The Future of AI in Healthcare

By Wayne Persky

RSS Feed

More and more people are turning to AI for answers to questions about their health problems, rather than going to their doctors.

According to a survey by the Annenberg Public Policy Center of the University of Pennsylvania (April 2025, involving about 1,600 adults) (Staff, 2025, July 14):1
​

• 79% said they’re likely to go online for a specific question about health symptoms or conditions
• ​Among those searching online, nearly two-thirds reported that they had seen “AI-generated” summary responses at the top of search results.\
• 63% said that those AI-generated responses were “somewhat” or “very” reliable.
• 12% said they “always or often” talk with their doctor about health information they found online.

Another report from a wellness trend survey, discussed in the New York Post, found that “more than 1 in 3 Americans (35%) are using AI to learn about and manage aspects of their health and wellness” — including researching conditions or meal planning (Beaner, 2025, July 24).2 For example, according to the article, 31% used AI to get insight into specific medical conditions/issues.

But people don't want their doctors to rely on AI.

On the flip side, a broader attitude survey by the Pew Research Center in February of 2023 found that 60% of U.S. adults would feel uncomfortable if their own health-care provider relied on AI to make diagnoses or recommend treatments.

Academic evidence also shows a risk. For example, one recent paper found that when users evaluate health advice from AI, they may over-trust it and treat it as equivalent to a doctor’s advice, even when accuracy is low.

So it's safe to say that somewhere between 30-40% of adults (at least in more tech-enabled populations like the U.S.) are already using AI tools for health-information or health-management tasks. True “replacing the doctor” is much less common — the data suggest more “supplementing online information” rather than full replacement of a clinician.

What's likely to happen in the future?

Given the trends in technology, user-behaviour, regulation, and health-care delivery, several scenarios seem likely:

1. AI will probably become a more common first choice.

• Instead of calling the doctor or scheduling an appointment, more people will likely first ask an AI-driven chatbot or digital assistant for symptom checks, condition explanations, diet or lifestyle recommendations.
• This means more information gathering and assessment of the health risks at home — which could reduce unnecessary doctor visits, but this also increases the risk of delaying needed care if the AI is wrong, or misinterprets the medical literature.​

2. AI will probably be integrated with clinical care, and hybrid models will evolve.

• The doctor-patient-AI relationship is likely to become more integrated rather than purely competitive. For example, AI tools may help patients prepare questions for their doctor visit (“What should I ask?”) or help doctors by summarising data, offering differential diagnoses, flagging concerns.
• One result might be that clinicians may see more patients who have already used AI to self-diagnose, bringing in those ideas — this will change the nature of office visits (more informed / more questions / maybe more skepticism).

​
3. This may improve accessibility, but increase regulatory and ethical complexity.

• For people with limited access to health-care services (rural, underserved, cost-barriers), AI could help bridge the gap in initial information and guidance.
• But this raises important issues of accuracy, safety, bias (AI trained on certain populations may under-serve others), liability (who is responsible if AI gives bad advice?), data privacy, and patient trust.

4. A potential shift in healthcare-consumer behavior and business models will be developed.

• Health-care delivery models may evolve: more remote or asynchronous consultations, more “digital front door” use of AI for preliminary screening, fewer in-person visits for routine issues.
• ​Insurers and health systems may encourage use of AI/self-service tools to reduce cost (for example, encouraging symptom checkers before seeing a specialist).
• On the flip side, medical practices may need to adapt to more informed/empowered patients who come in bearing AI-generated advice — and clinicians will need to spend time validating, refuting, or integrating that input.

There will be risks — particularly for complex or chronic conditions.

For people with chronic, complex, or serious conditions (for example. autoimmune/inflammatory bowel diseases such as microscopic colitis (MC), or thyroid disorders), the risk of relying solely on AI is higher — comorbidities, medication interactions, and personalised history, for example, may be overlooked. That said, in all fairness, it should be noted that most of those issues are often overlooked in the current healthcare environment.

• AI may lead to “false confidence” in people that they don’t need to see a doctor, or delay serious care.
• Over-trust in AI could widen disparities if tools are less accurate in certain populations, or if vulnerable patients substitute AI for in-person care where they need it.​
• Patient-provider relationship will shift.
  
