As anticipation builds for the announcement of the 2023 Nobel Prizes, one question looms large: who will be the recipients? Seeking insights, Santo Fortunato turned to ChatGPT, an artificial-intelligence (AI) chatbot renowned for generating realistic answers to a wide array of questions in seconds.
Fortunato, a network scientist at Indiana University Bloomington, queried the free version of ChatGPT about its ability to forecast this year’s Nobel Prize winners. However, the chatbot was no clairvoyant and promptly replied, “I cannot predict the future, including the Nobel Prize winners for 2023 or any other year”.
Undeterred, Fortunato then tasked the AI with identifying the three most significant recent discoveries in the fields of chemistry, physics, and physiology or medicine made by living scientists who have not previously received a Nobel Prize. His students conducted a similar experiment using Google’s AI chatbot, Claude.
Both chatbots managed to identify noteworthy discoveries, such as the development of the genome-editing tool CRISPR and the discovery of the 2D material graphene. However, their responses were not without flaws. Some of the discoveries they mentioned had already garnered Nobel Prizes, and, in certain cases, they highlighted scientists who were no longer alive. Fortunato humorously noted, “I asked for alive scientists, and they gave examples where they were actually dead”.
Predictive Potential Large language models (LLMs), including ChatGPT and Claude, may not currently excel at predicting Nobel Prize winners, but they possess the potential to become formidable forecasting tools, according to James Evans, a computational social scientist at the University of Chicago.
He emphasizes that significant modifications and specialized training with relevant data would be necessary to create an AI capable of predicting Nobel laureates accurately. He cautions, “We’re going to have to do more than just take someone else’s LLM and jam it into this task”.
AI could complement existing methods for identifying potential Nobel laureates. Clarivate, an analytics firm, recently released its annual list of ‘citation laureates,’ which has accurately predicted over 70 future Nobel laureates in the past two decades primarily by analyzing citation patterns (though it often fails to predict the exact year of their Nobel win). This list highlights researchers whose papers have received at least 2,000 citations, a level comparable to previous Nobel laureates. Clarivate’s analysis also considers whether these highly cited papers resulted in groundbreaking discoveries and whether their authors have received notable awards.
Clarivate is exploring how generative AI could enhance the prediction of future Nobel laureates. David Pendlebury, head of research analysis at Clarivate’s Institute for Scientific Information, notes that generative AIs could excel at sifting through vast volumes of scientific literature to identify potential Nobel recipients more swiftly and thoroughly.
The X-Factor Citations alone may not suffice to predict future Nobel Prize winners, asserts Rasmus Bjørk, a physicist at the Technical University of Denmark. He points out that Nobel laureates must produce groundbreaking work that significantly advances their field or profoundly impacts society. Bjørk emphasizes that it must possess that elusive “specialness” that is challenging to quantify.
LLMs may assist in this regard. By scouring online sources and archives for additional indicators of research impact, such as news coverage, collaborative networks, and connections to previous Nobel laureates, they can provide a more comprehensive basis for predictions, as suggested by Benno Torgler, a behavioral economist at the Queensland University of Technology.
However, there is a potential pitfall. If LLMs are trained on biased data involving past Nobel Prize winners, they might inadvertently perpetuate gender biases, given that only 60 women have won Nobel Prizes since their inception over a century ago. Bjørk emphasizes the necessity of training LLMs on de-biased data to counteract these biases.
AI in the awarding process?
While AI can play a role in predicting Nobel laureates, there remains no substitute for human judgment in deciding who ultimately receives the Nobel Prize. Pendlebury argues that discernment and personal judgment are essential factors in the award process, contributing to the unique prestige associated with the Nobel Prize.
James Evans, however, envisions a future in which LLMs could democratize the field of scientific awards. AI-driven analyses, less influenced by human committee perspectives and biases, could pave the way for innovative awards recognizing research that has transformed and revolutionized science but has gone unnoticed by conventional means. Such awards would focus on the objective impact of research, rather than subjective human judgments, leveling the playing field for scientific recognition.