Possibility of errors grows with complexity of queries, model used, phrasing:Educator
PETALING JAYA: Artificial intelligence (AI) is increasingly being used by the public for health advice but experts say it cannot replace doctors and carries serious risks if relied on blindly.
Universiti Malaya Computer Systems and Technology Department Prof Dr Ainuddin Wahid Abdul Wahab said AI tools, such as ChatGPT, are fundamentally limited because they rely on statistical patterns rather than genuine clinical understanding.
“AI is essentially a language tool, not a doctor. It predicts which words should come next based on patterns in the data. This means it could sound confident even when it is wrong,” he said.
Unlike human doctors, AI cannot examine patients or interpret subtle signs, such as skin tone, breathing patterns or other non-verbal cues.
“It is like using a highly advanced dictionary to interpret a poem. AI can define every word but it may miss the deeper meaning, in this case, the actual health condition.”
He said AI could produce plausible-sounding but inaccurate medical information.
“The system is designed to be helpful and conversational, not strictly factual. It is similar to asking a very good writer to fix a car engine. The words may sound right, but without hands-on checks, mistakes are easy and potentially dangerous.”
The risk of errors grows with the complexity of medical queries, the model used and how questions are phrased.
“While AI could reliably explain general medical concepts, it may mislead in complicated cases. Errors usually stem from biased or incomplete training data.
“If the data lacks information on certain demographics or rare conditions, the AI’s advice may not apply to every patient.”
Ainuddin said without human oversight, such gaps could lead to harmful or even fatal conclusions.
Ethical concerns also arise when AI provides medical guidance directly to the public, he added.
Safety is a major issue because chatbots may fail to recognise emergencies or serious conditions, delaying critical care.
Accountability is also unclear.
“If AI advice causes harm, it is uncertain whether responsibility falls on the user, the developer or the platform.”
Data privacy and fairness are also at stake.
He said patients may share sensitive information without knowing how it is used and AI could unintentionally provide lower-quality guidance to certain groups because of biases in its training data.
“It is like replacing a physical courtroom with an automated system. The speed and accessibility may improve, but without human judgement, fairness and safety are compromised,” he added.
Ainuddin stressed that certain aspects of healthcare must remain in human hands.
“Tasks requiring complex judgement, emotional support and physical intervention, such as surgery or delivering difficult news, cannot be delegated to AI.”
Humans can understand a patient’s context, values and needs in a way AI cannot replicate.
However, AI could serve as a powerful support tool as it excels at processing large volumes of data, recognising patterns, summarising medical records, highlighting drug interactions or flagging areas of concern in imaging scans.
“Think of AI as a high-powered microscope. It allows doctors to see what is invisible to the naked eye, however the microscope cannot make treatment decisions. The doctor remains the ultimate decision-maker,” he noted.
He warned against using AI for self-diagnosis or treatment as misinterpretation may cause unnecessary anxiety or a false sense of security, delaying proper medical attention.
“AI should be treated as a supplementary tool. Use it to generate questions for discussion with healthcare professionals and always verify information with trusted, peer-reviewed sources.
“AI has enormous potential to enhance healthcare by supporting doctors and improving access to information. But it should never replace the human judgement and expertise essential for safe, accurate and ethical medical care.”
