We repeatedly hear and see the slogan “AI will take doctors’ jobs.” This anxiety has its roots. In 2016, Geoffrey Hinton – one of the “fathers” of modern artificial intelligence – suggested that within a few years neural networks would outperform radiologists, so training new specialists would make little sense.
Nearly 10 years have passed. AI in medicine is real – it exists not only on conference slides. Radiology is, in fact, one of the most “saturated” areas: in Europe, more than 100 commercial, certified AI tools for medical imaging are already available, and both their number and the quality of supporting evidence grow year by year.
And yet, despite this:
- the number of radiologists has not decreased – on the contrary, demand for them is rising;
- reviews of studies from recent years show that newer solutions more often increase radiologists’ workload than reduce it.
So what is actually happening?
- The system orders more examinations because they are cheaper and faster.
- AI generates a new type of work: verifying alerts, handling false positives, and supervising system performance.
- The importance of tasks untouched by algorithms is growing: protocol selection, correlation with clinical history, participation in multidisciplinary case conferences.
This is not an exception. From the perspective of technological history, we much more often see the following scenario: technology lowers the unit cost of a task → we do more of it → the work does not disappear; instead, its structure and volume change. Before returning to medicine, it is worth looking at the experience of other industries.
1. Lessons from other sectors
ATMs and bank tellers
ATMs were supposed to symbolize “the end of bank tellers.” Yet analyses by James Bessen show that after the widespread rollout of ATMs in the United States:
- branch operating costs fell,
- banks began opening more branches,
- the number of tellers increased, but their role shifted from dispensing cash toward advising customers, sales, and handling more complex cases.
Technology did indeed “take away” some repetitive tasks, but the profession itself evolved.
Excel and accountants
A similar story played out with Excel. Spreadsheets were meant to “eliminate accountants.” In practice, they:
- radically reduced the cost of producing reports and analyses,
- led companies to expect increasingly sophisticated models, forecasts, and scenario analyses,
- shifted accountants and analysts from simple data entry toward analysis and business advisory roles.
Less time spent on tedious calculations, more on interpretation – but no overall loss of work.
Industrial robots
Industry has been undergoing robotization for years. A large study across 16 European countries showed that the introduction of robots:
- did not lead to mass unemployment,
- had a slightly positive effect on employment,
- particularly in low- and middle-wage countries.
Tasks changed. Some activities were taken over by robots, while new roles emerged: operators, technicians, and process engineers.
The Jevons paradox: the more efficient, the… more
As early as the 19th century, William Stanley Jevons observed that more efficient steam engines did not reduce coal consumption in Great Britain. On the contrary, they led to a dramatic increase in consumption, because energy became cheaper and was used everywhere.
This Jevons paradox is highly relevant to AI today: if a single report, description, or consultation becomes cheaper, the system usually orders… more reports, descriptions, and consultations.
With robotization, automation, and now AI, occupations tied to intellectual but routine work are often the most exposed. Algorithms can replace or facilitate various calculative and process-oriented tasks: data processing, information handling, memorization and aggregation, authentication, and search.
2. How technology changed physicians’ work before generative AI
Telemedicine and e-health
Telemedicine was intended to:
- improve access,
- shorten waiting times,
- relieve primary and outpatient specialist care.
However, a systematic review of eHealth in European primary care paints a mixed picture: some solutions genuinely improve efficiency, while others increase workload – among other things by adding communication channels and requiring parallel use of multiple systems.
Studies on inboxes in EHR systems (mainly from the U.S., but with a universal mechanism) show that:
- an increase in messages from patients and staff is strongly correlated with a higher risk of burnout,
- physicians with the highest message volumes had up to several times higher burnout risk compared with those with the fewest messages.
Alongside traditional visits, a new stream of work emerged: patient portal replies, chats, teleconsultations – often without a designated slot in the schedule, happening “in between” or after hours.
A crucial conclusion should be stated plainly: with telemedicine, physicians took on not only a new category of tasks, but also a greater burden of responsibility.
It is the physician who:
- decides whether a case can be safely handled remotely,
- is responsible for documentation completeness,
- initially chose tools independently (apps, messengers) or even called patients from private phones.
The system rarely provides a vetted list of solutions or clear standards that would relieve individual physicians from having to assess the legal, technical, and data security aspects of every new tool on their own.
Electronic medical records (EMR/EHR)
Electronic documentation was supposed to free physicians from paper. In practice, studies show that:
- for every hour of direct patient contact, there may be up to two hours of system and desk work,
- many physicians spend 1–2 hours daily on documentation after clinic or hospital hours,
- intensive EHR use is a major contributor to burnout.
This is a good example of how technology implemented without genuine co-design with staff can become an additional burden rather than support. Does that mean it should be avoided or abandoned? No – it simply means solutions must be introduced thoughtfully and adapted based on feedback.
