If you've ever typed a weird symptom into Google at 1 a.m. and scared yourself half to death, you already know both the promise and the danger of AI diagnosis. Now imagine doing it in a hospital room, with a parent who's unconscious and fading. That's exactly what a British anesthesiologist did last May — and it worked.
What Happened: A Doctor, a Phone, and a Hunch That Beat the Clock
She'd Lived With Parkinson's for Years. Then the Seizures Wouldn't Stop.
Ed Patrick puts people to sleep for a living. He's a British anesthesiologist — the doctor in the operating room who keeps you unconscious and breathing while a surgeon works. In May, he got the call nobody wants: his mother, Ann, was dying.
Ann had lived with Parkinson's disease for years. In plain English, Parkinson's is a brain condition that slowly damages the cells that control movement; as it progresses, it can take memory and sharp thinking along with it. Now she was in continuous seizures — one rolling into the next — and the anti-seizure drugs weren't touching them.
By the time Patrick reached her bedside, Ann was unconscious and her kidneys had started to fail. That's the body's way of saying the clock is running out.
He Asked Gemini One Question
Rather than stand there helpless, Patrick opened Google Gemini — Google's AI assistant — and asked it to find and summarize recent medical research on symptoms that matched his mother's.
That's the whole move. Not “diagnose my mum.” Just: read everything published on this pattern, and tell me what fits.
The AI surfaced a suspect nobody in the room had raised: a vitamin B6 deficiency, possibly caused by the Parkinson's medication Ann had been taking for years.
Why Almost Nobody Would Have Guessed Vitamin B6
Vitamin B6, also called pyridoxine, helps your brain run its chemical messaging. Think of it as the oil in an engine — you never notice it until it's gone, and then everything grinds to a halt.
Carbidopa/levodopa, the standard Parkinson's drug sold as Sinemet, can in rare cases drain the body's B6. When that happens, some patients get seizures that simply ignore normal seizure medication.
How rare is rare? Only 14 similar cases have ever been documented in the medical literature. Fourteen. That's not the first thing a busy medical team reaches for at 3 a.m.
The Seizures Stopped Within Minutes
Patrick took the idea to Ann's medical team. They gave her vitamin B6 intravenously — straight into a vein, so it reaches the bloodstream immediately instead of waiting on digestion.
“What we did not anticipate was the speed and magnitude of the response. Within minutes of administering vitamin B6, Ann's seizure activity stopped,” said neurologist Dr. Bruno Gran.
Ann hasn't had another seizure since. And cognitively — memory, thinking, the person behind the eyes — Patrick says she seems to have gone back five to ten years.
The story has been reported by Piq Markets, Parkinson's Insights, Chortle and Custom Map Poster. There's a regulatory footnote too: in March 2026 the FDA flagged the link between carbidopa/levodopa and B6 deficiency causing seizures, and noted that B6 supplements resolved the seizures in all 14 reported cases.
Patrick now wants researchers to dig into whether vitamin B6 could help other people with Parkinson's — not only in emergencies like his mother's, but possibly against the slow decline in memory and thinking. Doctors say proper clinical trials would be needed first. A clinical trial, by the way, is just a controlled test: you give the treatment to one group and not another, then compare, so you're not fooled by luck.
What It Means for You: Not a Doctor, a Tireless Research Assistant
The headline here isn't “AI is a doctor now.” It's this: AI is the assistant who has read everything, never gets tired, and never makes you feel stupid for asking. That's useful in a hospital room — and it's just as useful on a random Tuesday afternoon.
At Home: Decoding a Parent's Pill Bottles
You're the one who drives your dad to appointments and carries the plastic bag of prescriptions. Type the full list into Gemini — names, doses, what each one is for — and ask for a plain-English summary of what interacts with what, plus the questions you should be asking. You're not diagnosing anything. You're walking into a 15-minute appointment with five sharp questions instead of a blank stare. Families who do this tend to get better care, for the simple reason that they ask better.
At Work: Turning Dense Documents Into Something Readable
Paste in a policy, a spec sheet, a benefits handbook or a contract clause and ask for the version a smart 14-year-old could follow. Then ask what's missing, what's ambiguous, and what you'd be committing to if you signed. You'll still send anything binding to a lawyer — but you'll walk into that conversation already knowing the terrain instead of paying someone to explain the map to you.
