AI Health Advice Risks - Humans and AI Working together may Produce Terrible Health Advice

1 day ago
2 min read

AI Health Advice Risks - If an AI provides poor health advice… maybe you are the problem?

🤓 This is interesting.

A study published in Nature Medicine tested whether large language models (LLMs) could help non-medical professionals, i.e. the general public, make better health decisions in everyday health scenarios.

📊 Study Summary

Researchers assigned 1,298 UK adults to either:

Use one of three LLMs
Or be part of a control group without AI assistance

Each participant worked through one of ten medically designed health scenarios, then selected:

A likely diagnosis
An action to take

🔎 Key Findings

When the AI models were tested alone, without human interaction, they performed extremely well.

They identified relevant conditions in 94.9% of cases.

However…

When paired with human users, performance dropped dramatically.

Humans using LLM assistance identified relevant conditions in just 34.5% of cases.

That is an enormous decline, bringing to life AI health advice risks

🧠 Why Did Performance Collapse?

The main failure point was the human-to-LLM interaction itself.

Common problems included:

Incomplete information provided to the model
Poor prompting
Misunderstood questions
Confirmation bias
Humans ignoring or misunderstanding correct answers generated by the AI

In many cases, the correct information actually appeared within the conversation…

…but users failed to recognise or apply it properly.

🩺 Why This Matters in Nutrition & Lifestyle Medicine

Many people are now using LLMs for:

Self-diagnosis
Dietary advice
Symptom interpretation
Blood test interpretation
Supplement recommendations
General health decision-making

And honestly?

I’ve seen some extremely problematic AI interactions from clients already.

Including:

➡️ Leading questions➡️ Partial information➡️ Missed context➡️ Confirmation bias➡️ Factual inaccuracies➡️ Poor interpretation of outputs

When you combine:

imperfect prompting
incomplete health data
inexpert interpretation

…that 35% success rate suddenly makes a lot more sense.

🤖 AI Is Powerful. But It Is Not Magic.

AI can absolutely be useful within healthcare.

In some areas, it is already exceptional.

But effective use requires:

Good information input
Critical thinking
Context
Clinical reasoning
Understanding limitations

Right now, the best outcomes probably come from:

Human expertise + AI assistance

Not humans outsourcing their thinking entirely to AI.

💡 Bottom Line

A powerful AI tool in the hands of an unskilled user may not improve decisions at all.

In some cases…

It may make them worse

My clients enjoy clear, specific, actionable guidance on how to use diet, supplementation, lifestyle and functional testing to reach their personal health goals and resolve their health issues.

Why not book a free health kickstart call to find out how we would enable better health for you? 📲