## Level 1: The Conventional View Doubt is generally regarded as an essential tool for the pursuit of knowledge in the western rationalist view, which advocates for the scientific method of proposing hypotheses and experimentation (_The Stanford Encyclopedia of Philosophy_). Knowledge is acquired through doubt -- questioning hypotheses through experimentation and testing. Doubt is central because it's fundamentally what pushes a claim from a unverified hypothesis from a truth or knowledge. Popper's falsification theorem illustrate this clearly. Popper believes that the purpose of scientific evidence collection is not to verify theories, but to refute them -- to eliminate them from contention (Popper, 2010). A scientist's job is to carefully examine all possible theories and rule out as many as possible, or as Popper put it, to "falsify" them. As a result, doubt is at the foundation of this scientific method that has been popular since Enlightenment and the Newtonian times. ## Level 2: The Hidden Foundation However, I would like to argue that doubt is important to the scientific method -- but it is not central. Meaningful doubt is dependent on un-doubted foundations. Meaningful doubt itself must rest upon some foundation of undoubted knowledge. First is what I call constructive naivete. Mathematical progress often springs not from skepticism, but from the courage _not_ to doubt. For example, in the AOK of Mathemtaics, the five postulates of Euclidean geometry are not doubtable -- they are statements assumed to be true, so that everything else -- the entire subject -- can be built upon them (“Euclid’s Five Axioms: The Simple Rules That Built All of Geometry,” 2025). Here, doubting has now point becuase what they actually need was imaginative courage—the audacity to say, "Let us assume this might hold and see what happens." Furthermore, logic -- the absolute foundation framework of scientific thinking -- itself should stand beyond doubt. If we doubt logic itself, nothing meaningful can be derived, and knowledge can come from nowhere. Progress happens through deductive certainty. Here, logic and construtive naivete is the foundational paradigm in which the system of science stands. Yes, doubt is important within the paradigm, but what people ignore is that it is based on the faithful belief in logic and science itself -- believing that if we trust logic and trust some foundational assumptions, we can build whatever we what. The ultimate insight is that doubt is important but only within a specific paradigm that trusts logic and assumptions, with *belief* in the paradigm itself. Doubt is important but not central. ## Level 3: An Alternative Epistemology Further, let's question our previous claim -- is doubt always important? Again, I would like to point that up until now, we have been thinking in terms of the western rationalist paradigm, where doubt plays an important role; but what if we jump out of the paradigm itself and see other possibilities for the pursuit of knowledge? Are we universalizing Enlightenment rationalist-reductionist-mechanistic-cartesian framework as the only legitimate knowledge framework? Is it possible to acquire knowledge without doubt? The recent rise in AI arises my interest in how LLM "learn knowledge". LLM, or more commonly chatbots like ChatGPT, demonstrate great knowledge -- it can answer almost questions from any domain level of knowledge with a good level of accuracy (despite ocassional hallucinations) and demonstrate a fairly good understanding of knowledge -- but how does it learn these all? Diving into how an LLM is trained reveals to me that doubt is almost non-existent in how an LLM pursues knowledge. Consider AlphaFold, DeepMind's AI system that revolutionized our understanding of protein structures in the AOK of Biology. For over fifty years, biologists approached protein folding through the traditional scientific method -- proposing hypotheses about how amino acids interact, doubting and testing each mechanism, refining theories through experimental validation. Progress was painstakingly slow (Jumper _et al._, 2021). However, AlphaFold took an entirely different path. It absorbed patterns from over 170,000 known protein structures through training on big data, learning to recognize deep correlations in this vast web of relationships. Crucially, it developed no explicit rules about why proteins fold the way humans do. There were no hypotheses to doubt, no proposed mechanisms to falsify (Jumper _et al._, 2021). Within eighteen months, it solved what had been considered one of biology's grand challenges and discovered hundreds of new Protein strucutres (_AlphaFold: a solution to a 50-year-old grand challenge in biology_, 2022). The speed of human's progress on this subbject is supercharged. Notably, what AI learned and discovered is not knowledge acquired through doubt, but through _immersion_ and _pattern recognition_ (Jumper _et al._, 2021). The AI doesn't question whether a pattern is true; it absorbs the statistical regularities of reality itself. There is no Popperian falsification here, no skeptical examination of competing theories. In fact it doesn't even know what previous theories or explanations are. In LLM's pursuit of knwoledge, reality is a web of connections of data. Correlation matter more than causation. Explanations are powerless (Anderson, 2008). This forms a stark contrast from how humans understand the world: we are better at understanding causal chains (because of X so Y) by breaking things down to their components (reductionism) and looking for explanations (Shipper, 2025). | Huamn | AI | | ------------------------------------------------ | ---------------------------------------------------------- | | Breaking