Theory of Knowledge Essay

“To what extent do you agree with the claim “all models are wrong, but some are useful” (attributed to

George Box)? Discuss with reference to mathematics and one other area of knowledge.”

AOKS: Math, Natural Sciences

May 2025

Word Count: 1476

are simplified representations of complex concepts, using assumptions and limitations to make them more

understandable or applicable, serving as tools for acquiring knowledge. These models help us clench a slight grasp

on large and complex topics, building the foundation for our ongoing pursuit of knowledge through various ways

of knowing. However, the use of assumptions and limitations leads to the paradox emphasized in the quote by

George Box, “all models are wrong, but some are useful.” In this essay, I will evaluate the claim by exploring

different models in mathematics and the natural sciences, examining their epistemological value and limitations on

the pursuit of knowledge. This evaluation reveals a key knowledge issue: How do our chosen models, shaped by

our ways of knowing and the assumptions of our knowledge frameworks, influence what we consider “truth”? It is

important to note that Box’s quote was said under the context of statistical models, and “usefulness” is ambiguous,

where it must be evaluated and objectively quantified across different contexts, like math and natural science.

My stance on the claim is based on my interpretations of several key terms, and prior to preceding, we

must define them, exploring how they function within different knowledge frameworks. By using the term “all,”

Box generalizes every single model, regardless of individual purpose or content, while “wrong” refers to flaws,

limitations, and oversimplifications within any knowledge framework. By using the term “some,” the statement

suggests that only some models have practical value, where others might not. This generalization underscores the

context-sensitive nature of knowledge, teaching us to value models for their usefulness while being aware of their

limitations. “Useful” could refer to solving problems or furthering understanding of the world. These terms

highlight the limitations of perceptions or language, where conceptual frameworks may constrain our

understanding of the world, which emphasizes the subjectivity of this interpretive investigation.

I will delve into two opposing perspectives on Box’s bold statement: one perspective that agrees to a high

extent of the statement, all models are wrong but useful, and another perspective that disagrees with the statement,

believing that models are not wrong. Mathematics and natural sciences are fundamental regarding models, as their

methods and conclusions strive for objectivity. I will argue that while models have inherent limitations in

real-world contexts, they prove to be successful in guiding the creation of further knowledge.

Mathematical models have some clear examples of models that might not be considered “wrong” but rather

contextually limited. One key example is Euclidean geometry, developed by the ancient Greek mathematician
Euclid (Artmann). This system is based on 5 suggested rules, and, for centuries, continued to be the definitive

model for understanding geometry. This powerful model perfectly describes the geometry of perfectly flat surfaces

and 2D spaces, and each law has proven to be true. This high degree of consistency could suggest that the model is

correct. However, the contextual limitation of Euclidean geometry is shown when we consider that its laws only

hold in perfectly flat, idealized spaces, and not in more complex, non-Euclidean contexts. Despite these

limitations, Euclidean geometry remains useful, especially in fields like architecture and engineering. In this sense,

Euclidean geometry is “correct” in its application to strictly Euclidean spaces, but in the broad context of 3D

reality, it is “wrong” but still useful. Therefore, Euclidean geometry is an important example of how a model can

both be “correct” and “wrong”, especially through different contexts.

Probability is a mathematical model that represents the randomness in the world and can be seen as an

incorrect but useful model. While some may learn probability as a simplistic model for randomness, some

frameworks, like Bayesian probability, demonstrate the sheer complexities of probability. (Hayes) Nonetheless,

even these models still rely on idealized assumptions that can lead to inaccuracies in real-world applications. This

model would be “wrong” because it relies on assumptions that can not calculate every single variable in a

situation. One strong example of this is a dice roll, where we assume that the system is ideal and that the outcomes

are purely random with the chance of rolling each side being 1/6. However, in reality, many factors come into play,

like the air, the angle they were tossed at, the slight irregularities in the dice, or even environmental conditions that

may all sway the results and introduce measurable biases, as mentioned by Einstein. (Hayes) This evidence

supports the claim that the model’s assumptions are idealized and do not fully capture real-world variability.

Although these factors seem to influence the outcome minimally, they still prove that the probability models

incorporate many assumptions that may not hold universally, hence introducing slight imperfections. This example

underscores that our knowledge of uncertainty and risk is mediated by models that gain authority through their

predictive success, despite their limitations. That being said, this model is still useful, it provides a strong

framework for making near-perfect predictions on uncertainty, and allows us to understand risks and problems.

Thus, probability is a strong example of a model being “wrong” but useful as well.

