Alexander Sheppert, DO, PhD, MBA, is an internal medicine resident physician and AI researcher and the founder of Matic.
getty
In a modern colonoscopy, a camera is inserted through the anus and into the intestine. Watching the feed on a screen, an endoscopist looks for suspicious growths and cuts them out before they can turn into cancer.
In late 2021, four Polish endoscopy centers switched on software that watched the feed alongside the doctors, flagging anything that looked suspicious. After a few months of using the AI, the doctors got worse at finding polyps on their own. The conclusion parroted across the world was that AI is making doctors worse.
I don’t dispute the finding. Months of leaning on the software mean less practice flagging growths manually, and unused skills get rusty. Hardly anyone can recite a phone number now that we have contact lists. If you took away the backup camera, you'd watch the modern driver struggle to back into a tight spot.
The important question to ask here is whether you would want this software at your own colonoscopy. Across 44 randomized trials and more than 36,000 patients, the data seems to suggest that the answer is yes. AI assistance in those trials raised the detection of adenomas, the precancerous growths the exam exists to catch, from 36.7% of exams to 44.7%. That means one-fifth more patients got a precancerous growth snipped out before it could turn into cancer.
Nobody Wants A Rigid Sigmoidoscopy
We have deskilled endoscopists before, repeatedly and on purpose. Before the colonoscope, a doctor looking for polyps performed rigid sigmoidoscopy: a straight hollow tube was inserted into the rectum, lit in the early days by candlelight or lamplight bounced off a mirror. Around 1970 came flexible instruments, bundles of fine glass fibers that carried the image up the tube to a little eyepiece the doctor pressed his eye to, the way you would with a microscope. Then came the camera and the screen.
Each step retired the skill before it. Doctors who moved to flexible scopes had little reason to keep their rigid-tube technique sharp; doctors who moved to screens stopped practicing at the eyepiece. The older technique no longer had a job to do. Most people don’t care whether their endoscopist has kept up the old rigid-tube craft. They want the exam that gives them the best chance of catching a dangerous growth early.
I believe using AI to flag growths is the next step. The methods we use to look through a patient’s rectum are not sacred. When a superior method shows up, we must adopt it. In my view, it is a mistake to treat every atrophied skill as a tragedy.
A Worry As Old As Writing
The complaint we're hearing is about 24 centuries old. In Plato’s Phaedrus, the god Theuth presents his invention, writing, to the Egyptian king Thamus, praising it as a cure for forgetting. Thamus is unimpressed. Writing, he warns, will “create forgetfulness in the learners’ souls, because they will not use their memories.”
He was right. Literacy dissolved the trained memory of the oral world; almost no one now holds the Iliad in their head the way a Homeric reciter once did. We rarely mourn it; literacy today is arguably more accurate and convenient.
What About The Day The Software Is Down?
I hear one common objection frequently: Won’t you be glad that your endoscopist can still do your colonoscopy if the software goes down?
According to the report linked above, manual colonoscopy is less effective than the AI-assisted version. If the software is down, rescheduling is a real cost, and nobody who has drunk the prep wants to drink it twice.
The answer to fragile software is the boring one used everywhere else in medicine: redundancy, maintenance and a vendor on the hook, not a standing corps of doctors kept in practice at what research has shown is a less effective way of doing the job.
The Right Measure
Let’s measure what matters: How many precancerous growths get caught with the tool running? We don't need to measure how sharp the doctor is with the software unplugged. I would bet that few patients book the unassisted version on purpose any more than they ask for the candlelit tube.
Doctors need to be trained on the most effective technique available, and today, that means being trained with the software on. I don’t want doctors to skip the exam that research shows catches more precancerous growths because a headline mourned a dying skill.
Your Version Of The Rigid Tube
This is bigger than gastroenterology. Whatever your work is, some part of it is becoming the rigid sigmoidoscope: a skill you were proud of that has been made obsolete by a tool that does the job better.
Chess players went through this a generation ago. When engines surpassed grandmasters, purists warned that players raised on computer analysis would never develop real board vision. Then the engine-raised generation became some of the strongest players in the world.
Across the board, the skill worth honing now is the process of working with the machine. Endoscopists may lose some of their manual skills, but they may catch more cancer, which was always the job. If the tools in your field are getting good, the same trade is coming for you. Take it.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
>> Home