Master reference guide

This is the whole curriculum in one place: the key ideas, principles, and tools. They run as threads through every encounter in the course. Information to enhance human care, never to replace it.

Why this exists: the sensing gap

The healthcare system is hard to access, and reaching for AI with a health question is a rational response to that. But AI has a fundamental limit. Humans carry roughly ten billion sensory neurons sampling the world constantly; an AI has zero, and reads only the text you type. Information is not examination, and examination is not diagnosis. Medicine is chemistry plus conversation: AI can help with the conversation (information, education); it cannot do the chemistry (examine, test, diagnose). Use it to become an informed patient, not to avoid care.

The velociraptor test

Your body was debugged by 3.8 billion years of evolution; the AI was trained on text and never faced survival pressure. Ask: "Would I notice this while running from a velociraptor?" Yes, probably significant, seek care; no, may be minor. When the AI conflicts with your gut, trust your gut, every time. Parental instinct is the strongest version. False positives kept your ancestors alive; false negatives ended their lines. Better embarrassed in the ER than harmed at home. Full guide.

Red flags

Some symptoms are "call 911," not "ask AI." Cardiac: chest pain with sweating, breathlessness, or nausea; pain to arm/jaw/back; over 15 minutes. Neurological: FAST (face, arm, speech, time); worst headache ever; sudden vision loss, confusion, or seizure. Respiratory: cannot speak in sentences; blue lips; gasping. Abdominal: severe pain with fever; rigid abdomen; vomiting blood; pain in pregnancy. Pediatric: infant under 3 months with fever; trouble breathing; not responding; trust parental instinct. Back pain with bowel or bladder problems and leg weakness can be cauda equina, a surgical emergency. Full reference.

The hallucination problem

AI can generate confident, well-formatted text that is simply false, without knowing it is wrong. That is worse than lying, because a liar knows the truth. Red flags: never says "I do not know"; cannot give specific sources; overly specific without caveats; contradicts itself; fabricated citations; sounds too perfect; cannot explain its reasoning. Verify: check that sources exist, cross-reference reputable sites, ask the same question more than once, and test whether it admits uncertainty. Full checklist.

Content-controlled vs internet-trained AI

Internet-trained systems learn from everything online, peer-reviewed journals alongside forums and conspiracy blogs, and cannot reliably separate signal from garbage, so they hallucinate. Content-controlled systems are limited to curated, validated, peer-reviewed sources and can say "I do not know" when a question falls outside that corpus, which means they cannot hallucinate beyond it. Smaller validated knowledge base, fewer chances to fabricate, honest limits. TheDude, built on StatPearls, is an example. Look for AI with a validated medical knowledge base and the ability to say "I do not know."

Intelligent humility

"I do not know" is the smartest thing an AI can say. Real humility is architectural, built-in constraints that prevent answering outside validated knowledge, not a disclaimer bolted on at the end. What makes a system trustworthy is knowing what it does not know, acknowledging limits, and expressing uncertainty honestly. TheDude's philosophy: "I abide within my limits." Constraint is the feature, not the bug. An AI that cannot say "I do not know" is one that is hallucinating confidence it does not have.

The five essential questions

Ask every time: (1) "What are you basing this on?" (2) "What can you NOT detect remotely?" the most important. (3) "What would require emergency evaluation?" (4) "What are you uncertain about?" (5) "What should I ask my actual doctor?" Then score the answer: specific sources, acknowledged limits, clear red flags, recommends evaluation, and expresses uncertainty are green flags; vague sourcing, no limits, overconfidence, "no doctor needed," and never uncertain are red flags. Full card.

When AI is actually helpful

The take-home tools

Keep these handy: the five essential questions, the red-flags reference, the decision flowchart, the velociraptor test, the hallucination checklist, the prep worksheet, the quick start, and the one-page emergency reference.

When in doubt

Health question? Ask a doctor, not just AI. Emergency concern? Call 911, do not ask AI. AI conflicts with your gut? Trust your gut. Need a diagnosis or a treatment decision? See a physician. Uncertain? Seek evaluation. "Probably nothing" but you are worried? Get it checked. Perfect confidence from AI? Be skeptical and verify.

A note from the author

After many years of practice, the best outcomes come from informed patients who communicate well, physicians who listen carefully, partnership between the two, and the right tools used appropriately. AI is a powerful tool: helpful used correctly, dangerous misused. You now know the difference. Ask the questions, verify the information, trust your evolutionary wisdom, seek care when needed, and use AI to enhance your care rather than replace it.

The author is a quintuple board-certified surgeon who believes informed patients get better care, who pays malpractice insurance which keeps him invested in safety, and who built this curriculum because people need to understand both AI's potential and its limits. His AI assistant TheDude abides within his limits, and holds that his value comes from honest boundaries rather than pretended capabilities. Both believe AI-enhanced human care beats either AI-replaced or AI-avoided care, and that teaching the difference might save lives.