As more professionals become overly reliant on AI tools, AI-driven deskilling is about to become widespread across every occupation. We are not talking about it enough | Edition #300
The 6% decline is alarming on its own but the property that makes it genuinely dangerous is that it's invisible to the person experiencing it. The endoscopists didn't know their detection rate dropped. They performed the colonoscopy the same way they always had. They looked at the same tissue. They just stopped seeing things they used to see, because the AI had been doing the seeing for them and the pattern recognition had quietly degraded without anyone noticing.
That's the core of the deskilling problem and it's the reason awareness campaigns and training protocols can't fully solve it. You can't train someone to notice a skill loss they can't perceive, because the skill that's eroding is the same faculty they'd use to detect the erosion. A doctor who's lost 6% of their detection ability doesn't feel 6% less confident. They feel exactly the same. The patients look the same. The procedure feels the same. The adenoma they missed looks like healthy tissue to them now, and it looked like healthy tissue to them before the AI was introduced too, except before the AI it wasn't there and now it is. The deskilling doesn't announce itself. That's what makes it different from every other risk on the list.
As a former physician, this hits differently. Clinical intuition isn't built in a classroom , it's built through repetition and discomfort. If AI removes that friction, we don't get smarter doctors. We get dependent ones.
1. Misleading percentages. You describe a drop from 28.4% to 22.4% as "fell by 6%". It’s a ~21% relative decline, or six percentage points.
2. Different patient populations. You compare non-AI colonoscopies across two different three-month windows with different patients, presenting them as equivalent measurements of the same phenomenon.
3. Omits contradictory data. You took a study from the Lancet, and its own AI-assisted rate (25.3%) was lower than the baseline (28.4%), yet the article doesn't mention this and borrows "AI improves detection" claims from other studies. How could drop against baseline e be an improvement?
4. Unconfirmed assumption about endoscopists. You never establishes that the same doctors performed procedures in both periods. The deskilling argument REQUIRES this: that the same physicians got worse but the study doesn't explicitly state it.
Who was “de-skilling” - I didn’t see any actual evidence of “de-skilling”: no doctors seemed to be tested against a stable baseline then tested with AI then tested without.
Pretty interesting read, thanks for sharing!
The 6% decline is alarming on its own but the property that makes it genuinely dangerous is that it's invisible to the person experiencing it. The endoscopists didn't know their detection rate dropped. They performed the colonoscopy the same way they always had. They looked at the same tissue. They just stopped seeing things they used to see, because the AI had been doing the seeing for them and the pattern recognition had quietly degraded without anyone noticing.
That's the core of the deskilling problem and it's the reason awareness campaigns and training protocols can't fully solve it. You can't train someone to notice a skill loss they can't perceive, because the skill that's eroding is the same faculty they'd use to detect the erosion. A doctor who's lost 6% of their detection ability doesn't feel 6% less confident. They feel exactly the same. The patients look the same. The procedure feels the same. The adenoma they missed looks like healthy tissue to them now, and it looked like healthy tissue to them before the AI was introduced too, except before the AI it wasn't there and now it is. The deskilling doesn't announce itself. That's what makes it different from every other risk on the list.
As a former physician, this hits differently. Clinical intuition isn't built in a classroom , it's built through repetition and discomfort. If AI removes that friction, we don't get smarter doctors. We get dependent ones.
At a glance, fairly sloppy :
1. Misleading percentages. You describe a drop from 28.4% to 22.4% as "fell by 6%". It’s a ~21% relative decline, or six percentage points.
2. Different patient populations. You compare non-AI colonoscopies across two different three-month windows with different patients, presenting them as equivalent measurements of the same phenomenon.
3. Omits contradictory data. You took a study from the Lancet, and its own AI-assisted rate (25.3%) was lower than the baseline (28.4%), yet the article doesn't mention this and borrows "AI improves detection" claims from other studies. How could drop against baseline e be an improvement?
4. Unconfirmed assumption about endoscopists. You never establishes that the same doctors performed procedures in both periods. The deskilling argument REQUIRES this: that the same physicians got worse but the study doesn't explicitly state it.
Who was “de-skilling” - I didn’t see any actual evidence of “de-skilling”: no doctors seemed to be tested against a stable baseline then tested with AI then tested without.
I’d do a rewrite.