A Short History of AI in Medicine: From First Algorithms to Today’s Robots
Dr. Chetan Kapila
Emergency Medical Officer, Fortis Hospital, Ludhiana, NCLEX and UKMLA Educator
GMC-Registered Doctor
Keywords: Artificial Intelligence, Healthcare, Drug Discovery, Medical Robotics, Machine Learning
It was a foggy night in the emergency room when the CT machine refused to load, and I—sleep-deprived, hungry, and betrayed by a cold samosa—looked at the ceiling and muttered, “AI, take the wheel.” That’s when my junior casually told me that a new AI tool had just diagnosed a stroke 12 minutes faster than the radiologist on call.
And it hit me. We’re no longer practicing medicine alone. We’re sharing the stethoscope with silicon minds.
Welcome to the story of Artificial Intelligence in medicine—a tale of promise, perseverance, pixels, and patients. This isn’t just about machines getting smarter. It’s about how we went from flowcharts to full-on robotic surgeons, and how AI is now the quiet genius in our clinical orchestra.
A TIME-TRAVEL THROUGH CIRCUITS: TIMELINE OF AI IN MEDICINE
Let’s rewind to a time when a “computer in the clinic” was as wild as a dinosaur in the OPD.
1965 – DENDRAL:
Created at Stanford, DENDRAL was one of the first rule-based systems. It didn’t diagnose humans but helped chemists identify organic molecules. I imagine it as a shy nerd who preferred beakers to blood pressure.
1972 – MYCIN:
This one tried to take on bacterial infections. MYCIN worked on if-then rules to suggest antibiotics. But it couldn’t prescribe if you sneezed off-script. Let’s say it would’ve failed an Indian viva with all our vague, poetic presenting complaints.
1980s – INTERNIST-1 & CADUCEUS:
These tried to mimic expert clinicians, with thousands of diagnostic rules. But they were like the overachiever batchmates—great on paper, froze during real-life chaos. You had to feed them textbook-perfect data.
1997 – IBM Deep Blue beats Kasparov:
Not medical—but symbolic. If a machine could outwit a chess grandmaster, could it also outsmart a stubborn differential?
2011 – IBM Watson enters healthcare:
Armed with all medical journals and a swagger, Watson was supposed to revolutionize diagnostics. But like many interns, it struggled with real-life uncertainty and, let’s face it, some of our legendary Indian handwriting.
2016–Present – Deep Learning Era:
This is where AI got smart. Algorithms learned to read ECGs, chest X-rays, CTs, even retinal scans. Some could detect diabetic retinopathy with better accuracy than ophthalmologists. I still remember a day when the AI flagged a lesion before I even saw the image—and I pretended like I spotted it first. Ego saved. Patient saved.
2020 – COVID-19 Pandemic:
AI went from an optional extra to a frontline warrior. Models were built overnight to predict ICU demand, assist in vaccine development, and track outbreaks. It was the geek finally stepping into the ring.
FROM “CLINICAL CALCULATORS” TO CO-CLINICIANS
Back in the day, medical AI was like that one senior who needed everything in bullet points. “If fever AND rash AND conjunctivitis = maybe measles.” There was no room for grey.
Today’s AI? It swims in grey.
It analyses thousands of data points, learns from patterns, and adapts. One system by Google can predict diabetic retinopathy from a retina photo with 90%+ accuracy. Another predicts in-hospital mortality better than most clinical scores.
Here’s how far we’ve come:
| Feature | Then(1970s-1990s) | Now (2020s) |
| Learning Method | Rule -Based | Deep Learning |
| Flexibility | Low | High |
| Accuracy | Variable | Near Human/Exceeding in Some Domains |
| Interface | Command Line | User Friendly, Integrated Into EMR |
| Scope | Focused (Single Disease) | Broad (Multiple Comorbidities, Real Time Prediction |
I remember a fake case from “my past life” as a rural intern. An old AI-based CD-ROM tried diagnosing a man with seizures as “cattle disease exposure.” Turns out, the patient had epilepsy, and the CD-ROM got confused by his background in dairy farming. But Now? AI wouldn’t just diagnose him—it’d book his MRI, message the neurologist, and tell me to eat lunch.
CHALLENGES THAT REFUSE TO DIE
Despite decades of upgrades, some problems remain stubborn as ever.