things down and understanding the parts | Understanding a web of connections and their relatinoships | | Causation (looking for explanation) | Correlation (no need for explanation) | We have been universalizing the Enlightenment rationalist framework —with its emphasis on reductionism, mechanism, and Cartesian doubt. We learn by breaking reality into simpler parts, questioning how they work indivdually (doubt), and try to put them back together (Shipper, 2025). This is our most powerful way to pursue knowledge up until now -- throuth the power of doubt, as explained in Level 1. The new worldview shown by LLM admits that certain things can’t be explained or put into words, let alone be doubted. Instead of asking "why" questions and trying to break things apart to explain, an LLM accepts the complexity as it is and try to understand the holistic relationships within the complexity, and end up giving powerful results without explanations. Up to now, we can conclude that doubt's centrality is fundamentally _paradigm-dependent_. Doubt is central to knowledge that humans are good at -- the one that operates through explicit propositions and hypothesis-testing, but not to knowledge in complex domains gained through pattern recognition (in which out method of doubt and rationalist thinking struggle). In domains amenable to formalization—classical physics, formal logic, controlled experimental science—doubt-driven refinement remains powerful. But in domains of irreducible complexity—protein folding, language, consciousness, markets, ecosystems—an alternative epistemology proves more effective, one based on pattern-trust and immersive engagement rather than skeptical questioning. ## Level 4: On the Verge of a Paradigm Shift? However, the next question is whether it is truly a "pick the paradigm based on the domain" question. What if instead of being just a smart substitute to the rationalist western scientific paradigm, what LLM demonstrates -- an emergent-relatinoal-context-based epistemology is a better alternative or even where the next paradigm shift might happen, just as how humans shift from feudal superstition toward scientific inquiry and doubt. For centuries, the rationalist-reductionist paradigm has driven extraordinary progress, but it is now hitting fundamental limits. The greatest challenges facing humanity—climate systems, consciousness, economic stability, pandemic response—are irreducibly complex, resistant to the reductionist methods that conquered classical physics (Mitchell, 2009). These are emergent phenomena where the whole cannot be understood by doubting and testing isolated parts like what Cartesian did. This is when AI points toward a necessary paradigm shift: from rationalist reduction to emergent, relational, pattern-based knowing. Complexity science already recognizes this—systems must be understood through their relationships and contexts, not decomposed into testable propositions (Mitchell, 2009). If humanity is to navigate the challenges ahead, we cannot rely solely on doubt-driven analysis. We must learn to _see_ patterns in irreducible complexity, to trust them, to develop knowledge through immersion rather than skeptical distance, to understand them without doubting. Doubt served us well in the age of steam engines and mechanism. But the age of modern complexity might demand a new epistemology -- and we are already seeing it through AI's pursuit of knowledge ## Sources _AlphaFold: a solution to a 50-year-old grand challenge in biology_ (2022) _Google DeepMind_. Available at: [https://deepmind.google/blog/alphafold-a-solution-to-a-50-year-old-grand-challenge-in-biology/](https://deepmind.google/blog/alphafold-a-solution-to-a-50-year-old-grand-challenge-in-biology/) (Accessed: December 8, 2025). Anderson, C. (2008) “The End of Theory: The Data Deluge Makes the Scientific Method Obsolete,” _Wired_, 23 June. Available at: [https://www.wired.com/2008/06/pb-theory/](https://www.wired.com/2008/06/pb-theory/) (Accessed: December 8, 2025). “Euclid’s Five Axioms: The Simple Rules That Built All of Geometry” (2025) _Kronecker Wallis_, 11 February. Available at: [https://www.kroneckerwallis.com/euclids-five-axioms-the-simple-rules-that-built-all-of-geometry/](https://www.kroneckerwallis.com/euclids-five-axioms-the-simple-rules-that-built-all-of-geometry/) (Accessed: December 8, 2025). Hepburn, Brian and Hanne Andersen, "Scientific Method", _The Stanford Encyclopedia of Philosophy_ (Summer 2021 Edition), Edward N. Zalta (ed.), Available at: <https://plato.stanford.edu/archives/sum2021/entries/scientific-method/>. Hepburn, B. and Andersen, H. (2015) “Scientific Method.” Available at: [https://plato.stanford.edu/entries/scientific-method/?utm_source=chatgpt.com](https://plato.stanford.edu/entries/scientific-method/?utm_source=chatgpt.com) (Accessed: December 8, 2025). Jumper, J. _et al._ (2021) “Highly accurate protein structure prediction with AlphaFold,” _Nature_, 596(7873), pp. 583–589. Available at: [https://doi.org/10.1038/s41586-021-03819-2](https://doi.org/10.1038/s41586-021-03819-2). Mitchell, M. (2009) _Complexity: a guided tour_. Oxford [England] ; New York: Oxford University Press. Popper, K.R. (2010) _The logic of scientific discovery_. Special Indian Edition. London: Routledge. Shipper, D. (2025) _Seeing Like a Language Model_. Available at: [https://every.to/chain-of-thought/seeing-like-a-language-model](https://every.to/chain-of-thought/seeing-like-a-language-model) (Accessed: December 8, 2025). Stryker, C. (2021) _What Are Large Language Models (LLMs)?_ Available at: [https://www.ibm.com/think/topics/large-language-models](https://www.ibm.com/think/topics/large-language-models) (Accessed: December 8, 2025).