Feeling chest tightness or shortness of breath and your first thought is to enter these symptoms into an Artificial Intelligence, or AI, website to see what might be going on? Hold that thought.
Artificial intelligence has become the next evolution of “Dr. Google.” It is fast, accessible, and often sounds confident. But there is an important distinction that needs to be understood.
Information is not the same as diagnosis. AI is trained on patterns, not patients.
Be careful when asking AI for medical advice. It may seem like an easy tool to use, and although it has shown impressive capabilities, many AI systems have failed medical board style examinations and struggled with highly specialized or regional testing formats. Performance often declines when questions become:
• Multi layered and complex
• Image based, such as radiology or dermatology
• Subspecialty focused
• Region specific
• Dependent on clinical nuance
There are several reasons why AI should not be relied upon for personal medical decision making. One major issue is that instead of looking at the full clinical picture, it can miss red flag symptoms that require urgent care.
AI systems generate responses based on patterns in data. They do not examine a patient, review laboratory results in real time, or fully understand clinical nuance the way a regulated health professional does. They also do not independently verify context unless it is clearly provided. This means the output depends heavily on the information included in the prompt. If key details are missing, the answer may be incomplete or misleading.
Framing Effect
This concept comes from behavioural psychology. The framing effect describes how the way a question is worded influences the response or decision. For example, asking “Is this medication safe?” versus “What are the risks of this medication?” can shift the emphasis of the answer, even though the topic is the same.
Prompt Sensitivity
In AI, this is called prompt sensitivity. Small changes in wording can shift the tone, depth, or focus of a response. This is not bias in a moral sense. It is a structural feature of how language models function.
Confirmation Bias
On the user side, confirmation bias also plays a role. If someone asks, “Why is this supplement dangerous?” the system may highlight risks. If they ask, “Why is this supplement beneficial?” the system may highlight benefits. The user may unintentionally steer the direction of the answer toward what they already believe. Humans naturally seek information that supports existing assumptions. If someone suspects a serious condition, they may phrase questions in a way that directs the response toward that concern.
Information Asymmetry
In healthcare, patients often do not know which details are clinically relevant. If key factors such as age, comorbidities, pregnancy status, current medications, or red flag symptoms are omitted, the output cannot be fully accurate. This is one reason AI should never replace clinical assessment.
The deeper issue in medicine is that clinical decision making depends on context, probabilities, and risk stratification. AI can provide general educational information, but it does not hold responsibility, cannot examine the patient, and does not practise within a regulated scope.
AI systems generate responses based on patterns learned from large volumes of text. They can summarize research, explain medical concepts, and provide general educational information, but not with guaranteed accuracy. How many times has AI produced research articles that do not exist?
What AI cannot do is:
• Interpret subtle non verbal cues
• Review your complete medical record
• Assess real time laboratory values
• Apply clinical judgement within a regulated scope of practice
Medicine is rarely black and white. Two people with identical symptoms may require completely different assessments depending on age, medical history, medications, pregnancy status, menstrual cycle, and individual risk factors.
AI does not see the full picture unless every relevant detail is provided. Even then, it cannot independently identify what is missing. A healthcare professional is trained to ask targeted follow up questions. AI can only respond to the information it receives. Missing one key detail can significantly change clinical interpretation.
Where AI Can Be Helpful
AI can still play a positive role in healthcare when used appropriately. It can help:
• Explain medical terminology
• Summarize general information
• Provide educational overviews
• Generate thoughtful questions to ask your provider
• Support health literacy
It should not replace individualized clinical assessment.
When to Seek Immediate Care
Certain symptoms should never be evaluated solely online. Red flag symptoms include:
• Chest pain
• Shortness of breath
• Sudden weakness
• Confusion
• Severe allergic reactions
• Uncontrolled bleeding
These require urgent medical evaluation.
Technology is a tool, not a clinician. Artificial intelligence is powerful, but it does not carry legal responsibility, clinical accountability, or regulated professional oversight. Clinical care involves judgement, experience, ethics, and human interaction. It considers probabilities, risks, and patient specific factors that extend far beyond text based pattern recognition.
AI can inform. It cannot diagnose.
As healthcare continues to evolve, the goal is not to reject technology. It is to understand its limits. Use AI to learn. Use healthcare professionals to decide. This blog is for educational purposes only and does not provide medical advice.
About the author:
Hi, I’m Abinaa, a fourth-year naturopathic medical student at the Canadian College of Naturopathic Medicine with a deep-rooted passion for natural healing, inspired by my South Asian upbringing. Through this blog, I hope to share my journey, explore topics in holistic health and wellness, and offer simple, thoughtful insights that support a more balanced and mindful way of living.
