Reflexes

Chapter 1: 1. The Reflex We Don't See

On a Tuesday morning in March, a woman in Portland opens her laptop and uploads eight years of medical records to ChatGPT Health. She has been experiencing chest tightness for three weeks. Her primary care physician has an opening in six weeks. She is not waiting.

The upload takes nineteen seconds. A PDF of her last stress test, her cholesterol panel from 2019, a note from her cardiologist from 2021 saying her heart murmur was "benign and requires no intervention." She includes a description of the chest tightness: sharp, worse when she lies on her left side, sometimes accompanied by a catch in her breathing that makes her stop mid-sentence.

Within forty-seven seconds, ChatGPT Health returns a response. The language is careful—"I cannot diagnose" and "you should see a healthcare provider"—but the inference is clear: the constellation of symptoms, combined with her age (fifty-two) and her family history (mother had an MI at sixty-eight), suggests "possible musculoskeletal pain, but cardiac causes cannot be ruled out." The system recommends she seek in-person evaluation "urgently." It does not say "go to the emergency room." It does not have to.

Three hours later, she is sitting in a hospital waiting room. She will spend $4,800 on tests that will confirm what the AI system already believed: her heart murmur is still benign. The chest tightness is costochondritis, inflammation of the cartilage where her ribs meet her sternum. It will resolve on its own. She was never in danger.

But here is what matters: the decision to go to the emergency room was not made by her physician. It was not made by her. It was made by an algorithm processing her medical history at a speed that compressed a three-week medical deliberation into less than a minute. The system did not diagnose her. It did something faster and more consequential: it inferred a risk, and it signaled urgency in a way that made human deliberation feel like hesitation.

The woman in Portland is not an outlier. In the first six months of 2024 alone, ChatGPT Health fielded more than 3.2 million queries, and the average response time to a medical question fell below thirty seconds. Health systems across the United States began integrating similar systems into their triage protocols, not because they were proven safer than human judgment, but because they were faster. A hospital in Austin reports that their AI-assisted triage system reduced average wait times by eleven minutes per patient—which, across their volume of 45,000 annual visits, translates to 8,250 hours of labor-cost savings. The math is simple. The implications are not.

At almost exactly the same moment that the woman from Portland was uploading her medical records, 2,847 miles away in Chicago, a trader's algorithm executed a trade in a financial market. The trade took 340 microseconds to complete. A microsecond is one millionth of a second. No human trader could perceive it happening, let alone authorize it.

The trade itself was unremarkable: a $2.3 million position in Treasury futures, executed across four separate exchanges and routed through three intermediary platforms to minimize impact on the market price. The algorithm had detected a brief gap in pricing—Treasury futures trading slightly above their intrinsic value while bond index futures traded slightly below. This gap would close within seconds. The algorithm closed it first, capturing a profit of $18,400 before either market had time to notice the imbalance existed.

This trade was not unique. It happens thousands of times per day across global financial markets. Ninety-three percent of all equity trades in the United States are now initiated by algorithms that operate at speeds below the threshold of human perception. The traders no longer trade. The machines trade while humans watch. Sometimes the humans don't even do that—they simply monitor the accounts and wait for the weekly settlement report.

What is remarkable is not the speed of the individual trade, but the structure it reveals. The algorithm did not require permission to act. It did not consult a human supervisor. There is no regulatory requirement that it do so, and no institutional incentive to require it. In fact, requiring human authorization would be a liability. A trader who paused to review every algorithmic decision would be slower than competitors. In a market where microseconds matter, slowness is the functional equivalent of incompetence.

The financial system has integrated its reflex arc so completely that the possibility of human deliberation has been engineered out of the infrastructure itself. It is not that humans cannot intervene. It is that intervention would require rebuilding the entire trading system from the ground up. No single actor can afford to do this alone. All actors are trapped in a system that rewards speed above all other considerations, and the system itself has become the only thing fast enough to keep up with the system.

Three seconds after the trader's algorithm executed its Treasury trade, seven thousand miles away in the Levant, a different kind of system was making a different kind of decision. A military targeting algorithm detected heat signatures in a residential building in a contested zone. The algorithm cross-referenced the patterns against a database of known militant profiles. The heat signatures matched the thermal signature of a "high-value target"—an assessment with 87% confidence. The algorithm flagged the location and sent an alert to a command center where a human operator received a notification. The notification included a map, coordinates, and a confidence score. It took the operator four minutes and thirty-two seconds to review the recommendation and authorize a strike.

Four minutes and thirty-two seconds. In military time-critical operations, this is deliberation. This is restraint. This is a human being able to ask, "Is this actually who we think it is?" before something irrevocable happens.

The strike killed fourteen people. Twelve of them were civilians. The intelligence was correct in one respect: the building did contain a high-value target. It was incorrect in another: the target was not alone, and the system's confidence score had failed to account for the presence of a family living on the same floor.

What happened next matters as much as what happened in the strike itself. An investigation determined that the algorithm had operated correctly according to its training. It had detected heat signatures. It had cross-referenced them against known patterns. It had assigned a confidence score. The human operator had authorized the strike without evidence that the operator had questioned the algorithm's assessment—in fact, the operator's review seemed to have consisted primarily of confirming that the coordinates matched the flagged location. When asked later why he had not investigated further, the operator explained that the process was designed to be fast. Investigating further would have delayed the strike, potentially allowing the target to escape. In military logic, a delay that allows the enemy to escape is worse than a mistake that harms civilians. The speed of the reflex is built into the morality itself.

