Has AI Achieved Nobel Supremacy? The "AI Moment" in Science Has Arrived?

Has AI Achieved Nobel Supremacy? The "AI Moment" in Science Has Arrived?On October 10th, the field of artificial intelligence research witnessed a landmark event as Demis Hassabis, CEO of Google DeepMind, and Geoffrey Hinton, hailed as the "godfather of AI," were awarded the Nobel Prize in Chemistry and Physics, respectively. This groundbreaking achievement has sparked widespread discussion: will AI usher in a revolution of scientific incentive mechanisms?Hassabis was ecstatic upon learning about his Nobel Prize in Chemistry

Has AI Achieved Nobel Supremacy? The "AI Moment" in Science Has Arrived?

On October 10th, the field of artificial intelligence research witnessed a landmark event as Demis Hassabis, CEO of Google DeepMind, and Geoffrey Hinton, hailed as the "godfather of AI," were awarded the Nobel Prize in Chemistry and Physics, respectively. This groundbreaking achievement has sparked widespread discussion: will AI usher in a revolution of scientific incentive mechanisms?

Hassabis was ecstatic upon learning about his Nobel Prize in Chemistry. His wife received a barrage of calls from a Swedish number, repeatedly answering and hanging up until she realized the origin of the calls and forwarded the phone to Hassabis. In subsequent press conferences, Hassabis attended alongside John Jumper, a colleague at Google DeepMind.

The news came as no surprise, as just the day before, Hinton, along with Princeton University professor John Hopfield, received the Nobel Prize in Physics for their contributions to machine learning. Clearly, we have entered the era of artificial intelligence. Today, it is entirely possible to win a Nobel Prize by researching AI and applying it to other fields, as evidenced by Hinton and Hopfield in physics and Hassabis and Jumper (who shared the award with David Baker, a genomics scientist at the University of Washington) in chemistry.

"This is undoubtedly the 'AI moment in science,'" remarked Eleanor Drage, a senior research fellow at the Leverhulme Centre for Future Intelligence at the University of Cambridge. Her office colleagues playfully suggested that Elon Musk, founder of xAI, is now a hot contender for the Nobel Peace Prize. Drage believes that awarding the physics and chemistry prizes to AI researchers "has ignited a passionate debate, not only within these disciplines but also externally." She attributes this to two reasons: the pervasiveness of AI in academic research, blurring the lines between disciplines, and "our profound admiration for computer scientists, to the point where we're willing to classify them into any field."

While Drage harbors reservations about the significance of the Nobel Committee's decisions this week, she and others firmly believe that these decisions will have a profound impact on future research. Matt Hodgkinson, former research integrity manager at the UK Research Integrity Office, commented, "A trend of using AI to achieve Nobel Prizes may be emerging, potentially influencing research directions." However, the question remains: will this transformation propel us in the right direction?

David Baker, one of this year's Nobel Prize winners in Chemistry, has long been a leading figure in using AI to predict protein structures. He has dedicated decades to tackling this challenge, gradually making progress. Baker recognized that the well-defined problem and the regularity of protein structures made it an ideal testing ground for AI algorithms. His success is no coincidence; Baker has published over 600 academic papers throughout his career. Similarly, Google DeepMind's AlphaFold2 project stands as a testament to this.

However, Hodgkinson expresses concern that researchers, in trying to dissect the success of this year's three Nobel laureates, might overly focus on the technical aspects and neglect the scientific essence. He emphasizes, "I hope this doesn't mislead researchers into believing that all AI tools are equally valuable, leading to the misuse of tools like chatbots." This concern stems from the influence of excessive interest in other so-called "disruptive technologies." Hodgkinson points out, "Hype cycles always exist, the most recent examples being blockchain and graphene."

Data from Google Scholar, a research paper search engine, reveals that after the discovery of graphene in 2004, the number of academic papers mentioning the material soared from 45,000 between 2005 and 2009. However, following Andre Geim and Konstantin Novoselov's Nobel Prize win for discovering graphene, the number of related publications skyrocketed, reaching 454,000 between 2010 and 2014 and exceeding 1 million between 2015 and 2020. Yet, the impact of this research frenzy on the real world remains limited.

Hodgkinson believes that the inspiring force of multiple researchers being recognized by the Nobel Committee for their work in AI might entice more individuals to enter this field, potentially leading to changes in the quality of science. He adds, "Whether the proposals and applications of AI have substantial content is another matter to consider."

We have witnessed the profound impact of media and public attention on AI on academia. Research from Stanford University reveals that the number of papers published on AI tripled from 2010 to 2022, reaching nearly 250,000 papers in 2022 alone, equivalent to over 660 new papers every day. Moreover, this data predates the generative AI revolution spearheaded by ChatGPT in November 2022.

Julian Togelius, an associate professor of computer science at New York University's Tandon School of Engineering, who conducts AI research, is concerned about the extent to which scholars will be influenced by media attention, financial incentives, and Nobel Committee accolades. He notes, "Often, scientists choose the path of least resistance and greatest reward." Considering the competitive nature of academia, dwindling funding, and its direct correlation with researchers' career prospects, the allure of associating a hot topic (such as a field that potentially grants Nobel Prizes to high achievers, as in this week's case) with acquiring resources might prove irresistible.

The risk lies in the potential suppression of innovative thinking. As Togelius points out, "Obtaining more fundamental data from nature and formulating new theories humans can comprehend is an extremely challenging task." This demands deep thinking and exploration. For researchers, employing AI for simulations, supporting existing theories, and processing existing data, while offering incremental improvements rather than paradigm shifts in understanding, is far more efficient. Togelius predicts that a new generation of scientists might ultimately opt for this path, as it is simpler.

Furthermore, there exists another risk: overly confident computer scientists, while driving the development of AI, begin to witness AI's Nobel Prize wins in unrelated scientific fields (like physics and chemistry) and decide to follow suit, encroaching on research spaces in other disciplines. Togelius warns, "Computer scientists often recklessly apply algorithms to areas they don't understand and then label it progress. Whether it's good or bad, they consider it a given." He confesses that, due to his limited knowledge of fields like physics, biology, or geology, he has also succumbed to the urge to apply deep learning to other scientific domains and "advance" before deeper contemplation.

Hassabis serves as an outstanding example of using AI to advance science. He holds a doctorate in neuroscience, obtained in 2009, and has leveraged this expertise to significantly contribute to AI development at Google DeepMind. However, even he acknowledges a shift in how the industry is improving efficiency. At the Nobel Prize press conference, he mentioned, "AI is now more engineering-oriented. We have a lot of techniques, and we're mainly making improvements at the algorithmic level rather than directly referencing how the brain works."

This transformation has profound implications for the type of research conducted, its subject matter, the researchers' knowledge of the field, and their motivations for entering it. We might witness more computer scientists participating in research, rather than those who dedicate their lives to a particular specialized field, detached from the reality they study. However, this hasn't diminished the joy of Hassabis, Jumper, and their colleagues in winning the Nobel Prize. Hassabis earlier stated, "We're close to finishing the code cleanup for AlphaFold3, and we plan to make it freely available to the academic community. We'll continue to move forward from there."

Today, the influence of AI permeates every facet of science. The Nobel Prize awards undoubtedly inject new vitality into the application of AI in science, but they also raise questions about the future direction of scientific research. Will AI truly lead scientific development, or will it become a transitional "hype cycle"? This is a question that will require continued exploration and reflection in the future.