Why Evidence Matters More Than Ever

It looked like a breakthrough. But researchers resisted the temptation to simply release it into widespread use and hope for the best. Instead, they organized one of the world's first randomized controlled trials.

107 tuberculosis patients in London were carefully selected. They were clinically similar and treated in the same way, except that 52 received streptomycin and 55 did not. The results were striking.

As the British Medical Journal reported in 1948:

"It is evident that at every stage of the disease there is between the two groups a great difference in the course of the disease."

The patients who received streptomycin improved significantly more than those who did not. But the most important outcome wasn't the drug. It was the method.

Medicine had found a way to separate genuine benefit from coincidence, bias, and unsupported extrapolation. Healthcare learned a critical lesson: Plausibility isn't enough. Evidence matters.

Evidence-based medicine was born.

AI has reopened an old question

For decades, healthcare has been refining its ability to separate what is true from what merely appears to be true. Now, AI has reopened that question.

The attraction is obvious. AI—especially generative AI—takes vast amounts of information and distils it into something manageable. Often, it does this remarkably well.

But it has also created a new challenge. For the first time in history, plausible answers can be generated at an extraordinary scale.

An answer can sound:

• fluent
• thoughtful
• authoritative
• evidence-based

...and still be wrong.

The quality of the answer depends on the quality of the evidence beneath it

See the two identical-looking glasses of water at the top of this article? Which glass would you drink from? One comes from a carefully monitored source. The other is from a contaminated one. The clarity of the water tells you nothing. The source tells you everything.

AI works in a similar way. Source hygiene matters.

When a conclusion can be traced back to trusted evidence, confidence becomes proportional to the strength of that evidence. But when that chain becomes unclear, confidence becomes guesswork.

Not all AI is created equal

Used responsibly, AI can be transformative for healthcare. But clinical AI is fundamentally different from general-purpose AI.

Trustworthy clinical AI:

• remains grounded in trusted evidence.
• communicates uncertainty honestly.
• is challenged, tested, and monitored.
• supports clinical judgement rather than replacing it.

Because evidence doesn't make decisions. Clinicians do.

The future of medicine still depends on evidence

Humans are naturally drawn to confidence. The history of healthcare is filled with treatments that sounded plausible, felt logical, and were later proven wrong.

The AI systems that will be most transformative for healthcare won’t be the ones that answer most confidently. They will be the ones clinicians can trust.

AI might be changing the way healthcare accesses information, but it’s not changing the principles that make medicine more reliable. Trusted evidence supports better decision-making. And better decisions create better patient outcomes.

Which, at the heart of it, is what really matters.