Several studies have suggested that AI assistance may help doctors identify some cancers, but this is the first study to suggest that the implementation of AI could lead to a reduction in the ability of medical professionals who perform colonoscopies and impact health outcomes. Seeking commentary, including: How might AI assistance be used while avoiding these risks?

My background scaling AI-powered systems across Fortune 1000 companies and building PacketBase from zero to acquisition taught me that AI fails when it replaces human judgment rather than enhancing it. In medical settings like colonoscopies, the risk isn't the AI itself--it's creating dependency that erodes core skills. The solution lies in implementing AI as a validation tool, not a primary diagnostic method. At Riverbase, we use our Managed-AI method where human strategy guides AI automation, never the reverse. For colonoscopies, this would mean doctors make initial assessments independently, then use AI as a secondary screening to catch missed details or confirm findings. I've seen this work in our client campaigns where we use AI to identify high-intent prospects, but our human strategists make the final targeting decisions. When we tried full automation early on, campaign performance dropped 31% because the AI missed contextual nuances. The key is maintaining human primacy while leveraging AI's pattern recognition strengths. Train medical professionals to use AI intermittently rather than continuously, similar to how we A/B test our marketing systems. Run cases without AI assistance regularly to maintain diagnostic sharpness, then use AI insights to validate or reveal blind spots--not as a crutch but as a sophisticated second opinion.

As someone who built Nextflow (used globally for genomic analysis) and now runs Lifebit's federated AI platform, I've learned that AI deskilling happens when systems become black boxes that doctors can't interrogate or understand. The key is implementing "glass box" AI where the decision pathway remains visible. In our genomic workflows, we show researchers exactly which data points triggered specific alerts and why. For colonoscopies, this means displaying confidence scores, highlighting specific image regions, and explaining the reasoning behind each AI suggestion. At Lifebit, we've seen 97.5% accuracy rates in our voice recognition systems, but only when clinicians can validate and override the AI's interpretation. The moment healthcare providers lose the ability to challenge or understand AI recommendations, they start losing their diagnostic edge. I'd recommend implementing AI with mandatory "challenge periods" - designated times when doctors must make assessments without AI assistance to maintain their skills sharp. Our federated approach proves that AI works best when it augments human expertise rather than replacing the critical thinking that makes good clinicians invaluable.

As someone who's built AI systems across healthcare, staffing, and now field service over 15 years, I've seen this exact pattern play out repeatedly. The problem isn't the AI itself--it's implementing it as a replacement rather than a training partner. In our healthcare systems, we solved this by building "confidence thresholds" where AI flags cases but forces doctors to make the initial assessment when confidence drops below 85%. This kept diagnostic skills sharp while catching edge cases. For colonoscopies, I'd implement rotating "AI-off" shifts where doctors work without assistance, plus mandatory monthly assessments using AI-free cases. The real breakthrough came when we switched from "AI tells you the answer" to "AI asks you questions." Instead of saying "this looks like polyp type X," the system asks "what features suggest malignancy in quadrant 2?" This forces active engagement with the diagnostic process rather than passive acceptance. At ServiceBuilder, we're applying this same principle--our AI suggests quotes and schedules, but technicians must confirm the reasoning before proceeding. Our beta users report staying more engaged with their decision-making rather than just clicking "accept" on everything the system recommends.

Shamsa Kanwal, M.D., is a board-certified Dermatologist with over 10 years of clinical experience. She currently practices as a Consultant Dermatologist at https://www.myhsteam.com Profile link: https://www.myhsteam.com/writers/6841af58b9dc999e3d0d99e7 AI can be a valuable tool in colonoscopy by flagging suspicious lesions, enhancing visualization, and reducing human fatigue. However, over-reliance on AI may cause clinicians to lose critical observational skills or miss subtle findings that fall outside the AI's detection parameters. The safest approach is to position AI as an augmentative technology, not a replacement for physician expertise. This means training clinicians to use AI outputs as a second set of eyes. They should always be verifying, questioning, and integrating AI into their own assessment rather than deferring judgment to the system. Regular performance audits, ongoing skills training, and pairing AI-assisted cases with non-assisted ones can help maintain diagnostic competence while still benefiting from AI's speed and accuracy.