3. New technologies in medicine: robots and AI notes
Robots in hospitals – operating rooms and logistics
In healthcare, robots appear mainly in two areas:
Operating rooms – robotic surgery systems.
Current robots (e.g., da Vinci systems) are fully controlled by surgeons; they are not designed as autonomous replacements, but as more precise tools.
Studies from European hospitals show a mix of clinical benefits and increased workload and stress during implementation: training, longer procedures, and new organizational duties.
Logistics – transport robots (e.g., Moxi).
In many hospitals in Europe and the U.S., robots transport samples, medications, and PPE, reducing the number of corridor trips made by nurses.
This genuinely alleviates workload – but again, new tasks emerge: coordination, operation, and maintenance.
Once more, we see the pattern: robots do not eliminate roles; they change the task mix.
AI as an assistant
From the workload perspective, the most interesting – and theoretically easiest to implement – solutions are so-called ambient AI scribes:
- the visit is recorded (e.g., via a phone) with patient consent,
- the AI system generates a draft note, recommendations, sometimes also coding suggestions,
- the physician becomes an editor rather than a “typing machine.”
Early studies, including randomized trials, show that such solutions:
- shorten documentation time,
- reduce after-hours documentation,
- are associated with lower burnout indicators.
This is one of the few tool categories where data show that well-designed AI gives physicians valuable time back instead of adding another panel to click through.
4. Data from Poland and Europe: how clinicians view AI
Polish physicians: readiness exists, frameworks are missing
In 2025, the study “Physicians and AI in healthcare” covering 86 Polish physicians was published. Results:
- 68% declared readiness to implement AI tools in their practice,
- but strongly emphasized the need for tailored training, clear implementation guidelines, and transparent rules of responsibility.
Studies among Polish medical students show a similar picture: high curiosity and awareness of potential, alongside concerns about liability, errors, and the impact of AI on clinical education.
European healthcare workers: they want AI, but do not trust current tools
According to the Corti & YouGov “First AId Report” (HCPs from Denmark, France, Germany, the UK, and the U.S.):
- 74% of European healthcare workers support using AI for at least one workplace challenge,
- yet 52% do not feel confident using currently available solutions,
- most frequently expected AI benefits include administrative time savings (42%), automated visit notes (29%), clinical decision support (32%), and diagnostic support (25%).
Statista data for Europe show that:
- nearly 60% of HCPs would feel confident using “new technology” in general,
- but only about 32% declare a similar level of confidence specifically toward AI.
A major barrier is therefore lack of trust and lack of legal frameworks, recommendations, or standards – one reason why so many projects remain stuck in “perpetual pilots.”
Polish workers overall: low usage, large skills gap
From PwC’s “Global Workforce Hopes and Fears 2025” and its Polish analysis:
- 9% of Polish workers use generative AI daily at work,
- 45% have used GenAI at least once in the past 12 months,
- globally, 32% of workers report regular (daily / several times per week) AI use at work.
At the same time, most regular users in Poland report improved productivity and work quality thanks to AI. The issue is therefore not lack of potential, but rather an implementation gap and a rapidly growing need for new skills.
“Shadow AI”: using AI outside official systems
Even where organizations officially “do not use AI,” the data suggest otherwise.
In employee surveys (e.g., Germany, UK, U.S.):
- about 75% of workers use some form of AI at work,
- about half are “shadow AI” users – tools not approved by the organization and used without IT consent,
- 59–75% admit to pasting sensitive data (client data, employee data, internal documents).
In healthcare, this translates into:
- “silent” AI use,
- risk of uncontrolled data leaks,
- inconsistency between official policy and reality.
A good European example is a recent study by the Nuffield Trust and the Royal College of GPs: nearly 30% of UK GPs admitted to using AI tools (including ChatGPT) during consultations, most often for visit summaries, diagnostic support, and administrative tasks.
5. What this means for physicians and healthcare workers
AI will not take doctors’ jobs – but it will change how they work
Considering:
- experiences from banking, accounting, and industry,
- data from radiology and robotics,
- studies of physician attitudes in Poland and Europe,
one can conclude that AI will not eliminate entire specialties or groups of healthcare workers. Instead, it will automate specific tasks: parts of documentation, triage, preliminary image analysis, information retrieval, discharge drafts. Some of these will reduce unit costs, which will likely lead the system to order more of them (more tests, more patient contact, more reports). New competencies will therefore be crucial, making it essential to support healthcare professionals in adapting to new technological possibilities.
The greatest risk: AI implemented “for doctors, but without doctors”
The history of EHRs and, to some extent, telemedicine shows what happens when technology is designed primarily for billing and reporting, focused on financial metrics, with minimal involvement of those working directly with patients.
A system that only formally “automates” processes often adds documentation burdens and shifts risk and responsibility onto physicians and other medical professionals.
