Patients are increasingly using artificial intelligence tools and large language models (LLMs) as an additional source of medical information. They prepare for medical appointments, interpret test results, and explore potential treatment options with the help of AI. What was once the exclusive domain of medical professionals is becoming a daily tool for patients. However, experts stress that clear frameworks are needed to maximize benefits and protect patients from risks.
“Patients have discovered large language models and AI as a source of easily digestible medical information. They use them to prepare for appointments, interpret lab results, or look for new research and treatment directions. These tools often help translate complex medical jargon into simple, accessible language. Yet, they are not always used with full awareness of their limitations — for example, their tendency to present incorrect information convincingly. Moreover, such models cannot admit when they lack knowledge in a given field. That’s why critical verification of AI-generated answers is essential,” says Natalia Baranowska, Section Lead Data Advisory & Insights at Roche, in an interview with Newseria.
AI in medicine: growing adoption, mixed trust
The medical AI market is expanding rapidly. According to an Ipsos survey, 31% of consumers already use generative AI for health-related topics, and 38% express at least moderate trust in it. In the UK, around 20% of general practitioners use similar tools for documentation and diagnostic support (29% and 28%, respectively).
One of AI’s key advantages lies in its ability to “translate” complex medical terminology into plain language. Studies have shown that LLM-based tools consistently improve the readability of patient education materials without compromising accuracy. However, they also risk producing incorrect or contradictory content that may deviate from established medical knowledge.
“Language models are trained on specific datasets. The quality of their responses depends on how those datasets were constructed — whether they contain stereotypes or underrepresented groups. Ultimately, these are technologies based on mathematics and probability, not human understanding,” Baranowska explains.
Healthcare Datathon: testing AI safely and collaboratively
To address these challenges, new initiatives are emerging to define safe directions for AI development. One of them is Healthcare Datathon — an open event where doctors, patients, data scientists, and technology experts jointly test AI’s potential in healthcare.
“Bringing together such diverse participants — clinicians, specialists, patients, and researchers — allows us to truly explore the depth of AI and large language models. These systems have limitations: they can ‘hallucinate’ or give incorrect answers. A patient without medical knowledge cannot always recognize that. By gathering experts in one place, we can analyze where AI performs well, where additional review is required, and when it should not be used at all,” explains Baranowska.
Healthcare Datathon is the first event of its kind in Poland where teams work on real, anonymized clinical data. Such initiatives help assess both the potential and limitations of AI while aligning findings with clinical practice and patient needs. They also serve an educational purpose — teaching participants how to critically assess AI outputs and recognize its strengths and weaknesses.
Poland’s position: strong potential, uneven healthcare access
Experts note that Poland is well positioned to develop such initiatives. By 2026, up to $400 million could be directed toward digital health and AI technologies. The adoption of AI tools in hospitals is also accelerating — in 2024, 13.2% of medical facilities had already implemented AI, mainly in imaging diagnostics and patient data analysis.
However, access to healthcare remains uneven. Poland has only 3.4 doctors per 1,000 inhabitants, below the OECD average, with rural areas facing the greatest shortages. AI could play a key role in identifying workforce gaps, optimizing resource allocation, and helping patients understand complex medical information.
“Access to healthcare depends on many factors — where we live, our education level, and our financial means. AI and language models may help reduce these inequalities, at least to some extent,” says Baranowska.
AI as a “translator” for patients
The need for accessible medical communication is particularly pressing. According to the PIAAC international survey by Poland’s Educational Research Institute, nearly 40% of Polish adults struggle with reading comprehension. For many, medical texts filled with specialized terminology remain unintelligible without support.
Language models could serve as “translators”, simplifying medical terms and improving health literacy. However, experts warn that AI must operate within clearly defined frameworks that minimize the risk of errors and bias replication.
“Prevention is key in medicine — it’s always better to prevent disease than to treat it. If AI and data analytics can help people better understand how lifestyle, diet, and medical choices affect their health, we could see fewer illnesses overall. And when illness does occur, access to care could become faster and more effective,” emphasizes the Roche expert.
Regulation: from innovation to legal safeguards
The advancement of AI in healthcare requires not only innovation and collaboration, but also appropriate legal frameworks. Regulatory changes are already underway. The EU Artificial Intelligence Act (AI Act), effective since 2024, introduces strict requirements for high-risk systems, including medical AI.
In parallel, the European Health Data Space (EHDS), operational since March 2025, aims to facilitate secure data sharing and enable medical research across the EU.
Together, these initiatives mark the start of a new era — one where artificial intelligence supports medicine not as a replacement for professionals, but as a trusted assistant that empowers both doctors and patients.
