are probably ko-evolution and therefore may be close to the 3D structure of a protein. Researchers have already used this knowledge to predict protein structures when Alphafold was developed, and some even began to implement the idea in deep learning networks.

"It was not easy that we went to work, pressed the AI ​​button and then everyone went home," said Jumper at a press conference at Deepmind on October 9th. "It was really an iterative process in which we developed, researched and tried to find the right combinations between what the community understood about proteins and how we could incorporate these intuitions into our architecture."

Alphafold would not have been possible if the protein database had not existed, a freely accessible repository of more than 200,000 protein structures-including some that contributed to previous Nobel prices-which were determined using X-ray crystallography, cryo-electron microscopy and other experimental methods. "Every data point is the result of years of effort from someone," said Jumper.

Since its foundation in 1901, Nobel prices have often been a reflection of the influence of research on society and have rewarded practical inventions, not just pure science. In this regard, prices in 2024 are not outliers, says Ananthaswamy. "Sometimes they are awarded for very good engineering projects. Data-label = "https://www.nobelprize.org/physics/1964/summary/" Data-track-category = "Body text link"> laser and PCR . "