The fundamental paradox of psychotherapy—and indeed of human development itself—concerns how sufficient continuity of self can be conserved across radical transformation. When does someone change within the scope of who they are, and at what point do they become a different person, if ever? These questions are the bread and butter of being a reflective, evolving person.

I’ve been mulling over this question recently through an unlikely lens: artificial intelligence research.

A Mathematical Hull for Containing Multitudes

On New Year’s Day 2026, the AI company DeepSeek released a technical paper on something called “manifold-constrained hyper-connections”—an architecture designed to stabilize the training of massively interconnected large language models. Their solution constrained network weight matrices to lie on what mathematicians call a Birkhoff polytope. This immediately struck me, at least as an analogy, as being very familiar with my sense of how the mind might work.

The technical language is worth a slow take. A polytope is essentially a geometric shape with flat sides existing in any number of dimensions—in 2D it’s a polygon, in 3D it’s a polyhedron, like a cube. The Birkhoff polytope has a special property: It’s “doubly stochastic,” meaning all the values in each row and column must sum to one. Think of it like a Sudoku puzzle where everything has to add to 1, called “normalization.”

Why does this matter? This constraint creates what’s called a “convex hull”—a bounded container that keeps everything inside stable, even when what is inside gets pretty wild. The system can redistribute its internal patterns without runaway amplification or collapse. Infinite complexity, quenched within defined limits.

I often poetically think of our sense of self as the hull we need to stay afloat on the sea of life. This mathematical structure maps remarkably onto what we actually experience in clinical work.

Containing Multitudes Within Complex Geometries

What DeepSeek built for engineering stability in AI resembles a constraint we’ve long recognized in psychotherapy. Wilfred Bion called it “containment.” Donald Winnicott called it a “holding environment.” Without these, transformation tends toward either fragmentation or rigidification—chaos or stagnation.

I once asked my psychoanalyst during training what I was supposed to get out of therapy. He said simply, “Yourself,” a parsimonious response that has stuck with me—precise yet open-ended. We enter depth-oriented psychotherapy with identity unclear or fragmented, rigidified by defensive patterns or destabilized through developmental transitions. We seek profound reorganization of psychological structure. Yet we must emerge recognizably ourselves—not a different person, but the same person differently organized.

The doubly stochastic algorithm provides an active geometrical envelope that continuously normalizes divergence, reflecting back within the system to preserve its own integrity. It’s the best of both worlds: the power of massive connectivity and the foundation of long-range identity.

The Ship of Theseus and Caterpillar-to-Butterfly Metamorphosis

This leads naturally to Plutarch’s Ship of Theseus—the philosophical paradox wherein if each plank on a ship is replaced one-by-one over time until no original pieces remain, is it still the same ship?

The polytope framework suggests an answer: identity isn’t located in any particular plank but in the geometric constraints that govern how planks relate to each other. Change any component, but if the structure maintains its doubly stochastic property—if each transformation preserves the overall balance—continuity is maintained. It adds up to one.

The caterpillar dissolving into cellular soup before reorganizing into a butterfly has long served as transformation’s most dramatic metaphor—yet we rarely ask what preserves continuity through that apparent annihilation. Remarkably, research from Georgetown University demonstrated that moths retain associative memories formed as caterpillars, even after most tissues have liquefied and reorganized. Not everything survives—memories formed too early in larval development are lost—but key neural structures persist through the dissolution, carrying learned experience into a radically different body. The polytope framework offers a way to think about this: Even when visible structure liquefies, certain geometric constraints governing how components relate remain intact, enabling radical reorganization without total loss of identity.

In clinical terms, this partially explains why insight alone rarely produces lasting change. Recognizing a pattern doesn’t automatically reorganize the underlying structure. The therapeutic relationship, in this framing, functions as a normalizing algorithm—continuously helping the patient’s internal dynamics stay within generative bounds while profound reorganization occurs.

Free Will Within Geometric Constraints

This framework offers an intriguing perspective on agency. Free will, if it exists, may represent choosing direction of movement within geometric constraints at any given moment. Many possibilities exist—some that might appear to undermine identity—while the formal structure continuously preserves or renormalizes who we are. The more things change, the more they stay the same.

What we call the “thickness” of the present moment might reflect polytope dimensionality—how many simultaneous relationships our system can hold in working memory, how rich the possibility space appears from our current position. This is where causality operates, where past meets future, where self meets world.

Clinical Implications

For clinical practice, this framework suggests that premature termination risks leaving patients in unstable configurations. Integration requires completing sufficient “iterations”—working through the material enough times for new patterns to stabilize within identity-preserving constraints.

Combining modalities—therapy plus neuromodulation, for instance—may accelerate this process if interventions are properly timed and coordinated. Understanding the geometric structure suggests when flexibility is therapeutic (exploring the interior) versus when boundaries are essential (maintaining manifold constraints).

A Caveat and a Musing

This framework is entirely speculative, without empirical validation. The mathematical formalism should be considered a compelling metaphor rather than a literal description. Real brains are noisy systems that may only approximate these dynamics.

Yet there’s something remarkable here. What DeepSeek built to stabilize AI training, evolution may have built to enable consciousness. Perhaps there’s no accident that human minds conjured powerful math for AI, given that our sense of mathematics may derive from the brain itself being our core contact with physical reality, via self-interaction. Perhaps the brain is already doing something like this, and AI applications are naturally neuromorphic.

Whether the Birkhoff polytope is literal description or powerful metaphor remains to be determined. Either way, it provides a framework for discussing transformations that preserve identity—perhaps the central challenge of therapeutic work, human development, and conscious experience itself.

The therapeutic relationship, in this framing, functions as a normalizing algorithm—continuously helping the patient’s internal dynamics stay within generative bounds while profound reorganization occurs. (For a fuller treatment of therapeutic mechanism, see below the integrated tripartite framework combining Personalized Self-State Mapping, Active Inference Therapy, and Experiential Field Theory, Brenner, 2024.)