Free tools Windows power users keep installed
One-click scans. No signup required.
A 2025 study found that adenomas were detected less often during colonoscopies performed without AI after doctors had begun routinely using an AI polyp-detection tool. The rate fell from 28.4% before AI exposure to 22.4% afterward among 1,443 patients at four Polish centers. That is a six-percentage-point absolute decline, or roughly a 20% relative reduction.
But the study did not show that doctors broadly “lost the ability to spot cancer.” It was a small, retrospective observational study of 19 endoscopists. It measured detection of adenomas—precancerous polyps—not cancer itself, and it cannot prove that AI caused the decline or that patients experienced worse long-term outcomes. The study was published in The Lancet Gastroenterology & Hepatology in August 2025.
What the study actually examined
The paper, titled “Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy,” analyzed standard colonoscopies performed while AI was turned off. Researchers asked whether endoscopists who had begun routinely working with AI behaved differently when they had to inspect the colon without computer assistance.
The analysis was nested in the ACCEPT (Artificial Intelligence in Colonoscopy for Cancer Prevention) trial. AI tools were introduced at four endoscopy centers in Poland at the end of 2021. Investigators compared non-AI procedures in the approximately three months before implementation—September 8 through December 2021—with procedures in the three months afterward, through March 9, 2022. The study involved 1,443 patients and 19 endoscopists. PubMed provides the study design and participant details.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →The main numbers
| Measure | Before AI exposure | After AI exposure |
|---|---|---|
| Patients undergoing non-AI colonoscopy | 795 | 648 |
| Adenoma detection rate (ADR) | 28.4% (226 patients) | 22.4% (145 patients) |
| Absolute change | — | −6.0 percentage points |
| Relative change | — | Approximately −20% |
| Adjusted odds ratio associated with prior AI exposure | — | 0.69 (95% CI 0.53–0.89) |
| p-value | — | 0.0089 |
The reported 20% figure is a relative reduction, not a 20-percentage-point fall. The researchers reported a 95% confidence interval for the absolute difference of −10.5 to −1.6 percentage points. The association remained statistically significant after multivariable adjustment, but statistical significance does not by itself establish causation.
What adenoma detection rate means
Adenoma detection rate is the proportion of screening or diagnostic colonoscopies in which at least one adenoma is found. Adenomas are growths that can develop into colorectal cancer, so finding and removing them is an important part of cancer prevention.
A lower ADR in this study means that adenomas were found less often in the later, non-AI procedures. It does not mean that 20% of cancers were missed, that every doctor’s skill fell by exactly 20%, or that the patients developed more cancer. The study did not measure cancer incidence, interval cancers, mortality, or other long-term outcomes.
What “deskilling” could mean here
The study raises a concern about what might happen when clinicians repeatedly rely on an automated second reader and then work without it. It did not directly measure the cause of the lower ADR. Several explanations remain possible:
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errors- Automation bias: clinicians may give too much weight to software alerts or their absence.
- Reduced vigilance: knowing that a system is scanning the image may change the intensity of the independent search.
- Behavioral adaptation: clinicians may learn to wait for alerts rather than maintain a full visual inspection.
- Skill decay: repeatedly outsourcing part of a perceptual task may reduce practice of that task.
- Workflow changes: AI could affect withdrawal speed, attention allocation, or responses to ambiguous findings.
These are hypotheses, not mechanisms demonstrated by the data. The investigators recorded detection outcomes, not eye movements, attention, motivation, or cognitive workload.
Does this conflict with studies showing AI can improve colonoscopy?
Not necessarily. The studies may be asking different questions.
- Immediate assistance: Does a clinician find more lesions while AI is active during the procedure?
- Human behavior: Does routine use change how the clinician searches?
- Skill retention: Does performance change when the tool is removed?
- Patient outcomes: Does any change lead to more interval cancers or worse survival?
The deskilling study primarily addressed the third question. A randomized trial that compares AI-assisted and unassisted procedures during a single examination does not necessarily test whether months of AI exposure changes later unaided performance. The authors suggested that repeated exposure could make the “non-AI” clinicians in later trials different from genuinely AI-naive clinicians; that is a hypothesis, not a settled explanation of all apparently conflicting evidence. A clinical-context review in Nature Reviews Gastroenterology & Hepatology discusses this distinction.
What the study does not prove
- It does not prove causation. This was a before-and-after observational comparison, not random assignment to long-term AI exposure or no exposure.
- It does not describe doctors generally. The procedures came from 19 endoscopists at four centers in Poland.
- It does not establish a rapid or permanent loss of skill. The comparison covered roughly three months on either side of implementation but did not identify when any change began or whether it persisted.
- It does not apply automatically to every AI system. The specific tools and interfaces are important, and public records do not establish that this result applies to newer products or other specialties.
- It does not show patient harm. No increase in colorectal-cancer diagnoses, interval cancers, complications, treatment delays, or deaths was measured.
- It does not show that AI-assisted colonoscopy is harmful overall. The study focused on procedures without AI after prior exposure, not the complete risk-benefit balance when the tool is operating.
There was also a later correction noting that the indication for colonoscopy had been omitted from a supplementary multivariable-analysis table. The correction record identifies the omission but, in the available record, does not state a revised interpretation of the headline result. See the publisher’s correction notice.
Why the headline is misleading
“Doctors quickly lose the ability to spot cancer” compresses several different claims into one dramatic sentence:
- “Cancer” is too broad: the measured endpoint was adenoma detection.
- “Lose the ability” is too strong: ADR declined; the clinicians did not stop being able to perform colonoscopy.
- “Quickly” is not established: the study compared periods before and after implementation but did not determine the exact timing of a decline.
- “Doctors” overgeneralizes: the sample was 19 endoscopists in four Polish centers.
- “AI” is overbroad: this concerned AI-based polyp detection in colonoscopy, not chatbots, radiology systems, pathology tools, or consumer apps.
The most accurate summary is narrower: prior routine AI exposure was associated with lower adenoma detection during later colonoscopies performed without AI in this group of endoscopists.
Questions hospitals and vendors should ask
The result is a reason to evaluate both assisted and unassisted performance, not a reason to reject every AI tool. A deployment review should ask:
- Does the system improve detection in the intended patient population and across different clinicians?
- What happens when the system is offline or produces no alert?
- Are clinicians periodically assessed during non-AI procedures?
- Does training emphasize independent inspection rather than treating the tool as a substitute for search?
- Are false-positive alerts causing distraction or alert fatigue?
- Does the system show uncertainty, or only binary prompts?
- Are audit logs available for alerts, clinician responses, missed lesions, and downtime?
- Are outcomes monitored by clinician experience, exposure duration, equipment, software version, and patient mix?
Practical safeguards
Potential safeguards include regular non-AI sessions or competency checks, training about automation bias, a defined downtime procedure, and monitoring of both AI-assisted and unassisted ADR. Organizations can also track withdrawal time, bowel-preparation quality, lesion size and location, histology, and—when feasible—longer-term outcomes such as interval cancers.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallThese measures were not tested by this study, so they should be treated as risk-management options rather than proven fixes. Their purpose is to preserve independent clinical judgment while retaining any benefit the AI provides during use.
The broader lesson
The important question is not simply whether AI “works” or “fails.” A system could increase detection while active and still create a risk that unaided performance changes after repeated exposure. Determining whether that trade-off exists requires larger, prospective studies across centers, clinicians, products, and longer follow-up, with patient outcomes as well as surrogate measures such as ADR.
For now, this study is best read as a warning signal about reliance and skill retention—not proof that AI made doctors broadly incapable of detecting cancer.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