Shifting to the area of knowledge of natural sciences, one powerful example of a model that is not

necessarily “wrong” is the double-helix DNA model, proposed by Watson and Crick in 1953 (Hernandez). This
model suggests that the structure of DNA is two strands of nucleotides connected by hydrogen bonds between

nitrogenous bases, taking the shape of a double helix. Experimental findings, the repeated empirical observations

and data, have not contradicted the model and it has proven to be crucial in advancing our understanding of

biology. This consistency with experimental proof, despite being a simplification, suggests that this model is

“correct.” However, this model simplifies some more complex molecular processes and does not capture the full

picture. These details are more extensions of the basic models and do not contradict it, and act as extensions to the

model. For example, many proven processes, like DNA replication, have stayed consistent with the double-helix

structure. In a simplified context, the model stands as correct, but has limitations when seen through a real world

lens. Thus, the double-helix DNA model is “correct” in the sense that it has proven to be objectively accurate, just

with several limitations that make the model unable to address the entirety of the topic.

A contrasting model, the Bohr model of atomic structure, developed by Niels Bohr in 1913, is an example

of a model that is “wrong” but useful for building foundational knowledge. This model suggests that electrons

orbit the nucleus of an atom in perfect orbits and that these orbits correspond to specific energy levels. This model

was significant because it gave an explanation of the lines observed and gave a clear and simple answer to atomic

structure. However, the model is “wrong” in that it is not accurate according to quantum mechanics. It does not

explain the wave-like nature of electrons and contradicts Schodinger’s equation that suggests the electrons do not

follow fixed orbits but just exist in spaces called orbitals (Promeet). The model is still useful, however, since it

provides a simpler introduction to the complex topic of atomic structure, letting people grasp the fundamental

ideas about atoms. It acted as a stepping stone to a more comprehensive framework, showing that our certainty

about knowledge and understanding of “truth” is provisional and evolves as we gain new evidence. The authority

of classical models is always challenged as we acquire new evidence and adopt more nuanced frameworks of

understanding. Thus, while this model is objectively incorrect, it helps lay a foundation for understanding.

​ In conclusion, I agree to a large extent with the claim “all models are wrong, but some are useful.”

Within both AOKs, the models have proved to be “wrong” but “useful,” where the inherent limitations of models

underscore that our knowledge is both provisional and context-dependent. On one hand, examples such as

Euclidean geometry as a mathematical model and the DNA double-helix structure as a natural science model

provide authoritative frameworks that appear “correct” in those specific contexts, offering a high degree of
certainty in certain applications. However, these models also reveal the constraints imposed by our chosen

knowledge frameworks. On the other hand, the examples of probability as a mathematical model and the Bohr

model of atomic structure as a natural science model are clear examples that illustrate that even when models are

“wrong” in a literal sense, they remain important for developing future knowledge, highlighting the strengths and

weaknesses inherent in all models and encouraging us to continuously refine our understandings as new evidence

emerges. Although some models might be seen as entirely correct within limited domains, we can appreciate that

the usefulness of a model depends on its ability to create more inquiry and adapt to evolving contexts of

knowledge.
 Bibliography

Artmann, Benno. “Euclidean Geometry.” Encyclopædia Britannica, Encyclopædia Britannica, inc., 18 Feb. 2025, ​
​ www.britannica.com/science/Euclidean-geometry.

Hayes, Adam. “Bayes’ Theorem: What It Is, Formula, and Examples.” Investopedia, Investopedia, 23 Feb. 2025, ​
​ www.investopedia.com/terms/b/bayes-theorem.asp.

Hayes, Adam. “The Ineffable Purpose of Randomness.” John Templeton Foundation, 23 Feb. 2025, ​ ​
​ www.templeton.org/news/the-ineffable-purpose-of-randomness#:~:text=Einstein%20believed%20in​
​ %20%E2%80%9Cperfect%20laws,Bohr%20and%20Werner%20Heisenberg%20disagreed.

Hernandez, Victoria. “‘Genetical Implications of the Structure of Deoxyribonucleic Acid’ (1953), by James ​
​ Watson and Francis Crick.” Embryo Project Encyclopedia, 13 Jan. 2020, ​
​ embryo.asu.edu/pages/genetical-implications-structure-deoxyribonucleic-acid-1953-james-watson-an​
​ d-francis-crick#:~:text=In%20April%201953%2C%20before%20Watson,scientific%20verification​
​ 20that%20DNA%20contained.

Promeet, Dutta. “Schrödinger Equation.” Encyclopædia Britannica, Encyclopædia Britannica, inc., 14 Jan. 2025, ​
​ www.britannica.com/science/Schrodinger-equation.