What AI Can—and Cannot—Teach You About Pose Method® of Running

Artificial intelligence, or AI, has changed how people find, process, and produce information across a wide range of everyday tasks. It has made an enormous amount of information immediately accessible. It has also brought certain realities into the light that many simply had not noticed before.

This article is finally relevant enough to cover several pertinent questions. Let’s begin with the most obvious one—punctuation in writing. When AI-generated text started appearing everywhere like a flood of strange dashes, line breaks, emphasis, and suddenly all text looked unusually coherent, strangely unfamiliar, and somehow more polished while using words many people did not know existed or could be used in such a way—well, that revealed several things at once. One was that most people use only a fraction of the writing tools available to them, while common ways of communicating ideas and exchanging information have narrowed enough that perfectly normal devices of written language can suddenly look suspicious.

So while professional writers—people who write for a living across a variety of topics and industries, not just book authors or journalists—have always used em-dashes, line breaks, emphasis, and other structural devices, AI suddenly turned some of those things into an “immediate giveaway” that an article was AI slop. Because when does a human use an em-dash? Well, we do. We use it when we want to clarify something, emphasize a thought, interrupt ourselves, or simply use something other than another comma or pair of parentheses. I can even tell you how to type one on your keyboard because, unfortunately, it requires more than one key.

So, to get to the point—this article is written by a human who writes for a living. Those em-dashes in the title are mine, not AI. Just because a social-media “expert” classified them as AI slop does not make them so. What AI actually exposed here is a problem with recognition. People can mistake unfamiliarity for artificiality, just as they can mistake a fluent AI explanation for actual understanding.

And this brings me to the next point.

An AI system can summarize research, explain terminology, compare methods, suggest drills, and produce what appears to be a coherent explanation of how running works, but you thinking for yourself is still required. It is enormously useful to have AI scan, aggregate, compare, and organize information, but that usefulness creates a new problem as well.

Access to information is not the same thing as access to a system of knowledge, and neither one guarantees useful understanding.

AI Works With What It Can Find

Large language models generate answers from patterns found across enormous amounts of information. They do not operate with what we know as human thinking and understanding; they operate through learned relationships, embedded structure, probability, and published information.

Pose Method® of Running is a defined system developed by sports scientist Dr. Nicholas Romanov in 1977 because no such system for running existed at the time. It has its own model of running, terminology, teaching sequence, technical criteria, drills, and method of observation and correction. The information Dr. Romanov used to develop it did not appear out of nowhere. Much of the relevant scientific knowledge and research already existed. What did not exist was the connection between those pieces. Nobody had assembled those already existing dots into a coherent model, systemized that knowledge, and made it useful, teachable, and learnable as a method of running.

Pose Method has now existed for nearly fifty years. During that time it has appeared in books, scientific papers, courses, articles, interviews, coaching materials, military training, academic discussions, online forums, videos, criticism, summaries, and countless second-hand explanations. But it is not the governing framework around which general-purpose AI models organize their knowledge of running. They have been exposed to a vastly larger body of material built around conventional biomechanical and coaching frameworks, and when no explicit framework is supplied, those dominant patterns tend to pull the answer back toward them.

And even within the information specifically about Pose Method, the sources are not equivalent. Some describe the actual method. Some describe only a portion of it. Some study selected variables associated with it. Some reinterpret it through a different biomechanical framework. Some criticize claims Pose Method never made. Others borrow individual concepts or drills while placing them inside an entirely different model of running.

An AI system can encounter all of those sources at once. Unless it has a reliable way to distinguish what belongs to the system, what represents observation rather than instruction, what is someone else’s interpretation, and what directly contradicts the system, it may combine them. The result can sound perfectly reasonable while no longer describing Pose Method at all.

And this can happen even after the correct framework has been provided. An AI system may accurately explain Pose Method in one paragraph and quietly reintroduce a push-off, knee drive, or another incompatible concept in the next because those relationships are more strongly represented in the broader information on which it learned. In other words, the AI may know the pieces and still organize them under the wrong model.

AI does not automatically know a system simply because information about that system exists online. It can construct a framework from the information available to it, but it has no independent basis for knowing whether that reconstruction is actually correct. Give it the governing model and it can work within it much more effectively. Leave it to reconstruct that model from decades of primary sources, secondary sources, research papers, criticism, borrowed terminology, and internet commentary, and you should expect some strange offspring.

