AI in the Exam Room Patient
PATIENT · VELOCIRAPTOR TEST

The velociraptor test

Trusting your evolutionary wisdom over the algorithm.

All resources
"Until AI has to wrestle a velociraptor for dinner or protect its kids from a saber-toothed tiger, it will never have the contextual awareness evolution gave us."

Your body was debugged by 3.8 billion years of evolution. An AI's pattern recognition was debugged by training on text. When your velociraptor brain says something is wrong, listen to it.

The test

Ask: "Would I notice this while running from a predator?" Yes: probably a significant signal. No: might be minor and can wait.

Would notice (survival-critical)

  • Can't catch my breath
  • Chest feels like someone is sitting on it
  • Vision going dark
  • Can't move my arm
  • Room spinning, about to vomit
  • Extreme pain
  • My child is limp and unresponsive

Wouldn't notice (may still need care)

  • Minor headache
  • Slight muscle soreness
  • Small rash
  • Occasional heartburn
  • Mild fatigue

Your ten billion sensors

You have roughly ten billion sensory neurons sampling the world constantly. An AI has zero; its only input is the text you type. It cannot see you, hear you, smell you, touch you, or measure a single vital sign.

Visual: ~126 million photoreceptors per eye, reading color, movement, subtle skin and facial changes. Olfactory: ~400 receptor types that can pick up ketoacidosis or infection. Tactile: millions of receptors for temperature, pain, pressure, and where your body is in space. Auditory: ~15,000 hair cells per ear catching breathing changes and voice tremor. Vestibular: balance and orientation. Interoceptive: the internal sensors behind "something doesn't feel right."

The match-up
Ten billion sensors against zero. Who do you think is better at detecting a threat?

Evolutionary debugging

Humans were debugged by 3.8 billion years of lethal selection: the ancestors with accurate threat detection survived and passed it on; the ones who assumed "it's probably nothing" about real dangers did not. You are descended from the cautious. The AI was debugged by training loss on internet text, with no survival pressure and no consequences. It has never had to run from anything.

Parental instinct is the strongest version

Parents evolved extreme sensitivity to threats to their children. "Something's not right with my child" often picks up subtle behavior, breathing, color, energy, or feeding changes that are too early or too subtle to name. Maternal concern is a documented predictor of serious illness, sometimes outperforming clinical scoring. Trust it.

When AI conflicts with your gut

You have symptoms. The AI says "probably nothing." Your gut says "something is really wrong." Trust your gut, every time. The AI is guessing statistically from your text and population averages; your body is integrating real-time signals from ten billion sensors and patterns you cannot fully describe.

Real example
"I have chest discomfort, is this serious?" AI: "Many causes, often anxiety or reflux, try relaxation and antacids." Gut: "This feels different, I'm scared." Correct action: trust the gut, go to the ER. It was a heart attack. Trusting the AI would have meant delay and more damage.

The "I don't know why, but..." test

If you catch yourself saying "I can't explain it, but something's wrong," "it doesn't sound serious when I describe it, but...," or "everyone says it's nothing, but I'm not convinced," that is your velociraptor brain. You are detecting something you cannot yet put into words. That is a feature, not a flaw.

Being unable to be reassured is data
If an AI or a clinician says "it's nothing" and you stay worried, your continued worry is itself a finding. "Patient cannot be reassured" means investigate further, not "patient is anxious."

Where the velociraptor brain stops

It is excellent at detecting threat, deciding "I need help now," and trusting "my child is sick." It is not built to diagnose the specific condition, choose treatment, or set a medication dose. Use it to sense danger, then get medical care to figure out what it is and how to treat it.

False alarms are fine. Missing a real threat is fatal. Trust the hardware. It has been tested for 3.8 billion years.