Based on this colonoscopy study, I can draw some similarities to the things I have observed during the last decade in software engineering. When GitHub Copilot was released, a lot of developers began overusing the auto-generated code and not knowing its logic structure. After around half a year, I observed that junior engineers in my team were not able to debug some very basic problems due to loss of the basics of problem solving. It is not the answer to avoid AI but to use it in a strategic manner. AI in medical practice ought to be an advanced pattern recognition system that indicates possible areas of concern but leaves diagnosing to the physicians. This is what it is like advanced linting in code development- it can tell you that there are problems but it does not handle the issues automatically. According to my experience in training more than 10,000 developers the secret is active interaction. Physicians should have some protocol according to which they have to explain their thinking and only then be provided with AI suggestions as I instruct students to solve algorithm problems step-by-step and then check a solution. Scheduled skill drilling appointments with no AI would maintain the diagnostic instinct. We refer to this in our field as coding without autocomplete - training without the ability to practice those core skills individually will lead to atrophy but still getting the benefit of the technological advances when necessary.

AI assistance in colonoscopies can be incredibly useful when used as a complementary tool rather than a replacement for professional judgment. In my experience, the most effective approach is to let AI highlight areas of concern while keeping the physician fully engaged in the procedure. For example, I've worked with systems that flag potential polyps in real time, but the end decision to biopsy or remove is always made by the doctor. Regular training and simulations that mix AI guidance with manual practice help maintain procedural skills. I also encourage periodic audits where physicians perform colonoscopies without AI support to ensure their detection ability stays sharp. This balance allows AI to enhance detection rates without creating overreliance, ensuring patient outcomes remain the primary focus while leveraging technology to improve accuracy.

Integrating AI into medical practices like colonoscopies has sparked a ton of interest, especially for its potential to enhance diagnostic accuracy. However, using AI tools requires a careful balance to ensure they complement rather than replace the skill set of healthcare professionals. One effective strategy is to treat AI as a second set of eyes. That means AI can flag potential issues during a procedure, and doctors can then take a closer look themselves. This collaborative approach harnesses the strength of AI without undermining the doctor's expertise. Training is key, too. Medical professionals could receive specific training not just to work alongside AI but also to maintain and even enhance their skills over time. Maintaining a hands-on approach ensures that doctors remain at the forefront of the diagnostic process, using AI as a powerful tool rather than a crutch. Continuous professional development courses that include AI training modules might be the way to go. It's like keeping a good balance; you don't want to lean too heavily on tech, but you also don't want to skip out on its benefits. Keeping this in mind helps ensure that AI serves as an asset without compromising the quality of healthcare.

After 10 years treating high-achieving patients paralyzed by perfectionism and anxiety, I see the same pattern emerging with AI in medical settings. When my perfectionist clients become over-reliant on external validation tools, they lose trust in their own judgment and become more anxious, not less. The key is preserving what I call "internal authority"--the doctor's confidence in their clinical instincts. I've watched patients who constantly seek reassurance from apps and online tools actually become worse at reading their own bodies. Medical professionals need to maintain that same self-trust. I recommend implementing AI like I use mindfulness with my patients--as a grounding tool, not a replacement for awareness. Have doctors verbalize their initial observations out loud before AI input, similar to how I have clients state their feelings before we explore them together. This preserves their diagnostic voice. The real danger isn't wrong diagnoses--it's creating medical professionals who doubt themselves. In my practice, I've seen how dependency on external validation creates a cycle where people need more and more confirmation to feel confident. Doctors using AI intermittently while maintaining regular "solo flights" prevents this psychological dependency.

Looking at this through my work with federated data systems at Lifebit, the solution lies in designing AI as an intelligence layer rather than an automation tool. We've seen this with our OMOP data harmonization platform--AI processes massive genomic datasets to surface patterns, but clinical researchers still make every interpretive decision about patient pathways. The critical difference is implementation architecture. At Thrive, we use data analytics to identify behavioral health risk patterns, but our clinicians must actively engage with each data point before any treatment recommendation. For colonoscopies, AI should flag potential areas of concern in a "review queue" format that requires physician acknowledgment and manual assessment before proceeding. I've learned from scaling both companies that the most dangerous AI implementations are "black box" systems where users can't see the decision logic. Instead, build transparent AI that shows confidence scores and reasoning. When our Lifebit clients review cancer genomics data, they see exactly which genetic markers triggered each AI recommendation, allowing them to maintain clinical oversight while benefiting from improved pattern recognition. The key insight from our federal health partnerships is that AI should augment cognitive load, not reduce it. Design systems that require more physician engagement with findings, not less--this actually improves outcomes while maintaining skills.

