The Evolving Definition of Discovery in the Age of AI
In 1974, the Nobel Prize in Physics was awarded to Antony Hewish for his groundbreaking work on the discovery of pulsars. However, the initial detection of the first pulsar signal was made by his graduate student, Jocelyn Bell Burnell, who also played a crucial role in constructing the telescope and analyzing the resulting data. Despite her significant contributions, Bell Burnell was not recognized with the prestigious award, a decision rooted in the prevailing view that she was fulfilling the expected role of a student under the direction of a senior researcher. This historical instance raises a fundamental question: when a pivotal insight arises not from the principal investigator’s direct efforts, but from the work of others, who rightfully deserves recognition for the discovery?
This question gains renewed relevance with the rapid advancement of artificial intelligence (AI). Consider a scenario where an AI system independently generates a solution to a complex scientific problem. If human scientists subsequently verify the accuracy of this solution, who should be credited with the breakthrough? The debate extends beyond the Nobel Prizes themselves, encompassing accolades like the Abel Prize, Wolf Prizes, and Lasker Awards – all recognizing exceptional scientific achievement.
OpenAI’s GPT-5.2 and the New Landscape of Scientific Contribution
The emergence of sophisticated AI models is prompting a re-evaluation of traditional notions of authorship and discovery. On February 13th, OpenAI announced that its latest model, GPT-5.2, assisted a team of scientists in achieving “a new result in theoretical physics.” The process involved scientists formulating the initial research question, with GPT-5.2 proposing a potential solution. Subsequently, an internal OpenAI model refined and ultimately delivered the complete solution, which was then validated by the human researchers.
This collaborative process mirrors the historical case of Andrew Wiles’s proof of Fermat’s Last Theorem, where computer verification was used to confirm the mathematical validity of his work. In that instance, the computer’s role was viewed solely as a verification tool, not a co-author. However, when an AI generates a proof that surpasses human reconstructability, the lines of contribution become blurred, potentially reducing human scientists to the role of curators rather than genuine collaborators. True discovery, it is argued, necessitates understanding, not merely validation.
The Problem of Arbitrary Boundaries in Assigning Credit
If understanding is the key criterion, then the award should go to those who possess it. This leads to recognizing the scientists who provided the necessary constraints, conducted sanity checks, and formulated the conceptual framework that rendered the AI-generated solution mathematically coherent. This approach seems logical, yet it implicitly acknowledges a distinction between the intellectual labor driving the solution and the infrastructure enabling it.
This raises a parallel to the original Hewish/Bell Burnell case: why was only Hewish honored, excluding the technicians who built the radio receiver or the engineers who mitigated atmospheric interference? The justification lies in the belief that these contributions were merely necessary conditions, not the discovery itself. The crucial act was recognizing the signal’s anomaly – an intellectual feat distinct from the engineering involved in building the telescope. However, this logic quickly unravels when considering the theoretical physicists of the 1930s who predicted the existence of neutron stars. Without their foundational work, Hewish and Bell Burnell might not have known what they were observing, prompting the question of whether they, too, deserved recognition.
Recognizing the Collective Nature of Scientific Progress
Ultimately, the determination of who constitutes a “discoverer” is inherently arbitrary. We inevitably draw a line, designating certain individuals as primary contributors while relegating others to the background. This line is often influenced by factors such as proximity to the final stage of research, institutional affiliation, national origin, and established scientific hierarchies. Labor that is distant in time, space, or social standing is frequently overlooked as merely a prerequisite for the discovery.
With AI’s increasing involvement, this arbitrariness becomes undeniable, as the normally invisible labor embedded within the model’s workflow is brought to light. The development of these AI systems relies on the contributions of countless machine learning researchers, data annotators, and organizations with the resources to support large-scale model training. The Nobel Prizes, while valuable as signals of achievement, are ultimately a reflection of a constructed narrative, rather than a perfect representation of reality. Science is a collaborative endeavor, built upon the collective efforts of vast networks spanning generations and continents. Recognizing this inherent complexity is crucial as we navigate the evolving landscape of scientific discovery in the age of AI.
