In the 14th century, William of Ockham, a medieval friar and philosopher, proposed a principle that has influenced scientific thought for centuries: Ockham’s Razor. The idea was simple—pick the simplest explanation. This principle, often referred to as the parsimony principle, suggests that when faced with competing hypotheses or models, the one with the fewest assumptions should be preferred. It’s a concept that has shaped how scientists approach problem-solving, particularly when developing theoretical models.
For many years, parsimony—the notion that simpler models are more likely to be accurate—was a guiding principle in scientific inquiry. It served as an elegant heuristic for building theories and conducting research. However, the recent rise of complex, data-driven artificial intelligence (AI) models has begun to challenge this long-standing assumption. Take, for instance, groundbreaking technologies like AlphaFold, which has revolutionized the field of biology by predicting protein structures with high accuracy, or ChatGPT, which generates humanlike text responses. These AI systems, which thrive on vast amounts of data and complexity, have outperformed simpler, more traditional models in numerous domains.
A recent paper published in Proceedings of the National Academy of Sciences (PNAS) argues that by adhering too rigidly to Ockham’s Razor, scientists may be missing out on valuable insights and making mistakes in their models. The authors suggest that complexity may not be the enemy of good science but, rather, a tool that, when used appropriately, can lead to more accurate and nuanced understandings of the world.
Rethinking Parsimony
Marina Dubova, the first author of the study and a Postdoctoral Fellow in Complexity at the Santa Fe Institute (SFI), is at the forefront of this challenge to the parsimony-driven approach. Dubova contends that the tendency to rely on simplicity in scientific modeling has historical roots. She argues that parsimony was once adopted as an easy-to-use tool, but it has not been scrutinized enough in recent years.
In an interview, Dubova explains, “Scientists need a tool to guide how they build models of the world. Parsimony was historically adopted as an easy tool to use. Since then, it’s not been questioned enough. Educational programs continue to teach parsimony as a key principle in scientific theory and model-building, but the justifications for its dominance haven’t stood the test of time.”
Indeed, Dubova’s own research challenges the very foundation of parsimony in modeling. In a recent computational simulation, she demonstrated that randomly selected experiments, unbound by the assumptions typically used to guide scientific inquiry, resulted in better models than those developed using the traditional, parsimony-driven approach. This revelation calls into question whether scientific progress has been unnecessarily constrained by an over-reliance on simplicity.
The Paradox of Simplicity vs. Complexity
Despite its historical significance, parsimony—while effective in some contexts—may limit our ability to understand the complexities of the world. Dubova’s work suggests that by focusing too much on simple models, scientists may overlook important variables and patterns that could improve predictions. In fact, parsimony may introduce bias into models, leading to misleading conclusions and poor predictions.
For example, in neuroscience, simple models used to interpret live brain scans often detect periodic patterns of brain activity, when, in reality, the brain’s activity is changing gradually over time. These models, which rely on simplified assumptions about brain function, fail to account for the full complexity of neural processes. Similarly, in pharmacology, leaving out essential characteristics—such as a patient’s age, genetic background, or pre-existing health conditions—from drug models could lead to inaccurate predictions about which individuals will respond well to new treatments.
On the other hand, complex models can be more flexible, accounting for a wider range of factors and interactions that simple models may miss. This flexibility is particularly evident in areas such as climate change research, where models often incorporate a vast array of variables, from atmospheric dynamics to ocean currents. Interestingly, recent findings suggest that complex, ensemble-based models—which combine multiple different models—are far better at making accurate climate predictions than relying on any single model. These models might contradict each other in some respects, but by pooling their insights, they offer a richer and more reliable understanding of climate patterns.
The Power of Ensemble Modeling
One area where complexity has proven particularly effective is in the study of climate change. Climate scientists have long struggled with the inherent complexities of the Earth’s climate system, which involves a multitude of interacting factors. To model climate behavior, researchers often develop their own individual models, each with different assumptions and structures.
At first glance, these models might appear incompatible with one another, as each captures different aspects of the climate system. However, recent research has shown that combining these diverse models into an ensemble—a group of models working together—results in better predictions. Even though the individual models may differ in their assumptions, their combined output can lead to more accurate forecasting of real-world phenomena.
Dubova explains this phenomenon: “Even when these climate models are incompatible, scientists decide to employ them all because they know each one is capturing some aspect of the world. The literature suggests that using them together helps us better predict the reality around us.” In the case of climate forecasting, models may represent different perspectives or simplifications of the complex system, but when they are combined, they provide a fuller and more reliable picture.
This approach could inspire new ways of thinking about scientific problems, where diverse models contribute to a more comprehensive understanding, rather than forcing researchers to adopt a singular, simplistic explanation.
Challenging the Status Quo
Dubova’s research underscores the importance of a more nuanced approach to model-building—one that recognizes the value of both simplicity and complexity. In her view, parsimony and complexity should not be viewed as opposing forces but as complementary tools. Scientists must carefully weigh the evidence, the context of their research, and the specific demands of their problem to determine when simplicity or complexity is appropriate.
As Dubova explains, “Relying on parsimony alone as our guiding principle limits what we can learn about the world and potentially drives us in wrong directions.” The key, she argues, is to maintain a flexible approach to scientific modeling, allowing for the incorporation of complex models when the situation demands it.
By challenging the traditional emphasis on parsimony, Dubova hopes to open new avenues of research into when complexity is beneficial and when simplicity should be favored. Her goal is to prompt a reassessment of how scientists approach model-building and encourage more open-mindedness toward complex solutions.
The Way Forward
As the scientific community continues to grapple with increasingly sophisticated challenges, the old adage of Ockham’s Razor may no longer be enough. Advances in AI, machine learning, and computational modeling have made it clear that simplicity does not always lead to the best solutions. Complex models, when used appropriately, can offer richer insights, better predictions, and more robust understandings of the world.
Dubova’s work is a timely reminder that science is a dynamic field, and our methods for exploring the unknown should evolve with the times. While simplicity will always have its place, scientists must also embrace complexity when it offers a more accurate and nuanced explanation. By doing so, they can better navigate the intricacies of the natural world, ensuring that their models—whether simple or complex—reflect the reality they seek to understand.
Ultimately, the challenge lies not in choosing between simplicity and complexity but in knowing when and how to use both. Dubova’s research encourages scientists to go beyond the limits of parsimony and explore the vast potential that complex models have to offer.
Reference: Marina Dubova et al, Is Ockham’s razor losing its edge? New perspectives on the principle of model parsimony, Proceedings of the National Academy of Sciences (2025). DOI: 10.1073/pnas.2401230121