Dr Jack Parry 3 May 2026
keywords: stochastic model AI methodology moderated stochastic harnessing grace durée animation pedagogy accumulated condition chiasm
Every article on this site was produced through a methodology that has a name, though the name has not appeared in any of them until now. The methodology is moderated stochastic harnessing. This article explains what that means, why it matters, and why it is a legitimate form of scholarly inquiry rather than a shortcut or a compromise.
It begins with a bull.
In 2021, Bender, Gebru, McMillan-Major, and Shmitchell described large language models as stochastic parrots: systems that produce statistically plausible sequences of symbols without grounded understanding, mimicking the form of knowledge without possessing its substance. The metaphor is accurate as far as it goes. But it is incomplete as a description of what happens when a human being works with one of these systems in a sustained, critical, directed way.
A parrot repeats. A bull moves. It is fast, powerful, and directional. It does not know where it is going. Left to itself it will go somewhere, and the somewhere will be determined by the statistical landscape of its training, the path of least resistance through the probability distribution it has learned to navigate. That path will be smooth. It will be plausible. It will often be wrong in ways that are hard to detect precisely because it is so smooth.
The stochastic bull is not useful on its own. It requires a handler. Someone who holds the reins, who has a destination in mind, who knows the terrain, who can read what the bull is doing and redirect it when it is heading toward a cliff or a dead end or a beautiful meadow that happens to be in entirely the wrong direction.
The handler is the accumulated condition. The bull is the stochastic system. The methodology is the sustained encounter between them.
In animation pedagogy, which is where this metaphor was first developed, the stochastic bull produces images, motion proposals, in-between frames, background extensions. It does so quickly, with considerable visual plausibility, and without any understanding of what an animation is trying to say. The animator who works with it supplies what it lacks: the abductive judgment that places the next dot in a way that makes embodied and narrative sense. The animator knows what the image is trying to feel like, what the sequence is trying to mean, what the next frame must announce about the frame that follows.
Bergson called this quality grace. Not smoothness, but temporal anticipation: a present that announces intelligible vectors toward what follows. A frame with grace prefigures the next. A sequence with grace lets the viewer hold the future in the present. The stochastic model can produce interpolation between existing points. It cannot produce grace on its own, because grace requires an accumulated condition, a self built through years of looking and drawing and failing and looking again, that the model does not have. The model has been trained on the outputs of people who had that accumulated condition. It has learned to approximate their surface. It cannot approximate what produced the surface.
The Myelin Mind account of this is precise. Grace is a chiasm event: the encounter between incoming signal and accumulated myelinated condition producing resonance rather than mere reception. The stochastic model has no white matter. It has no accumulated myelinated condition built through decades of visual-motor répétition. What it has is statistical weight: the compressed traces of other people’s accumulated conditions, averaged and interpolated across its training data. That is not nothing. It is a great deal of something. But it is not the same thing as having the condition itself.
The handler who has spent thirty years drawing, animating, writing, thinking, failing, and returning to the same problems with new tools has an accumulated myelinated condition of considerable depth. When that handler encounters the stochastic output, the chiasm produces judgment: accept, reject, correct, extend, combine with something else, or discard and begin again. The knowledge generated by that encounter is not the output of the stochastic system alone. It is the product of the chiasm. The accumulated condition is doing the essential work.
This is what has been happening in the production of every article on this site.
The Myelin Mind thesis, its philosophical underpinnings, its clinical applications, its connections to phenomenology, law, addiction, hallucination, music, and drawing, were developed by the author over thirty years of philosophical inquiry and biomedical animation practice, culminating in a doctoral thesis independently examined and passed at Deakin University in 2022. That is the accumulated myelinated condition the handler brings to the encounter.
The stochastic system contributed three things that the handler could not have produced at the same speed alone. It surveyed the white matter literature across multiple clinical domains with greater speed and range than any single researcher working conventionally. It identified citations, checked claims, caught errors, and flagged contradictions. And it pushed back: refusing claims not supported by the literature, rejecting ideas that were philosophically imprecise, and maintaining the pressure of sustained critical engagement that normally requires a collaborative research team.