1. Bias and Data Discrimination:
An AI trained on Western populations might misread an Indian ECG. Once, a predictive tool flagged a Punjabi uncle as having a “rare genotype” because it hadn’t seen enough brown skin in its dataset. Solution? Train locally, test globally.
2. Black Box Syndrome:
Doctors don’t like being told “Trust me, it’s cancer” by a machine that can’t explain why. We need interpretable AI. One that says, “Because of these 7 data points and that CT feature.” Not just vibes and voltages.
3. Liability & Ethics:
If AI misdiagnoses, who gets sued? The doctor? The developer? The hospital? Or the ghost in the machine? As of 2025, we’re still figuring that out. Till then, we treat AI as an assistant—not a scapegoat.
4. Doctor-AI Trust Issues:
I once saw a senior consultant glare at an AI tool and mutter, “This bot thinks it knows more than me?” Spoiler: It did. But egos heal slower than wounds.
AI IN DRUG DISCOVERY & VACCINE DEVELOPMENT
One of AI’s most underrated talents is in biochemistry.
During COVID, AI helped design mRNA vaccines in record time. What used to take 5–10 years took 12–18 months. It predicted protein structures, optimized compounds, and ran virtual trials before we touched a petri dish.
Today, companies use AI to:
• Discover new antibiotics
• Design personalized cancer vaccines
• Simulate how a molecule will behave before making it in real life
• Repurpose existing drugs for new diseases
Imagine AI telling you: “This TB drug might actually work for pancreatic cancer.” That’s not magic. That’s math.
THE ROAD AHEAD: WHERE WE’RE GOING
In the next decade, AI will:
• Write discharge summaries
• Predict sepsis hours before vitals drop
• Guide robotic surgery with real-time feedback
• Coach junior doctors during CPR (yes, like Siri yelling ‘Push harder!’)
• Deliver telemedicine in villages without doctors
But the heart of medicine? That will always be human.
Because no machine can console a grieving mother, break bad news with softness, or hold a hand before surgery. AI can process vitals, but it can’t feel pulse.
FAKE ENDING, REAL HOPE
I once asked an AI model, “Will I be replaced?”
It replied: “Not if you keep learning.”
So I smiled, got back to work, and whispered: “You read the labs, I’ll break the news to the family.”
That’s what medicine should be—human and machine, mind and metal, data and compassion.
References
- Cedars-Sinai Staff. AI’s Ascendance in Medicine: A Timeline. Cedars-Sinai. April 20, 2023. Available from: https://www.cedars-sinai.org/discoveries/ai-ascendance-in-medicine.html
- Gastrointestinal Endoscopy Journal. History of artificial intelligence in medicine. 2020. Available from: https://www.giejournal.org/article/S0016-5107(20)34466-7/pdf
- Keragon. When Was AI First Used in Healthcare? The History of AI in Healthcare. 2024. Available from: https://www.keragon.com/blog/history-of-ai-in-healthcare
- Fowler GA. AI in Drug Discovery and Biomedical Research. Medium. 2024. Available from: https://gafowler.medium.com/ai-in-drug-discovery-and-biomedical-research-3229353b7519
- National Center for Biotechnology Information. The Role of AI in Drug Discovery:Challenges, Opportunities, and…2024.Availablefrom:https://pmc.ncbi.nlm.nih.gov/articles/PMC103020/
- IRB Barcelona. AI in Drug Discovery and Biomedicine. 2024. Available from: https://www.irbbarcelona.org/en/events/ai-drug-discovery-and-biomedicine
- UW Clinical Trials. AI in Biomedical Research Is Revolutionizing Drug Development. 2023. Available from: https://uwclinicaltrials.org/2023/11/06/ai-in-biomedical-research-is-revolutionizing-drug-development-clinical-innovation/
- Wikipedia. Isomorphic Labs. 2025. Available from: https://en.wikipedia.org/wiki/Isomorphic_Labs
- Wikipedia. AlphaFold. 2025. Available from: https://en.wikipedia.org/wiki/AlphaFold
- Financial Times. Google DeepMind duo share Nobel chemistry prize with US biochemist. 2024. Available from: https://www.ft.com/content/ba14c3a1-ac8e-42b9-a5ba-9d73cc1fff4c