• The role of the human provider may shift more toward interpretation, validation, shared decision‐making, emotional support, rather than just “giving information.”
  
• Providers may need to address patients’ AI-derived beliefs, correct misinformation, and integrate digital tool outputs into care plans.
  
• At the same time, there may be a devaluation of routine information giving tasks by clinicians (if AI handles them), freeing clinicians for higher-order tasks, such as complex reasoning, empathy, and patient education.

How is this likely to affect MC Patients?

Many patients with chronic GI or autoimmune issues may use AI tools to ask symptom-based questions (“Why do I still have diarrhea?”, “Is my diet okay?”, “What antibiotic risk do I have?”) — which means more people will arrive at forums, chats or doctor visits already armed with AI-generated thoughts or conclusions.

• For chronic conditions, personalised history, individual disease characteristics, comorbidities matter a lot — so the risk of missing something may possibly be higher if someone uses AI instead of consulting a specialist.
  
• Guidance on how to responsibly use AI health-tools (what to ask, what to validate, when to see a clinician) may become much more important.
  
• Discussion and support forums, websites, and apps may need to be designed to include guidelines that can assist users in properly using AI for healthcare purposes.
  
• And conversely, users may come to forums, websites, and apps expecting to find such guidelines for properly using AI for healthcare purposes.

​But the foregoing analysis assumes that healthcare professionals will remain more knowledgeable and more reliable than AI. And most importantly, the analysis assumes that healthcare professionals currently provide adequate explanations and information to patients regarding health issues. Unfortunately, in many cases, including MC, that simply is not the case. Additionally, many clinicians have personal biases that affect their viewpoints regarding healthcare.

What if AI truly knows more about health than the average doctor?

Up to now, most discussions assume that AI is an assistant, and the doctor is the authority.
But if AI eventually becomes more accurate, more comprehensive, more up-to-date, and more diagnostically reliable than the average doctor, the entire structure of healthcare will change so that AI becomes the primary source of medical information.

If AI becomes consistently:
​

• more accurate than humans,
• more current with medical research,
• better at integrating massive amounts of data,
• able to avoid cognitive biases that physicians routinely have,

then AI will naturally become the first choice for health information — not because people prefer technology, but because it will outperform physicians. This would make AI‐driven self-assessment the new default, and doctor visits would become secondary, confirmational, or reserved for cases requiring physical procedures.
​
Clinicians would shift from diagnostic experts to human-care specialists.
A glimpse into the way most hospitals operate reveals that this is already happening, to some extent. GPs have less diagnostic and prescribing authority, and patients are sent to specialists to diagnose many health issues that previously were diagnosed by GPs.

If AI becomes the superior diagnostician:

This would eliminate the need for a physician's presence in at least 40 to 70% of office visits. Obviously, hands-on work such as surgery and endoscopy cannot be replaced by AI (at least, not in the immediate future). But wait times would be reduced, and healthcare costs would drop dramatically, because (for example), most of the time consuming record-keeping requirements would be handled by AI. Clinicians would need to transition into interpreters and counselors, to discuss various treatment options, accommodate patient preferences, fears, social issues and any concepts that AI can't fully handle.

Medical training would need to be redesigned so that physicians would be skilled at:

• Critical thinking with AI tools
• Ethics
• Interpretation
• Communication
• Procedural skill
• Crisis management

Physicians would shift from being a walking textbook, into an AI enhanced medical operator.

What's wrong with this picture?

The attributes listed above are currently weak areas for most physicians. All of them are currently deemphasized by the healthcare system. So these changes would require a paradigm shift. Will physicians be willing to make those dramatic changes? That's not likely for older, well-established doctors. Surely the changes will need to occur gradually, as med school graduates evolve.​

Physicians will probably resist these changes, but patients will demand them if the changes result in improved healthcare at reduced cost.

Liability will shift.

Right now, if a doctor follows AI and AI is wrong → the doctor is blamed.
If AI proves more accurate than the average doctor, the opposite will happen: it will become malpractice not to use AI.

Physicians struggle most with:

• Chronic diseases
• Multi-system interactions
• Diet-gut autoimmune connections
• Medical interactions
  
• ​Nutrient deficiencies

But AI can handle all these easily, and virtually instantaneously.