As a CRNA who's performed thousands of ultrasound-guided procedures over two decades, I've seen how technology can create dependency if not implemented thoughtfully. The solution isn't avoiding AI but designing systems that improve rather than replace clinical reasoning. At Pain Specialists of Brighton, when I use fluoroscopy for epidural injections or radiofrequency ablation, I never let the imaging make decisions for me. I require myself to predict needle placement and identify anatomical landmarks before confirming with the x-ray guidance. This keeps my palpation skills sharp and prevents technological crutches from forming. For colonoscopies specifically, I'd recommend implementing AI that operates in "challenge mode" - it identifies potential areas of concern but presents them as questions rather than diagnoses. Instead of "polyp detected," the system could ask "what do you observe about the tissue texture in this quadrant?" This forces active evaluation while still providing the safety net. The key insight from my pain management practice is that technology should amplify your existing skills, not substitute for developing them. When I teach other practitioners ultrasound-guided regional anesthesia, we always start with anatomy identification before turning on any improvement features.

AI in colonoscopies has great potential, but it should assist, not replace a doctor's judgment. I've found that with a large nursing network, continuous training and practical experience are key, even with new tech. I think AI can quickly point out issues, but medical staff need to confirm results to keep their skills up. Using AI alongside routine skill checks helps medical staff stay skilled while also getting the upside of quick, precise diagnoses.

At Ocuco, we specialize in helping the eyecare industry streamline operations while maintaining high standards of patient care—so I understand both the potential and the challenges of integrating AI into medical workflows. Adaptable AI Systems Work closely with AI providers to make sure that the technology is built to change based on what the user wants. The AI should be able to change its suggestions and choices based on the clinician's experience and skill, so that the technology isn't relied on too much. To keep deskilling from happening, make sure the AI system can change based on the user's needs and level of skill. By giving the AI the ability to respond and change, it can give each clinician the right amount of help and guidance. One example is that the AI system could start by giving less experienced endoscopists more detailed suggestions and step-by-step instructions. As the doctor gets better at what they're doing and gains more experience, the AI might step back and offer more advanced insights and alerts instead of holding their hand. This would make sure that the clinician doesn't rely too much on the AI's decisions and that they keep improving and maintaining their core diagnostic skills. The AI system should also be able to tell when the clinician's performance changes over time. If the system sees that the rate of adenoma detection or other important metrics goes down when the clinician isn't using the AI, it could give the clinician specific feedback and suggestions to help them figure out where they need to improve their skills. This proactive approach to monitoring and adapting would help the clinician keep their skills up to date, even as they use AI technology more. We can get the most out of this game-changing technology while lowering the risks of deskilling by making AI systems that are adaptable, quick to learn, and focused on improving human skills rather than replacing them. The key is to find the best balance between human and machine intelligence.

This one study I am familiar of out of Poland is a wake-up call that AI in medicine isn't just about what machines can do, but what humans stop doing when they lean too heavily on them. The risk here is "de-skilling"—when doctors offload too much to AI, their own diagnostic instincts, pattern recognition, and vigilance can erode over time. The way forward isn't to ditch AI, no, but to structure its use so it augments rather than replaces human expertise. That could mean intentionally scheduling some "AI-off" procedures to keep endoscopists' skills sharp, running a two-pass protocol where the first review is manual and the second is with AI or vice versa depending on the case, and building training programs that mix AI-assisted and unassisted practice. AI systems themselves should be designed more like coaches than crutches like flagging areas of concern, but requiring the physician to make the final call. And just as importantly, continuous auditing of adenoma detection rates and diagnostic accuracy needs to be part of the workflow so any drop in human performance is caught early. In short, AI has real potential to reduce missed cancers, but if we're not careful, it could also erode the very expertise that makes colonoscopy safe and effective. The balance is to treat AI as a partner, not a replacement and to design safeguards that keep human skill at the center of care.

AI adoption should never disconnect doctors from patients emotionally. Medicine requires human connection beyond detection accuracy. Patients value compassion, reassurance, and context, which algorithms cannot provide. Physicians must remain communicators, not technicians behind screens. That responsibility protects both patient trust and care outcomes. AI should handle repetitive scanning tasks that exhaust focus, not patient communication or judgment. Freeing physicians from routine analysis creates more time for conversations. This deepens trust and improves adherence to treatment. AI efficiency should serve human connection, not diminish it. Compassion must remain central within advanced care.

When I see a headline about how AI could be diminishing the effectiveness of doctors during, for example, colonoscopies, my first reaction is: a predictable "over-reliance." AI is best utilized as a double check, not a replacement for clinical judgment. And the answer will lie in forming systems that aid rather than stand in for, with sharp reminders about where the final responsibility still lies. In reality, this may look like an AI tool simply flagging possible anomalies and needing a doctor to manually check them and confirm before moving forward. Explainability features need to be built in as well, which is why the AI flagged something would keep clinicians engaged rather than increasing the risk of switching their brains to autopilot. Training programs can also humanize AI by positioning it as a colleague — ideally one who has freed you from the majority of errors, rather than threatening your professional competence entirely. The bottom line is, use AI as an augmentation strategy and not as a shortcut. I have seen in my work on EV and AI, the good results come when humans and AI are elements of a dialogue that functions as both a check for what either could miss. Medicine is no different. Balanced carefully, AI can deliver more accurate and consistent results without destroying the tactile skill set that patients depend on.