With AI, the stakes are even higher, because tools not only organize data but also propose diagnoses and decisions. If the medical community does not assume the role of co-authors, safety standards and usage scopes will be defined instead by systems, vendors, payers, and finance departments.
Rather than pretending “we don’t use AI here,” it seems reasonable to:
- clearly define what must not be entered into generative tools (identifiers, medical data, employee data),
- specify approved tools and use cases,
- develop ways to document situations where AI influenced a clinical decision.
System-level support and vetted tool lists
It is unrealistic to expect every physician to be an expert in technological risk analysis. What is needed:
- national or at least institutional lists of recommended, vetted solutions (AI, telemedicine, patient apps),
- procedures for evaluating new tools (clinical, legal, IT security) at the hospital, regional, or national level, instead of shifting this burden onto individual physicians,
- regular training covering typical risks (model hallucinations, bias, privacy), boundaries of tool responsibility versus physician responsibility, and safe AI use for both patients and clinicians.
In such a model, physicians do not need to know every technical detail – they operate within a trusted catalog of verified tools.
Summary
The question “Will AI take doctors’ jobs?” is understandable, but based on past experience, it may not be the most relevant one. In healthcare, an aging population and growing medical complexity will generate ever greater challenges. Even if AI takes over some tasks, the system will quickly fill that space with new patient and provider needs.
More appropriate questions are:
- Which parts of our work do we want to delegate to AI, and under what conditions?
- Will we use AI to reclaim time for patients and ourselves, or allow it to become yet another layer of bureaucracy?
Change and new technologies always bring uncertainty. It is natural for physicians to react to AI with a mix of curiosity and concern. But one thing seems clear: if physicians and other healthcare workers do not proactively engage in shaping the rules for AI use, the system will do it for them. And then, in a few years, the shape of our work will be decided elsewhere. From the perspective of patient safety and our own professional security, it seems better to be among those who co-create new tools than among those who later have to adapt to them.
References
- Kowalewska E. i in. Physicians and AI in healthcare: insights from a mixed-methods study in Poland (2025).
- Ratajczak P. i in. Perceptions of AI-based tools among Polish medical university students (BMC Med Educ, 2025).
- Corti & YouGov. First AId Report: AI’s impact on healthcare (Europa i USA, 2025).
- Statista. Artificial intelligence (AI) in healthcare – trust and confidence among European HCPs (2024).
- PwC. Global Workforce Hopes and Fears 2025 oraz komunikaty dla Polski (2025).
- Keuper J. i in. The impact of eHealth use on general practice workload in the pre-COVID-19 era: a systematic review (BMC Health Serv Res, 2024).
- Keuper J. i wsp. Use of E-Health in Dutch General Practice during the COVID-19 Pandemic. IJERPH, 2021.
- Mandal S. i in. Quantifying the impact of telemedicine and patient medical advice requests on physician work (npj Digit Med, 2024).
- Akbar F. i in. Physician Stress During Electronic Health Record Inbox Work (2021).
- Kwee T.C., Kwee R.M. Workload of diagnostic radiologists in the foreseeable future and role of AI (Insights Imaging 2021; aktualizacja 2025).
- van Leeuwen K.G. i in. Artificial intelligence in radiology: 100 commercially available CE-marked AI solutions (Eur Radiol, 2021) oraz aktualizacje (Antonissen N. 2025).
- Liu H.Y., Hayton J. Expectation vs reality: impact of the da Vinci surgical robot on healthcare professionals’ work experiences (Soc Sci Med, 2025).
- Artykuły nt. robotów szpitalnych (np. Moxi), m.in. Diligent Robotics, doniesienia z europejskich i amerykańskich szpitali.
- Microsoft (2025), Rise in ‘Shadow AI’ tools raising security concerns for UK organisations, Microsoft UK Stories. https://ukstories.microsoft.com/features/rise-in-shadow-ai-tools-raising-security-concerns-for-uk/
- Nuffield Trust & Royal College of GPs. AI in UK primary care (2025).
- UpGuard. The State of Shadow AI – Trends, Insights & Statistics / State of Shadow AI report, 2025
- Olson K.D. i wsp. Use of Ambient AI Scribes to Reduce Administrative Burden and Professional Burnout. JAMA Network Open, 2025;8(10):e2534976.
- Ivanti (2025), 2025 Technology at Work Report. Reshaping Flexible Work, raport badawczy, Ivanti. https://www.ivanti.com/resources/research-reports/tech-at-work
Author:
Małgorzata Maj – Innovation Manager at the Institute of Mother and Child. Graduate of medicine at the Medical University of Lublin and law at the University of Warsaw. She gained professional experience at one of the largest Warsaw law firms specializing in Life Sciences. In her work, she seeks to combine legal expertise with medical knowledge. Co-author of publications and reports on medical innovation, artificial intelligence, and telemedicine.