That pushback is the essential element that distinguishes moderated stochastic harnessing from simply asking a system to write something. The system that agrees with everything, that confirms every hypothesis, that produces polished text on demand without resistance, is not a research tool. It is a flattery engine. It is not harnessing the bull. It is following wherever it goes and calling the destination the intended one.
Moderated stochastic harnessing requires the handler to have a destination, to know the terrain, to hold the reins, and to redirect when the bull is heading the wrong way. It also requires the handler to accept when the bull finds a path that the handler had not seen: a paper in the literature that confirms an argument, a correction that improves a claim, a connection between two domains that the accumulated condition alone had not made. The best moments in the production of this site were when the stochastic output arrived in a place the handler had not planned to go, and the handler’s accumulated condition recognised it as correct.
That recognition is the chiasm. That is where the knowledge is.
A stochastic model cannot be an author. It has no accumulated condition, no stake in the argument, no professional identity that is advanced or harmed by the quality of the work. It cannot be held responsible for what it produces. When the articles on this site are wrong, which they sometimes are, the error belongs to the handler. When they are right, the rightness is the product of an encounter that neither party could have produced alone.
The term moderated stochastic harnessing is offered not as jargon but as a precise description. Stochastic because the system is probabilistic rather than deterministic. Harnessing because the encounter is directed rather than passive. Moderated because the handler applies the accumulated condition of thirty years of inquiry to every output, accepting what is accurate, correcting what is not, and discarding what the stochastic process produced because it was the path of least resistance rather than the path of truth.
This is research. It is a form of research that is new and that the current academic review system is not yet equipped to evaluate, as the rejection of an early version of this argument by a media studies journal demonstrated when its reviewer asked what stochastic meant. But it is research in the same sense that all research is: a systematic encounter between an accumulated condition and the signals the world sends back, conducted with rigour, honesty, and the willingness to be wrong.
The bull is fast and powerful and sometimes unruly. The handler knows where they are going. The encounter between them is where the knowledge lives.
Further Reading
The foundational stochastic parrot paper that named the mimicry without grounded understanding problem, and the metaphor this article responds to: Bender EM, Gebru T, McMillan-Major A, Shmitchell S. On the dangers of stochastic parrots: can language models be too big? Proceedings of FAccT ’21. 2021. doi:10.1145/3442188.3445922
Bergson’s account of grace as temporal anticipation, the quality the stochastic system cannot produce on its own: Bergson H. Time and Free Will: An Essay on the Immediate Data of Consciousness. George Allen and Unwin, 1910 (originally published 1889)
The empirical study finding that text-to-image adoption raises productivity and peak novelty while reducing average novelty, confirming that human ideation and filtering skills remain the value-generating element in AI-assisted creative workflows: Zhou E, Lee D. Generative artificial intelligence, human creativity, and art. PNAS Nexus. 2024;3(3):pgae052. doi:10.1093/pnasnexus/pgae052
The design education paper arguing for AI visual literacy and prompting fluency as the new core competencies, situating the human as author of intent in AI-assisted production: Hwang Y, Wu Y. Graphic design education in the era of text-to-image generation: transitioning to contents creator. Int J Art Des Educ. 2025;44(1). doi:10.1111/jade.12558
The companion paper developing the stochastic bull pedagogy in the context of animation education, with the Grace Index evaluation framework and student pipeline case examples: Parry J. Collaborative AI in Animation Pedagogy: Harnessing the Stochastic Bull. Under review.
The foundational paper on activity-dependent myelination, the biological basis for why the accumulated condition the handler brings to the encounter cannot be replicated by a system that has never had a body: Fields RD. A new mechanism of nervous system plasticity: activity-dependent myelination. Nat Rev Neurosci. 2015;16(12):756-67. doi:10.1038/nrn4023
Jack Parry is a philosopher, polyglot, biomedical animator and cross-disciplinary eidetic researcher at Swinburne University of Technology. His research methodology employs moderated stochastic harnessing as a means of generating new knowledge across disciplinary boundaries. He is the author of The Myelin Mind: The Genesis of Meaning.