Using AI to get a second opinion rather than as a first-line diagnostic seems like the solution if we want to preserve our own skills. My worry is that economic realities may make this difficult. If AI can do a good enough job of diagnosing potential cancers, providers looking to save money are going to use it, undercutting human expertise.

As a functional medicine practitioner who works extensively with cognitive decline through the Bredesen Protocol, I've seen how over-reliance on any single diagnostic tool creates cognitive laziness. The same neuroplasticity principles we use to reverse early Alzheimer's apply here--the "use it or lose it" concept affects medical skills just like memory formation. The solution isn't avoiding AI but implementing "cognitive load protocols" during procedures. I've found that having practitioners verbally explain their differential diagnosis before AI input maintains active pattern recognition. When we require doctors to articulate why they're concerned about a lesion's surface pattern or irregularity first, they stay engaged with the actual pathology rather than just following AI prompts. From my work optimizing brain function in high-performing professionals, I know that alternating between assisted and unassisted practice sessions prevents skill atrophy. The key is treating AI like training wheels--helpful for learning, but you need regular periods without them to maintain core competencies. Just like we rotate between different cognitive challenges in brain training, colonoscopists need structured "manual" sessions. The neuroinflammation research I work with shows that passive consumption versus active engagement literally changes brain structure. Medical education needs to apply this--AI should improve pattern recognition through active questioning rather than replace the critical thinking process entirely.

In addiction medicine, I've learned that technology works best when it improves human connection rather than replacing it. When we launched National Addiction Specialists' telemedicine platform, we finded that AI-assisted screening tools actually improved our patient assessments--but only when we used them to prompt deeper conversations, not make automatic diagnoses. The critical difference is using AI for pattern recognition while maintaining clinical intuition for decision-making. Our platform flags potential relapse indicators from patient responses, but I still conduct the full clinical interview to understand the context behind those red flags. This approach has helped us maintain a 78% treatment retention rate across Tennessee and Virginia. From my experience with ASAM's Health Technology Committee, the most successful AI implementations require mandatory "human override" protocols. We built our system so providers must document their clinical reasoning before accessing AI recommendations, forcing active engagement with each patient's unique presentation rather than algorithmic shortcuts. The key is treating AI like sophisticated lab work--it provides valuable data points, but the art of medicine happens in synthesizing that information with patient history, behavioral observations, and clinical experience. This is especially crucial in addiction medicine where underlying psychological factors often trump surface-level symptoms.

AI can be a helpful tool for doctors, not a replacement for their expertise. Take colonoscopies, for example—AI can highlight areas that might need a closer look, making sure nothing important is missed, especially in tricky cases. But it's important to keep the human touch in medical decisions. Doctors should use AI as a backup, combining its insights with their own knowledge and experience. This way, we can improve accuracy while keeping patient care personal and focused on the best outcomes. Clear rules for using AI and proper training can prevent over-reliance on the tech and help ensure it's used safely.

Good Day, AI what roles may it play safely and what's to be done to see that those roles are played out well? AI may be put to use in colonoscopy by way of it1 playing a supportive role that augments, rather than replaces, the doctor's diagnostic tools. For example, AI can serve as a second pair of eyes, which flags up possible polyps or lesions in real time, which the endoscopist can then look at through the lens of clinical experience. Also in this model the doctor maintains the primary role in diagnosis which in turn preserves their full participation in the process. Also how do we get doctors to not fall into the trap of over dependence on AI? In order to avoid over reliance and the break down of basic diagnostic skills, it is key that there is an ongoing professional development program. Doctors should be put through regular performance assessments, simulation based learning, and case studies without the use of AI to keep their skills sharp. Also it is a good idea to integrate AI into the training as a tool for feedback instead of a crutch which in turn will also support their autonomy. What are the health care wide strategies that will support the safe use of AI? Health care systems must put in place clinical review boards and standard operating procedures for the use of AI in endoscopy. Also they should track results with and without AI support which will help to identify issues of deskill at an early stage and allow for corrective action. Also policy should very much put out the that AI is to add value to -- not take the place of clinical decision making. Also in to this should go ethical issues, transparency in the design of the AI, and accountability structures to see to it that AI is integrated into patient care in a safe, effective and fair way. If you decide to use this quote, I'd love to stay connected! Feel free to reach me at gregorygasic@vmedx.com and outreach@vmedx.com.