TLDR: This paper examines how AI impacts creativity in science, distinguishing between creative approaches and products. Through mathematical case studies, it argues that while AI can extend human disciplinary creativity (e.g., Four Color Theorem), certain AI-driven methods (e.g., Cap Set Problem with LLMs) can displace it by solving problems without relying on discipline-specific human expertise, potentially diminishing the value of scientific pursuit.
A recent paper explores the evolving relationship between artificial intelligence and human creativity in scientific problem-solving, particularly focusing on how AI might “decenter” disciplinary creativity. The author, Eamon Duede, delves into a philosophical tradition of worrying about new technologies displacing human capacities, drawing parallels from Socrates’ concerns about writing to modern anxieties about computation.
Historically, new technologies have often been seen as both extensions and amputations of human abilities. Writing, for instance, extended our capacity to store knowledge beyond individual memory but also potentially diminished the need for active recall and genuine wisdom. Similarly, the introduction of mechanical calculators raised concerns about a loss of “number sense” among mathematicians and physicists. Today, the increasing reliance on AI in science brings new epistemological concerns, such as scientists adopting beliefs not fully justifiable due to the opacity of complex AI models, and a potential limitation of scientific understanding.
The paper introduces a crucial distinction between “creative approaches” and “creative products.” While a product (like a scientific theory or a work of art) can be judged as creative, the focus here is on the approach taken to achieve it. Disciplinary creativity is defined as the creative application of discipline-specific expertise to a valued problem within that field. This means an approach must not only be creative in a general sense but also deeply rooted in the knowledge, methods, and norms of a particular scientific discipline.
Computation’s Dual Role: Extending and Displacing Creativity
The paper presents two cases from mathematics to illustrate how computation can both extend and displace disciplinary creativity. The first case examines the proof of the Four Color Theorem (4CT) by Kenneth Appel and Wolfgang Haken in 1976. This theorem states that any planar map can be colored with at most four colors such that no two adjacent regions share the same color. The proof involved a significant computational component, where a supercomputer verified the four-colorability of a finite set of configurations. Crucially, the mathematicians conceived the mathematically creative approach, and the computer acted as a mechanized tool to carry out a part of that approach. This was seen as an extension of human mathematical capacity, preserving the disciplinary creativity of the mathematicians.
In contrast, the second case involves a more recent breakthrough in the Cap Set Problem, a challenge in extremal combinatorics. Researchers used a multi-step algorithmic framework involving a large language model (LLM) like Codey (fine-tuned on general computer code, not specific mathematical texts) and an evaluator program. The LLM generated code for a “priority” function, which was then tested and refined recursively. This process led to the discovery of the largest known cap set for n=8. However, the paper argues that this approach, while creative in a general or engineering sense, displaces disciplinary creativity. The LLM was not guided by mathematical insights from human experts, nor did it apply combinatorics-specific techniques. Instead, it iteratively generated and tested code against a mathematical constraint, operating largely detached from the discipline’s background knowledge and normative practices.
Also Read:
- The Rise of Autonomous AI Scientists: A New Paradigm in Discovery
- AI-Generated Chess Puzzles: Experts Weigh In
Implications for Scientific Pursuit
The author concludes that while AI offers powerful new ways to scale up our collective problem-solving capacity, it also has the potential to radically alter, and perhaps diminish, the value of scientific pursuit. The framework used in the Cap Set Problem is general and can be applied to various mathematical problems, suggesting an approach that is “indifferent to the particularities of any given problem.” This detachment, the paper argues, not only pushes humans further from the center of the epistemological enterprise but also displaces disciplinary creativity and expertise from the core of our scientific problem-solving strategies.
For a deeper dive into this fascinating discussion, you can read the full research paper here: AI and the Decentering of Disciplinary Creativity.