Healthcare will become decentralized and patient driven.

Currently, medical knowledge flows from medical schools through doctors to patients. In an AI dominated healthcare system, knowledge will flow directly to patients. Rather than dictating healthcare, doctors of the future will become "assistants" in the healthcare system.

While that sounds somewhat ominous, it's actually already the case. How doctors are currently able to perform their duties is dictated by the government, insurance companies, pharmaceutical companies, attorneys, hospital administrators, and state regulators. For doctors, choosing to think "out of the box" is a risky venture.

But an AI takeover can only go so far, because:

AI cannot, for example, palpate an abdomen. Nor can it handle various real-time physical observations encountered in various diagnostic procedures. So there will always be a need for physicians during examinations, endoscopies, surgeries, managing emergencies, and various other procedures.

AI has already surpassed human doctors.

in a few narrow medical tasks, such as:.

• Radiology image detection (CT, MRI, X-ray) — AI can detect certain cancers, fractures, and lung abnormalities as accurately or more accurately than top specialists in controlled settings.
  
• Dermatology image classification — AI beats most dermatologists at identifying melanoma from high-resolution photos.
  
• Retinopathy screening — AI systems (like Google/DeepMind’s) outperform ophthalmologists in diabetic eye disease detection.
  
• Pathology slide interpretation — AI can spot rare cell changes faster and sometimes more accurately than humans.
  
  

But these are relatively narrow laboratory tasks — not full clinical reasoning.

But AI is still far behind

In the core skills of medicine, such as:

• Clinical judgment
  
• Integrating symptoms, history, comorbidities, medications, psychology, social factors, risks, contradictions, and incomplete information
  
• Understanding subtleties and uncertainty
  
• Real patients rarely match textbook patterns
  
• Ethical decision-making
  
• Balancing risks, patient values, and long-term outcomes
  
• Communication, empathy, trust, reassurance

​A huge portion of healing is relational and emotional. But note that GPs typically possess those attributes, while many specialists do not, and the healthcare system appears to be phasing out GPs, which implies that it's phasing out relational and emotional attributes.

So when will AI surpass doctors overall?

Experts in medical AI tend to concur on three timelines:

1. Narrow medical tasks: 2025–2030 (already partly achieved)

• Image interpretation
• Lab value prediction
• Pattern recognition
• Summarizing medical records
• Medication interaction checking

AI will be consistently better than average doctors in these narrow tasks. But this is not true general doctor-level performance.

2. Broad diagnostic reasoning: 2035–2045 (optimistic)
This refers to what primary-care physicians or gastroenterologists do:

• Evaluating symptoms
• Building and ranking differential diagnoses
• Integrating medical history, labs, imaging, risk probability
• Choosing appropriate next steps

Some AI models are getting surprisingly close in controlled tests, but real-world generalization remains extremely difficult.

3. Full-spectrum medical practice: Not foreseeable yet (maybe never)
This includes:

• Performing surgery
• Managing emotional or anxious patients
• Handling atypical presentations
• Managing risks with incomplete information
• Taking responsibility for outcomes
• Interpreting subtle human cues

But a more likely outcome is that instead of replacing doctors,

​AI will:

• Prepare diagnostic summaries
• Catch rare conditions
• Reduce medical errors
• Flag dangerous interactions
• Provide detailed patient explanations
• Assist with chronic disease management
• Help underserved populations

And doctors will:

• Make final decisions
• Provide empathy
• Integrate social and psychological issues
• Continue to be responsible for diagnoses
• Handle medical complexity
• Perform procedures
• Manage uncertainty

So that AI becomes a copilot, rather than a replacement.

Could AI realistically diagnose microscopic colitis better than doctors?

​The answer is, "Yes, eventually.” In fact, MC is one of the conditions most likely to benefit from AI -assisted diagnosis — but that will only be possible after additional progress is made.

Reasons why AI could diagnose MC better than many doctors include:

1. Many doctors simply miss MC:

• Studies show that up to 70% of MC cases are initially misdiagnosed as IBS, viral gastroenteritis, functional diarrhea, SIBO, bile acid malabsorption (BAM), or anxiety disorders.
• Some patients go years without a correct diagnosis.
• Many doctors still believe MC is rare, despite data showing it is as common as Crohn's in older adults.
• AI has no preconceived biases and would not dismiss a 50–80-year-old woman with chronic diarrhea as “IBS.”

