What is it about?
For over a century, mathematicians have used a famous thought experiment to explain how order can emerge from randomness: given enough time, a monkey randomly hitting typewriter keys would eventually type the complete works of Shakespeare. The conclusion is that if you wait long enough, structure will appear on its own, simply through trial and error. This paper challenges that conclusion. The author argues that real-world systems, whether artificial intelligence, biological organisms, or human communication, do not behave like infinite random processes. They are bounded, recursive, and limited. Order does not emerge from endless possibility; it emerges from the systematic elimination of what cannot work. Drawing on the work of cyberneticist Gregory Bateson, the study reframes structure as the product of constraint, not the residue of chance. The argument has real consequences. When AI systems are trained repeatedly on their own outputs, they collapse into repetitive, low-diversity behavior, a phenomenon known as model collapse. The paper shows this is not a technical glitch. It is the predictable outcome of removing the constraints that keep any system meaningful. Meaning, the author concludes, is not what remains after everything is tried. It is what survives when most possibilities are excluded.
Featured Image
Photo by Igor Omilaev on Unsplash
Why is it important?
This study is the first to systematically critique the epistemological foundations of the Infinite Monkey Theorem using the cybernetic logic of Gregory Bateson, and to apply that critique directly to contemporary artificial intelligence. That pairing has not been attempted before. The work is timely because probabilistic and scaling-based assumptions dominate current AI research. The paper offers a formal, theoretical counterweight: a Constraint-Driven Model of Intelligence in which bounded, recursive, selective processes, not unbounded randomness, are the source of structured meaning. This reframing provides a principled explanation for why recursive AI training collapses, why scaling alone does not guarantee intelligence, and why constraints, long treated as obstacles, are actually generative. The framework is transdisciplinary. It connects probability theory, information theory, cybernetics, computability, and bounded rationality into a single coherent account of how structure, diversity, and meaning emerge in finite systems. For researchers in AI, cognitive science, philosophy of information, and complexity science, it offers a unifying lens. For practitioners, it offers design principles: if you want intelligent, diverse, and stable outputs from AI, you need to engineer, preserve, and curate constraints rather than remove them.
Perspectives
I am a Distinguished Professor of AI Ethics and Cybernetics, a licensed clinician, and a programmer, and I do not treat any of these as secondary. The Fifth Industrial Revolution is not a future to wait for. It is a demand on scholars working right now to integrate human-centered practice, formal systems thinking, and computational modeling in a single working epistemology rather than carrying them as separate identities/committments. That is the position from which this paper is written. The Constraint-Driven Model of Intelligence is not a metaphor. It is a synthesis produced by someone who has spent decades operating inside both psychotherapy and digital technology, and who has watched the same structural truth surface in each: meaningful, stable, adaptive outcomes arise in systems whose limits are well chosen, applied recursively, and revisited by an observer capable of adjusting them. Borel's monkey is the modern field's favorite thought experiment, and it has been quietly updated for the scaling era: throw enough randomness (and money, and compute, and data, and likewise) at a problem, and structure will eventually appear. The math is elegant. The intuition is appealing. The conclusion is wrong. For any real-world system that is bounded, recursive, and self-consuming, which is exactly the regime modern AI now operates in. Gregory Bateson saw this in advance. Shannon, Simon, and Turing supplied the formal machinery. The clinical and ethical traditions I work in have been saying it differently for a long time: structure is what survives when noise is excluded, and the observer is always inside the constraint system doing the excluding. What was missing was a unification that took both halves seriously at once. This paper is an attempt at that unification. I hope it also serves as a small instance of the 5IR imperative itself: a clinician-ethicist-cyberneticist-computational scholar producing a formal result that none of those four identities could have produced alone.
Assoc. Prof. Ezra N. S. Lockhart
National University
Read the Original
This page is a summary of: Constraint, Asymmetry, and Meaning: A Cybernetic Reinterpretation of Probabilistic Emergence Across Complex Systems, Symmetry, March 2026, MDPI AG,
DOI: 10.3390/sym18030518.
You can read the full text:
Contributors
The following have contributed to this page