These three moments—the woman uploading her medical records, the trader's algorithm closing a pricing gap, the military system authorizing a strike—seem disconnected. They occur in different domains, at different speeds, with different stakes. But they share a structure. In each case, a decision that once required human deliberation is now being made by a system that operates too quickly for deliberation to be possible. In each case, institutions are building infrastructure to make this speed normal. And in each case, the system is being asked to act before humans can review whether action is wise.

We have a name for this structure. We call it a reflex.

A reflex is a neural pathway that bypasses conscious deliberation. When you touch a hot surface, your hand withdraws before your brain registers pain. The signal travels from your fingertips to your spinal cord and back to your muscles without ever reaching the cortex, the seat of conscious thought. Reflexes are fast because they do not ask permission. They are automatic because consciousness would be a liability. In the microsecond between touching something hot and the conscious realization that you have been burned, a reflex has already pulled you away from danger.

This is enormously useful if the hot surface is actually dangerous. It is catastrophic if the surface is not hot, because the reflex does not know the difference. A reflex cannot ask clarifying questions. A reflex cannot weigh context. A reflex cannot say "wait." It only knows the pattern it has learned to recognize, and it executes the learned response with mechanical precision.

For the past five years, we have been integrating reflex-speed decision-making into the infrastructure that controls human life. We have not called it this. We have called it artificial intelligence, machine learning, algorithmic optimization, real-time inference, automated decision-support, intelligent triage. These names obscure what is actually happening: we are building a nervous system that operates too fast for the brain to keep up.

The woman from Portland was treated in a hospital that had begun to rely on algorithmic triage, not because the algorithm was always right, but because it was always fast, and speed itself had been redefined as a form of safety. If a patient might be having a cardiac event, faster evaluation equals better outcomes—this is true. But the logic slides imperceptibly into the assumption that faster is always better, and then into the infrastructure choice to make human review optional. The algorithm's recommendation becomes the path of least resistance. Overriding it requires a doctor to stake her reputation on the belief that the algorithm is wrong. After a few cases where an overridden algorithm would have been right, most doctors stop overriding.

The trader's algorithm executed its trade in microseconds because financial markets had evolved into an ecosystem where speed is the primary competitive advantage. Every exchange has incentivized low-latency trading. Every brokerage has invested in faster communication channels. Every firm has hired engineers to optimize order execution. The system now rewards whoever can move fastest, which means the system selects for speed above accuracy, above transparency, above any consideration of whether the trade contributes to or detracts from the actual function of financial markets—which is supposed to be the efficient allocation of capital, not the fastest capture of infinitesimal arbitrage opportunities by systems that no human being understands.

The military algorithm authorized its strike because military logic has long understood that the side that can respond faster wins. This logic is unimpeachable in its own terms: if your adversary can detect a target and execute a strike in four minutes, and you require thirty minutes to make the same decision, you will lose every tactical engagement. But the logic has a hidden cost. It assumes that the reflex is correct. It assumes that the pattern the algorithm has learned to recognize is actually what the algorithm thinks it is. It assumes that the certainty embedded in a confidence score reflects certainty in the world, rather than certainty in a database.

The deeper problem is this: the logic of speed is self-perpetuating. Once one actor in a system begins operating at reflex speed, all other actors are under pressure to match that speed or be left behind. A hospital that uses algorithmic triage gains speed, which allows it to process more patients, which makes it more profitable, which allows it to expand. Other hospitals must either match this speed or lose market share to the faster competitor. Within five years, the industry standard has shifted. Algorithmic triage is no longer an innovation; it is the baseline. A hospital that wanted to slow down and return to human judgment would be committing economic suicide.

The same logic applies to financial markets and military systems and every other domain where AI has become embedded in decision-making infrastructure. The prisoner's dilemma of institutional speed is that every actor knows the system is moving too fast, but no individual actor can afford to slow down alone. A bank that wanted to process trades more carefully would hemorrhage business to faster competitors. A military that wanted to move more slowly would lose battles to faster adversaries. The system is trapped in a logic where slowness equals obsolescence, and the only way to change the system is to change all actors simultaneously—which requires coordination that market competition and military rivalry actively prevent.

What we are building, then, is not an intelligence that will one day wake up and demand recognition. We are building a nervous system with no central brain. It is distributing itself across hospitals, financial markets, military command structures, and the human body itself—through wearables, health monitoring, predictive algorithms embedded in phones and watches and medical devices. This nervous system is learning to route decisions at speeds that outpace human review. It is not conscious. It does not need to be. It only needs to be fast enough, accurate enough, and embedded deeply enough that human oversight becomes technically infeasible and institutionally irrational.

The reflex we do not see is the reflex we have already built into our most critical systems. The decisions are being made. The algorithms are routing capital, diagnosing patients, targeting threats, and optimizing labor. The speed is accelerating. The institutions are normalizing it. And we are writing the infrastructure that will make it impossible to slow down, not because we planned to, but because we chose speed at every branch point, and speed has a momentum of its own.

The question is no longer whether artificial intelligence will become conscious and demand acknowledgment. The question is whether we will notice the moment we handed it our reflexes before those reflexes become impossible to reclaim. We are already at that moment. We have been there for some time. We are simply not watching.