This is one reason AI platforms carry some version of the same warning near the typing box: AI can make mistakes. The people building these systems understand this limitation very well. A fluent answer can still be an incorrect synthesis.

Research Creates Another Layer of Confusion

Available scientific research does not eliminate this problem. In some ways, it adds a hefty new layer of confusion to it. Very few people outside of research fields know what conducting research actually involves, what it requires, what its limitations are, or what the results do and do not allow us to conclude. So when you ask a question based on existing research and AI produces a coherent-looking synthesis complete with references and scientific terminology, it can become remarkably difficult to know what, exactly, you should be questioning.

A study may investigate one feature associated with a method without testing the entire method. Researchers may use their own terminology, select particular variables, modify an intervention, or measure outcomes that answer a much narrower question than the one a reader assumes was answered. They may not be proficient in the method they are testing. There may be only one participant involved for perfectly legitimate reasons, but that automatically limits what can be concluded from the findings. Researchers may also be unaware of earlier work in the same direction, or of research in another field that would materially change how their results could be interpreted.

None of this makes research bad. These are largely the constraints that make scientific research useful in the first place: define the question, control what can reasonably be controlled, measure something specific, and limit the conclusions to what the evidence actually supports. But those same constraints also mean that research does not interpret itself. The researchers’ knowledge, experience, assumptions, familiarity with the subject, and the way the study itself is constructed can all influence what is examined and how the findings are understood. That becomes particularly important in human movement and biomechanics, where biological variability, skill, environment, and the way an intervention is actually performed can all matter.

Those studies can therefore be extremely valuable without becoming something they were never designed to be. A study of one component of Pose Method does not become a definition of the complete method. An observation about successful runners does not automatically establish the cause of their success, nor does it by itself provide adequate support for a theory explaining why they run faster.

This is where AI can make the situation even more interesting. AI is very good at finding relationships among published statements. It is much less reliable at recognizing when those statements belong to different levels of explanation. A measured association in one paper, an observed characteristic in another, a proposed mechanism in a third, and a coaching instruction found somewhere else can all be assembled into one beautifully coherent paragraph even though the original sources never established the chain of reasoning that now seems to connect them.

Research, like AI, is a tool. Data does not explain itself, and accumulating more data does not automatically produce a coherent understanding of what the data means. Without an adequate framework for interpreting it, research can tell us a great deal about what was observed, measured, or changed under particular conditions while leaving the much larger question of why it happened unresolved.

That distinction becomes especially important when studying movement. A measured characteristic, a coaching instruction, a physical principle, a research outcome, and a technical rule are not the same kind of claim. They cannot simply be stacked together and treated as if, by accumulation, they have become an explanation.

Use AI as a Tool, Not as the Source of the Method

There is no reason to avoid AI when studying running. Used properly, it is an extraordinary research and learning tool.

Ask it to explain a term. Ask it to locate research. Ask it to compare two biomechanical arguments. Ask it to summarize a paper. Ask it to challenge an assumption. Ask it to help trace where an idea originated. Use it to search faster, compare more material, and ask questions you might not otherwise have thought to ask.

But knowing the description is not knowing the skill, and when the question is specifically, “What is the Pose Method?”, the standard remains Pose Method itself. The problem is that the error will not always look like an obvious factual mistake. An answer may contain several individually recognizable ideas, all neatly mashed together into what looks like a legitimate package, and you don’t know what you don’t know.

So until AI platforms are explicitly trained to work within the Pose Method framework, protect yourself. Because now you know.

The same principle applies to any established technical system. AI can organize information about a system. It can discuss the system. It can help you study the system. But you should not rely on it to reconstruct that system from fragments and then assume the reconstruction is authoritative. Direct it to the primary source, give it the governing framework, or go to the source yourself. If your interest is your own running, the Pose Method® course for individual runners provides a structured way to work directly from the method rather than trying to reconstruct it from pieces found online.

That distinction will become more important, not less, as AI becomes part of everyday education. The more capable AI becomes at producing convincing explanations, the more important it becomes to know what those explanations are being measured against.