The Best Way to Prepare for the Unknown Is to Practice Near It

May 21, 2026

9 min read

2 views

91%

The strange problem with getting better

What if the fastest way to become more resilient is not to train harder on the thing you already know, but to practice receiving surprise?

That question sits at the heart of two ideas that seem unrelated at first glance. One comes from machine learning, where a model is deliberately exposed to synthetic, self-generated variations so it can adapt more smoothly when reality shifts. The other comes from elite performance training, where Navy SEAL candidates were taught to shrink overwhelming missions into short goals, rehearse success mentally, talk themselves through fear, and regulate their breathing.

Both are really about the same hidden problem: when the future changes, raw skill is not enough. A system, whether a neural network or a human nervous system, can fail not because it lacks capability, but because it has not been prepared for a new pattern of conditions. The challenge is not simply learning more. It is learning to stay functional while the world stops behaving as expected.

That is a deeper and more interesting problem than conventional improvement. Most training assumes the goal is to maximize performance under a fixed set of rules. But the world rarely rewards that kind of narrow optimization for long. Markets shift, teams reorganize, technologies mutate, anxiety spikes, and the situation that once made you excellent becomes obsolete. The real question is not, “How do I do this better?” It is, “How do I stay adaptive when the game changes?”

Training for the adjacent possible

A useful mental model comes from the idea of the adjacent possible: the next set of states that are not random, but also not fully familiar. In machine learning terms, this means exposing a model not only to its existing data distribution, but to slightly displaced versions of it, so the model does not become brittle. The network effectively practices living one step beyond the world it has already mastered.

That sounds abstract until you translate it into human terms. Imagine learning to swim only in a calm pool, then being dropped into open water with waves, currents, and cold shock. You might know the strokes, but you have not learned how to operate under environmental disturbance. Now imagine a more intelligent training process. You still practice in the pool, but you also rehearse in choppy conditions, you learn to manage your breathing under stress, and you build the habit of recovering from disorientation. That is training in the adjacent possible.

This is where the Navy SEAL techniques become more than motivational advice. Short goal setting narrows attention to the next controllable unit of action. Mental rehearsal preloads the mind with a successful trajectory. Self talk gives language a regulatory function, interrupting panic with instruction. Arousal control keeps physiological noise from hijacking execution. Each technique is a way of stabilizing performance not by removing uncertainty, but by keeping the system within a range where adaptation remains possible.

The strongest performers are not those who avoid surprise. They are those whose minds and bodies are organized to stay usable under surprise.

That is the bridge between the two domains. The machine learning method and the SEAL training methods both aim to create a kind of preparedness elasticity. They do not merely optimize outcomes inside a known landscape. They expand the range of conditions under which the learner can still function.

Why ordinary training fails when the world shifts

Traditional training tends to assume that improvement is mostly about reducing error on current examples. That works well when the future looks like the past. But the moment conditions change, yesterday’s accuracy can become tomorrow’s fragility.

Think about a chess player who has memorized openings without understanding the structure of the board. Or a manager who has mastered one company's culture but cannot operate in another. Or a student who can solve textbook problems but freezes when a question is phrased differently. In each case, the problem is not insufficient repetition. It is overfitting to the familiar.

Overfitting is a technical term in machine learning, but it is also a beautiful description of human vulnerability. We overfit to stable routines, predictable feedback, and familiar emotional states. Then, when novelty arrives, we do not merely lack a solution. We lose composure. The system is not just wrong. It is disoriented.

This is why the SEAL methods matter so much. Goal setting reduces the psychological size of the problem. Instead of “survive the whole ordeal,” the recruit thinks, “make it to lunch.” That is not a motivational trick. It is a state management strategy. The brain performs better when uncertainty is chunked into actionable units.

Mental rehearsal serves a different function. It creates a kind of internal simulation, allowing the nervous system to encounter the shape of the challenge before the body has to face it. In effect, the person is borrowing from the future. That is conceptually close to the learning algorithm that lets a model generate synthetic sequences and train on them. The learner is not waiting passively for the next event. It is rehearsing probable futures.

Self talk and arousal control complete the loop. One governs meaning, the other governs physiology. Fear becomes especially destructive when the body interprets a challenge as threat, because then cognition shrinks. Breathing, phrasing, cadence, and inner instruction are not cosmetic. They determine whether the learner remains in a state where information can still be processed.

In short, ordinary training asks, “How do I perform?” Better training asks, “How do I remain trainable while under pressure?” That is a much more valuable question.

