Creativity in Machine Learning

A framework for AI systems that can change the rules of the game

New WHITE PAPER

There is growing evidence that the use of AI tools is associated with a narrowing of the diversity of scientific inquiry[1] [2]. This effect may largely reflect changing research incentives. For example, researchers may increasingly focus on problems where AI methods are most effective, such as areas with abundant data. That explanation is plausible, but this paper addresses a different and more far-reaching question about AI: whether artificial intelligence systems in general, and large language models (LLMs) in particular, possess the creative capabilities required to generate genuinely new directions of inquiry or to challenge the assumptions and framings that define existing problems. This question extends beyond scientific discovery. If such limitations exist, they may also constrain the use of AI in other complex domains such as business, policy, and technological innovation, where progress often depends on questioning existing frameworks rather than finding new solutions within them.

To investigate this possibility, we focus in this paper not on diversity of outputs alone but on the creative capabilities that generate new approaches and problem framings. We draw on Margaret Boden’s distinction between three forms of creativity: combinational, exploratory, and transformational[3]. We will define these categories more fully in the next section. In summary, we suggest that AI models can recombine existing ideas and explore deeply possible solutions to a defined problem. However, without further capabilities, they struggle with transformational creativity—the ability to revise the problem formulation itself.

Understanding why this limitation arises is one aim of this paper. By analysing the training objectives and optimisation dynamics of modern AI systems, we identify structural reasons why transformational creativity does not naturally emerge. Secondly, we then suggest approaches that may introduce the conditions required for transformational creativity.

If you’re interested in reading the full white paper, please see the file below:

[1] Hao, Q., Xu, F., Li, Y. et al. Artificial intelligence tools expand scientists’ impact but contract science’s focus. Nature 649, 1237–1243 (2026). https://doi.org/10.1038/s41586-025-09922-y

[2]Traberg, C.S., Roozenbeek, J. & van der Linden, S. AI is turning research into a scientific monoculture. Commun Psychol 4, 37 (2026). https://doi.org/10.1038/s44271-026-00428-5

[3] Boden, M. A. (2014). Creativity and artificial intelligence. In E. S. Paul & S. B. Kaufman (Eds.), The Philosophy of Creativity). Oxford University Press.

Edited and adapted by: Marika Niihori