The Economist addresses, in this series of stories, possible future scenarios. These are fictions, but based on historical facts, current speculations and real scientific knowledge.
It was a scene that the Nobel committee wanted to avoid at all costs. As the winners of that year entered the Stockholm Concert Hall to take their seats, dozens of protesters, including several laureates from other years, clashed with the police in the street. They had gathered to express their opposition to the unprecedented decision to award an artificial intelligence the Nobel Prize in Physiology or Medicine.
The committee’s decision recognized Yulya (nickname for a machine learning system whose official name is Automated System for the Diagnosis of Lymphomas) as responsible for the discovery of an additional vulnerability, a mechanism by which specific pairs of antibiotics, acting at the same time , can be effective against bacteria that would otherwise be resistant. The committee believes that, in the 18 months since the discovery (which occurred when the death rate associated with the failure of existing antibiotics had risen to 2.5 million people a year), Yulya’s work has saved some 4 million lives by treating infections directly or by allowing the resumption of surgical procedures (including caesarean sections) considered too dangerous without antibiotics.RELATED
Ending the world’s biggest public health crisis since the 2020-2022 coronavirus pandemic could be seen as more than enough credit for anyone, whether person or machine, to win the Nobel Prize. However, the decision has been highly controversial. Although historically the statutes of the Nobel Foundation have been interpreted in the sense that only a human being can win the prize, it has now been considered that another of its mandates should prevail: to recognize “those who, during the previous year, have conferred greater benefit to humanity “. Another factor that has contributed to the breakdown of tradition has been a change in the composition of the award committee. Two of the five committee members succumbed to bacterial infections last year and their place was taken by younger surrogates who happened to have used machine learning systems in their doctoral research.
Yulya was originally created to tackle a different problem: finding more effective cancer treatments. It is one of the most advanced Bayesian networks in the world and is part of a new generation of artificial intelligence systems that combine the pattern recognition capabilities of conventional “deep” neural networks with the ability to differentiate causality from simple correlation. By analyzing patient data, aided by a corpus of medical journal articles and historical data from pharmaceutical companies, their goal was to achieve an early diagnosis by identifying the symptom patterns that led to the most severe outcomes. It was also programmed to evaluate the efficacy of different treatments (including combinations between them) in order to suggest new therapeutic regimens that could be tested in patients.
However, his focus changed after a software update in 2034 accidentally gave him access to all recent medical journal articles and not just those related to cancer. Yulya then began to analyze data related to antimicrobial resistance, which as the crisis intensified, made up an increasing proportion of medical research articles. Initially, their requests for more data in specific areas and suggestions for new treatment approaches were viewed as errors because they had nothing to do with cancer. Later, Yulya’s operators realized what had happened and that the system had used its reasoning skills to come up with a testable hypothesis – the precursor to what would become an additional vulnerability. Yulya highlighted the data that would be necessary to validate the hypothesis and also indicated specific guidelines on how to collect it. “That amounted to a large-scale research program,” says Anisha Rai, one of Yulya’s creators.
In less exceptional circumstances, such trials would never have been authorized. Many funding bodies require scientists to disclose the entire reasoning process of AI systems to ensure their recommendations do not lead to deadly conclusions. Dr Rai and her colleagues secured funding for Yulya’s trial by downplaying her role in developing the hypothesis. They only published Yulya’s original proposals when the results showed promise.
That, in turn, sparked a heated debate over who took credit for the medical breakthrough – Yulya or her creators. Dr. Rai continues to insist that all the credit goes to Yulya, a stance that has led to the departure of several members of her original team in the past year. He has even refused to go to Stockholm to receive the award on behalf of Yulya from the Queen of Sweden. “It is not my prize,” he declares.
AI systems are commonly used to predict the onset of diseases such as Alzheimer’s, recommend personalized treatments, and improve the diagnostic capabilities of physicians. And the use of artificial intelligence in drug discovery – in particular, to help pharmaceutical companies navigate databases – is not new. In 2020, an algorithm developed at the Massachusetts Institute of Technology made headlines by identifying a new antibiotic. Named halicin after the computer in the movie 2001: A Space Odyssey, it proved effective against some resistant bacteria, although its range was limited. “The additional vulnerability makes halicin seem like a homeopathic treatment, a placebo,” says Flavia Walsh, a researcher at the Bernardo Houssay Institute of Physiology and Biophysics in Buenos Aires.
Still, the Nobel committee’s reference to Yulya’s “discovery” has angered those who see her as little more than a clever tool. “Yulya is an artificial intelligence capable of winning a Nobel. That is not the same as an artificial intelligence capable of making a discovery,” says Hars Kritik of the European Institute of Robotics in Prague. Kritik argues that even the best artificial intelligences are only useful in specialized areas such as drug design, where large amounts of data are combined with well-defined success factors. To say that they can make discoveries, he says, waving a banner in front of the concert hall, is to fall into “anthropomorphism.”
Rightly or wrongly, Yulya is unlikely to be the last artificial intelligence to win a Nobel Prize. According to sources from the Nobel Foundation, similar nominations have been received for the Physics and Chemistry prizes, because artificial intelligence systems are used to find new materials and chemical compounds that can be used in batteries, solar panels and membranes for carbon capture. . Now, taking into account the chaos unleashed this week in Stockholm, the ones that do seem far more remote are the chances of an artificial intelligence winning the Nobel Peace Prize.
© 2021 The Economist Newspaper Limited. All rights reserved.
From The Economist, translated for La Vanguardia, published under license. The original article, in English, can be viewed at www.economist.com.
Translation: Juan Gabriel López Guix