The hidden similarity between a neural network and a human mind

It is tempting to think of artificial learning and human performance as separate worlds. But both are essentially systems that must learn from imperfect signals while avoiding collapse under novelty.

A neural network exposed only to one narrow statistical regime becomes brittle. A person who only rehearses ideal conditions becomes brittle too. The structure of the problem is the same: the learner must balance two forces that pull in opposite directions. One force is stability, the preservation of what already works. The other is plasticity, the ability to incorporate what is new.

Too much stability, and you cannot adapt. Too much plasticity, and you lose coherence. Growth lives in the tension between the two.

Here is a useful framework: every robust learner needs three layers.

Core competence: the stable skills that should remain intact.
Stress adaptation: the mechanisms that preserve performance under pressure.
Novelty absorption: the capacity to integrate unfamiliar patterns without breaking.

The SEAL techniques are mostly about the second layer. The machine learning method is mostly about the third. But both depend on the first. Without core competence, rehearsal is empty. Without stress adaptation, competence collapses in the wild. Without novelty absorption, a system becomes a museum of past success.

This tri-layer model explains why some people seem brilliant in theory but inconsistent in life. They have competence, maybe even high competence, but not enough adaptation infrastructure. When conditions drift, they do not know how to downshift, reset, or widen their internal bandwidth. They confuse excellence with invulnerability.

Real resilience looks less like hardness and more like controlled permeability. The mind stays open enough to incorporate change, but not so open that it loses structure. That is exactly what synthetic exposure and psychological regulation are trying to preserve.

How to practice near the edge without falling off it

The most practical insight here is also the most neglected: you do not need to simulate catastrophe to prepare for it. In fact, doing so too aggressively can backfire. The goal is not to flood the system. The goal is to approach novelty in digestible increments.

That is the logic of the adjacent possible. The next challenge should be unfamiliar enough to require adaptation, but familiar enough that the learner can still organize around it. If the leap is too small, nothing changes. If the leap is too large, the system shuts down.

You can apply this principle in many domains:

A writer can practice by changing genre, constraint, or form instead of waiting for inspiration to appear in a vacuum.
A manager can rehearse difficult conversations at low stakes before having them in real life.
A student can solve problems in slightly altered formats, not just repeat canonical examples.
An athlete can train under light fatigue, noise, or timing pressure to preserve technique under disturbance.
A founder can run scenario planning for plausible disruptions, not just optimize for the current market.

The point is not exposure for its own sake. It is calibrated exposure. The learner visits the borderlands of competence often enough that novelty stops feeling like a cliff.

This is also where language matters. Saying “I am overwhelmed” often freezes action. Saying “What is the next small move?” changes the problem from global to local. The recruit who can breathe, shorten the horizon, and speak constructively to himself is not doing positive thinking. He is preserving function. Likewise, a model trained on a broader neighborhood of possible inputs is not becoming mystical. It is becoming more robust.

Preparedness is not the absence of surprise. It is the ability to keep learning while surprise is happening.

That sentence may sound obvious, but most training systems ignore it. They optimize for correctness at rest, not adaptability in motion.

Key Takeaways

Do not train only for the average case. Practice slightly altered conditions so your skills remain usable when reality shifts.
Shrink the time horizon under stress. Replace huge goals with the next visible milestone, because composure improves when the brain has a concrete target.
Use rehearsal to borrow certainty from the future. Mentally simulate success before the moment arrives, so the body has a script to follow.
Treat self talk as a control system. The words you repeat under pressure influence whether fear becomes signal or noise.
Build a novelty buffer. Regularly operate one step outside your comfort zone, not so far that you break, but far enough that you expand.

The real lesson: resilience is a shape, not a trait

We tend to speak about resilience as if it were a personality characteristic, something you either have or do not have. But the deeper view is more interesting. Resilience is not a trait. It is a relationship between a system and change.

Some systems are brittle because they are too specialized. Others are chaotic because they never consolidate. The best systems have learned how to remain coherent while still welcoming disturbance. They do not simply resist disruption. They absorb just enough of it to evolve.

That is what makes the connection between these two ideas so powerful. Whether you are designing an algorithm or preparing a human being for extreme conditions, the mission is similar: create a learner that can visit the edge of the unknown without losing itself there.

So the next time you train, do not ask only whether you are getting better at the task. Ask whether your method is making you more capable of meeting the unexpected. The future rarely arrives in the exact form you rehearsed. What matters is whether your system has practiced enough adjacent possibilities to stay awake when it does.

The highest form of preparation is not prediction. It is organized adaptability. And once you see that, improvement stops being about mastering a fixed world. It becomes about learning how to remain alive to change.