In Business: Research That Used to Cost You a Consultant
Supplier risk, a new market, a competitor's pricing page, a regulation you've never heard of. Ask for a summary, then the three biggest risks, then what a skeptical CFO would say about all of it. That's roughly a day of junior-analyst work compressed into a coffee break. The output isn't the deliverable — your judgment still is — but you start from a running head start instead of zero.
For Students: A Tutor Who Never Sighs
Paste a chapter, ask for an explanation at the level of a curious high-schooler, then ask it to quiz you on what you just read. The trick is to make it explain, not to make it write your essay. Explaining is where the learning actually happens. If you want more free tools to stack on top of Gemini — summarizers, prompt helpers, writing and image utilities — the collection at mykreatool.com is a good place to browse without spending a cent.
For Creatives: A Sparring Partner at 2 a.m.
Stuck on a name, a hook, a shot list, or a chapter that isn't working? Give it the constraints and ask for ten deliberately bad options. Bad options are faster to react to than a blank page, and your own taste does the rest. Nobody's taking your voice here — you're skipping the part where you stare at the wall for 40 minutes.
For Extra Income: Selling the Skill, Not the Answer
What people pay for isn't “knows how to use AI.” It's knowing which question to ask and how to check the answer. Freelancers charge real money for research briefs, plain-English summaries of technical documents, and content outlines built on genuine sources. Same tools, same afternoon, billable work — provided you verify everything before it goes out under your name.
How to Try It Right Now: Five Steps, Free First
Step 1 — Start with the free version of Gemini. Go to gemini.google.com, sign in with a Google account and just start typing. The free tier costs nothing, asks for no card, and is more than enough for research summaries.
Step 2 — Give it context, not a one-liner. Weak: “what causes seizures?” Strong: “My mother has Parkinson's and takes carbidopa/levodopa. She's having seizures that don't respond to medication. Summarize recent research on unusual causes.” Context is what turns a fortune cookie into a useful answer.
Step 3 — Ask for sources and dates. “Give me the studies or case reports, with dates.” Then check that they exist. AI can sound very confident about things that were never published, and citations are the fastest place to catch it.
Step 4 — Ask it to argue with itself. “What are the three strongest arguments against this explanation? What else fits these symptoms?” You want the second opinion, not just the first one.
Step 5 — Take it to a human expert. AI handed Patrick a lead; a neurologist and a medical team made the call. Keep that order. Use AI to ask better questions, never to replace the person holding the stethoscope.
The Upsides: What Actually Changes
Speed. Research that used to mean a week of reading becomes a 20-minute summary you can actually act on.
Access. You don't need a medical library card or a high-priced consultant to understand the words on a lab report anymore.
Plain language. Jargon turns into English, and English is where decisions get made.
A second pair of eyes. Doctors are busy and human. A careful, specific question from you adds one more pass over the details.
Rare things get spotted. Fourteen cases in the entire literature is exactly the kind of needle a fast reader can pull out of a haystack.
The Limits: Read This Before You Trust a Chatbot With Your Health
Gemini didn't diagnose Ann Patrick — it pointed at a possibility that a trained team then evaluated, tested and treated. That order matters. These models can be confidently wrong, they don't know your full medical history (kidney function, allergies, every other drug in the cabinet), and they carry no liability for what you do with their answer. They also can't feel the pulse or see the patient. Fourteen documented cases is a signal worth chasing, not proof that B6 fixes Parkinson's — that's precisely why doctors say clinical trials come first. And be careful with privacy: don't paste names, ID numbers or anything that identifies a real person into a chat window.
Conclusion: One Thing to Do Today
Write down one health question you've been carrying around — yours or a family member's — and spend 15 minutes turning it into a better question with Gemini. Not an answer. A better question. Then take that question to a real professional and watch how differently the conversation goes. Ed Patrick did the same thing, from a hospital room, with the clock running. He didn't have special access or a secret tool. He had a phone, a hunch and the discipline to hand the idea to the people who could actually act on it.
Comments 0