The AI Nobel Prizes Might Shift the Direction of Research
This week has been significant for advertising artificial intelligence research. However, might significant victories for Demis Hassabis and Geoffrey Hinton alter wider scientific motivations?
Demis Hassabis was unaware that he would receive the Nobel Prize in chemistry from the Royal Swedish Academy of Sciences until his wife began receiving numerous calls from a Swedish number on Skype.
“Several times she would set it aside, but they continued to push,” Hassabis stated today during a press conference held to honor the prize presentation, along with his fellow Google DeepMind colleague John Jumper. “Then I believe she understood it was a Swedish number, and they requested my number.”
His victory in securing the prize—the foremost accolade in science—might not have been too surprising: Just a day prior, Geoffrey Hinton, frequently referred to as a “godfather of AI,” along with Princeton University’s John Hopfield, received the Nobel Prize in physics for their contributions to machine learning. “Clearly, the committee chose to make a statement, I suppose, by having the two together,” stated Hassabis during a press conference held after his victory.
To clarify: AI is present, and one can now achieve a Nobel Prize by researching it and impacting other disciplines—such as physics with Hinton and Hopfield or chemistry with Hassabis and Jumper, who shared the honor with David Baker, a genome scientist from the University of Washington.
“There’s no question this is a significant ‘AI in science’ moment,” states Eleanor Drage, senior research fellow at the Leverhulme Center for the Future of Intelligence at the University of Cambridge. “Based on renowned and distinguished computer scientists receiving a chemistry and a physics prize, we’re all anticipating who will get a peace prize,” she states, noting that coworkers in her office were joking about xAI’s owner Elon Musk being favored for that honor.
Drage describes the granting of physics and chemistry prizes to AI researchers as “a significant controversy, both internally within those fields and externally to observers.” She proposes that the awards might be attributed to one of two factors: either a significant change in disciplinary lines made possible by the omnipresence of AI in research, or because “we’re so enamored with computer scientists that we’re ready to place them everywhere.”
She is uncertain about the direction that this week’s choices indicate. However, she and others are confident that it will significantly impact the future of research.
“Securing a Nobel through AI might be a ship that has already sailed, but it will shape the future of research,” states Matt Hodgkinson, an independent specialist in scientific research integrity and former research integrity manager at the UK Research Integrity Office. The issue is if it will affect them positively.
Baker, a recipient of this year’s Nobel Prize in chemistry, has been a prominent figure in the research of AI applications for predicting protein structures. For decades, he had been diligently working on the issue, achieving small advancements, understanding that the clearly defined problem and format of protein structure provided an effective testing ground for AI algorithms. This was not a fleeting success narrative—Baker has authored over 600 papers throughout his career—nor was AlphaFold2, the Google DeepMind initiative that received the accolade from the committee.
However, Hodgkinson is concerned that scholars in the discipline will focus on the method instead of the science while attempting to understand why the trio was awarded the prize this year. “What I hope this doesn’t result in is researchers misusing chatbots, by mistakenly believing that all AI tools are the same,” he states.
The anxiety that this might occur is based on the surge of interest in other allegedly revolutionary technologies. “Hodgkinson states, ‘There are always hype cycles, with the latest ones being blockchain and graphene.’” After the discovery of graphene in 2004, Google Scholar reports that 45,000 academic articles discussing the material were released from 2005 to 2009. However, following the Nobel Prize victory of Andre Geim and Konstantin Novoselov for their discovery of the material, the published papers surged, reaching 454,000 from 2010 to 2014, and exceeding one million from 2015 to 2020. This increase in research has arguably had only a small effect on the real world thus far.
Hodgkinson thinks that the motivating effect of several researchers being acknowledged by the Nobel Prize committee for their contributions to AI might lead to more individuals gathering in this area, potentially leading to science with variable quality. “Whether the proposals and applications [of AI] have any real substance is a different issue,” he states.
We have witnessed the influence of media and public focus on AI within the academic community. Research from Stanford University indicates that the volume of AI publications has increased threefold from 2010 to 2022, with close to 250,000 papers released in 2022 alone: averaging over 660 new works daily. That was prior to the November 2022 launch of ChatGPT, which ignited the generative AI movement.
The degree to which scholars are inclined to pursue media focus, funding, and accolades from the Nobel Prize committee is a matter that troubles Julian Togelius, an associate professor of computer science at New York University’s Tandon School of Engineering, who specializes in AI. “Researchers typically adhere to a mix of the path of least resistance and maximal value,” he states. Considering the competitive landscape of academia, where funding is dwindling and closely tied to the job opportunities of researchers, it appears that the allure of a fashionable subject that—as of this week—might lead high achievers to win a Nobel Prize could be impossible to ignore.
The danger is that this might hinder creative new ideas. “Obtaining additional fundamental information from nature and developing new theories that people can comprehend are challenging tasks,” states Togelius. However, that necessitates profound contemplation. It is much more effective for researchers to perform AI-driven simulations that bolster current theories and utilize available data—leading to incremental progress in comprehension, rather than massive breakthroughs. Togelius predicts that a fresh cohort of researchers will ultimately take this route, as it is more convenient.
There’s a possibility that overly self-assured computer scientists, who have contributed to the progress of AI, might notice AI-related achievements being recognized with Nobel Prizes in different scientific areas—in this case, physics and chemistry—and choose to pursue similar paths, infringing on others’ domains. “Computer scientists are often known for meddling in areas outside their expertise, implementing algorithms, and labeling it progress, whether positively or negatively,” states Togelius, who confesses he once contemplated applying deep learning to a different scientific discipline to “advance” it, but reconsidered due to his lack of knowledge in physics, biology, or geology.
Hassabis exemplifies the effective use of AI to further scientific progress. He trained as a neuroscientist, earning a PhD in the field in 2009, and attributed that expertise to aiding the progress of AI through Google DeepMind. However, he also recognized a shift in the way the industry achieves efficiencies. “Nowadays, [AI] has shifted to being more focused on engineering,” he stated during his Nobel Prize press conference. “We possess numerous techniques now that we’re enhancing purely algorithmically, no longer referencing the brain.”
This could also influence the type of research conducted—and who conducts it, their expertise in the area, and the motives that drive them to pursue it. Instead of researchers who focus their lives on a specific area, we might observe increased research from computer scientists, disconnected from the actual context of their studies. However, that is expected to take a secondary role to the celebrations for Hassabis, Jumper, and the coworkers they both expressed gratitude toward for assisting them in securing the Nobel Prize this week. “We’re nearly finished refining the [AlphaFold3] code so it can be made available for the academic community to use freely,” he stated earlier today. “Next, we will continue to advance from that point.”