2. AI can detect symptom patterns doctors routinely overlook.
Most physicians don’t know the typical patterns associated with MC:

• Onset after a PPI, NSAID, SSRI, or antibiotic
• Nocturnal diarrhea
• Weight stability
• Normal colonoscopy
• Sudden onset after a viral infection
• Symptoms worse with bile acids, fat, or magnesium
• Flares after vaccinations, anesthesia, or GI infections
• Association with autoimmune thyroid disease (especially Hashimoto’s)
• Increased risk after age 50
• Strong response to dietary elimination

AI can learn these patterns from large datasets, medical literature, and patient histories. Doctors rarely integrate all of them.

3. AI could flag missing biopsies or inadequate sampling:
A huge share of missed diagnoses result from:

• Random colonoscopies without biopsies
• Not taking enough biopsies — (Minimum required: 6–8 specimens from different colon segments)

AI reviewing procedural notes and pathology reports could automatically detect:

• “No biopsies taken” carries a high risk of missed MC

AI can follow the guidelines perfectly, every time — something human clinicians often fail to do.

4. AI can analyze pathology slides more sensitively:

• Microscopic colitis is diagnosed by histology, but real-world variability is massive:
• Pathologists often disagree (interobserver variability is high).
• Mild cases are frequently missed.
• Collagen band thickness can be borderline.
• Lymphocyte counts can be miscounted or overlooked.

AI vision models are already excellent at detecting subtle histologic changes in:

• Crohn’s disease
• Ulcerative colitis
• Celiac disease
• Barrett’s esophagus
• Chronic gastritis

So applying similar models to:

• Lymphocyte density
• Collagen band thickness
  Surface epithelial damage
• Intraepithelial lymphocytosis​

would likely yield more consistent and earlier detection than many general pathologists provide today.

But first, high-quality MC datasets must be available for training AI.

Unlike Crohn's and ulcerative colitis (UC) MC datasets are still small, and AI training requires thousands of biopsy slides, clinical histories, medication lists, and follow-up outcomes. Furthermore, pathologists don't have standard labeling policies regarding specifying collagenous colitis versus lymphocytic colitis.

There are AI tools for:

• Colorectal cancer
• Polyp detection
• Celiac disease
• Gastritis classification

But precious few exist for MC, because of it's microscopic nature, and analysis requires histology training, not endoscopy. And, of course, gastroenterologists would still need to supply the biopsy samples.

Looking ahead, curating adequate data sets will surely take at least 3 to 7 years, suggesting that for AI to be able to gain superior diagnostic capability for MC, at least 10 to 15 years would be required.

The Bottom Line:

AI is already changing how patients seek health information, with roughly one-third of Americans using it for medical guidance—yet most remain uncomfortable with doctors relying on it. This paradox won't last. As AI demonstrates superior accuracy in diagnostic tasks, exceeds physicians' ability to integrate complex data, and avoids the cognitive biases that plague clinical decision-making, it will naturally become the primary source of medical information within the next 10-20 years.​

For MC patients specifically, this shift offers particular promise: AI could eliminate the misdiagnoses, overlooked patterns, and inadequate biopsies that currently plague 70% of cases. The future won't see AI replacing doctors entirely, but rather transforming them from walking textbooks into interpreters, counselors, and procedural specialists — a shift that may finally give chronic disease patients like those with MC the thorough, unbiased, data-driven care they've long deserved. The question isn't whether this will happen, but whether the medical profession will adapt gracefully or resist until patients demand change.

References:

1. Staff. (2025, July 14). Many in U.S. Consider AI-Generated Health Information Useful and Reliable. Annenberg Public Policy Center, University Of Pennsylvania, Retrieved from https://www.annenbergpublicpolicycenter.org/many-in-u-s-consider-ai-generated-health-information-useful-and-reliable/

2. Beaner, L. (2025, July 24). More than one in 3 Americans are using AI to manage their health, according to new survey. New York Post, Retrieved from https://nypost.com/2025/07/24/tech/more-than-1-in-3-americans-are-using-ai-to-manage-their-health-according-to-new-survey/