Will AI ever win a Nobel Prize?

Artificial intelligence is already disrupting industries from banking and finance to cinema and journalism, and scientists are exploring how AI could revolutionize their fields – or even win a Nobel Prize.

In 2021, Japanese scientist Hiroaki Kitano proposed what he called the “Nobel Turing Challenge,” inviting researchers to create an “AI scientist” capable of autonomously conducting research worthy of a Nobel Prize. here 2050.

Some scientists are already hard at work trying to create a fellow AI worthy of a Nobel Prize, with this year’s winners expected to be announced between October 7 and 14.

And in fact, there are already around 100 “robot scientists,” according to Ross King, professor of artificial intelligence at Chalmers University in Sweden.

In 2009, King published a paper in which he and a group of colleagues introduced “Robot Scientist Adam,” the first machine to make scientific discoveries independently.

“We built a robot that discovered new science on its own, generated new scientific ideas, tested them and confirmed that they were correct,” King told AFP.

The robot was configured to autonomously formulate hypotheses and then design experiments to test them.

He would even program laboratory robots to carry out these experiments, before learning from the process and repeating it.

“Not trivial”

“Adam” was tasked with exploring the inner workings of yeast and discovered previously unknown “gene functions” in the body.

In the document, the creators of the scientific robot noted that while the findings were “modest,” they were also not “trivial.”

Later, a second scientific robot, named “Eve”, was created to study drug candidates for malaria and other tropical diseases.

According to King, scientific robots already have several advantages over the average human scientist.

“It costs less to do science, they work 24/7,” he explained, adding that they are also more diligent in recording every detail of the process.

At the same time, King admitted that AI is far from comparable to a scientist worthy of a Nobel Prize.

To do this, they would have to be “much smarter” and able to “understand the bigger picture.”

“Nowhere”

Inga Strumke, associate professor at the Norwegian University of Science and Technology, said that for the moment the scientific profession is safe.

“Scientific tradition is far from being supplanted by machines any time soon,” she told AFP.

However, Strumke added that “that doesn’t mean it’s impossible,” adding that it is “absolutely” clear that AI has and will have an impact on the way science is conducted.

An example of how it is already being used is AlphaFold, an AI model developed by Google DeepMind, which is used to predict the three-dimensional structure of proteins based on their amino acid.

“We knew there was a relationship between amino acids and the final three-dimensional shape of proteins…and then we could use machine learning to find it,” Strumke said.

She explained that the complexity of these calculations was too intimidating for humans.

“We kind of have a machine that does something that no human can do,” she said.

At the same time, for Strumke, the AlphaFold case also demonstrates one of the weaknesses of current AI models such as neural networks.

They are very good at processing huge amounts of information and coming up with an answer, but not very good at explaining why that answer is correct.

So while the more than 200 million protein structures predicted by AlphaFold are “extremely useful,” they “don’t tell us anything about microbiology,” Strumke said.

Aided by AI

For her, science seeks to understand the universe and is not simply about “making the right guess.”

Yet the groundbreaking work done by AlphaFold led the experts to top the rankings for a Nobel Prize.

Google DeepMind director John Jumper and CEO and co-founder Demis Hassabis have already received the prestigious Lasker Award in 2023.

The Clarivate analysis group, which monitors potential winners of the scientific Nobel Prize, places the two men among the top choices for 2024 candidates for the chemistry prize, announced October 9.

David Pendlebury, head of the research group, admits that even if a paper by Jumper and Hassabis published in 2021 was cited thousands of times, it would be out of place for the Nobel jury to reward work so quickly after its publication, such as the most of the discoveries honored. date back decades.

At the same time, he is confident that it won’t be long before AI-assisted research wins the most coveted scientific prizes.

“I am sure that in the next decade there will be Nobel Prize winners who will be helped in some way by computation and, these days, computation is increasingly AI “, Pendlebury told AFP.

© 2024 AFP