By Michael J. Critelli | MakeUsWell Newsletter, 

For all of the advances in medicine, nutrition science, and artificial intelligence, one of the most poorly understood conflicts in health remains the tension between what is “evidence-based” and what is “personalized.”

We often speak as if they are the same thing. They are not.

Governments regulate medicine, medical devices, foods, and beverages largely on the basis of evidence generated through research. In medicine, the gold standard is the randomized clinical trial: carefully designed studies that isolate the effects of a drug, therapy, device, or procedure and compare outcomes against a control group. The goal of both the randomized clinical trials and “evidence-based medicine” is to assess the effects of whatever is being tested with respect to a large population.

This framework has saved millions of lives.

But the best evidence is never as certain as we want it to be. It never completely enables us to assess the complete effects on every individual. 

Drugs have side effects. Treatments work differently across populations. Therapies approved as breakthroughs may only help a subset of patients. In cancer care, for example, many therapies are approved not because they work universally, but because they perform statistically better than the alternatives available at the time.

Sometimes that improvement is meaningful even if the therapy only works for a minority of patients.

We accept that imperfection because a probabilistic improvement is still better than no improvement at all.

That reality points to something deeper: evidence-based medicine is fundamentally based on population-level probabilities, while health outcomes are experienced individually.

A therapy that works for 50% of one group may work for 20% of another. A medication tolerated well by most people may create severe complications for someone else. Every person brings different genetics, environmental exposures, stress levels, metabolic conditions, sleep patterns, and behavioral habits into the equation.

The same principle applies to food and nutrition.

Foods may begin with relatively standardized nutritional profiles, but their actual effects vary enormously based on processing methods, preparation techniques, combinations with other foods, individual metabolism, stress levels, medications, supplements, and still imperfectly understood variables such as the gut microbiome.

Two people can consume the same meal and experience dramatically different physiological responses.

This is where the future of health becomes both exciting and dangerous.

Artificial intelligence can help synthesize enormous amounts of nutritional and medical information. It can identify patterns humans might miss. It can accelerate discovery. It can help consumers navigate complexity.

But AI must be used with a strong dose of humility.

At best, most nutritional and health guidance, even that emerging from randomized clinical trials, remains probabilistic rather than absolute. The quality of conclusions depends heavily on the quality of the underlying evidence. That evidence itself depends on multiple factors:

1. The size and representativeness of the populations studied
2. The rigor of the research methods
3. Whether findings can be replicated consistently
4. Whether conclusions align with other credible evidence
5. The reliability of the systems used to gather and analyze the data
6. Whether the research is current and up-to-date

One of the underappreciated risks of AI is that large language models often rely heavily on summaries, headlines, abstracts, and secondary interpretations of research.

If those summaries are incomplete, biased, exaggerated, or politically distorted, the AI may confidently reproduce flawed conclusions.

The old computer science warning still applies: garbage in, garbage out.

We saw this problem repeatedly during the pandemic. Headlines often overstated the significance of early findings, oversimplified nuanced research, or ignored studies that conflicted with prevailing narratives. Public trust suffered because certainty was often projected where uncertainty actually existed.

That experience reinforced an important lesson: evidence itself must be evaluated.

Not all research is equally reliable. Not all publications maintain the same standards. Not all interpretations are honest. And not all scientific consensus is as settled as media narratives suggest.

That is why the product we are building uses multiple AI agents to validate and cross-check the conclusions generated by other AI agents. It is also why we believe some form of research-rating framework will eventually become essential.

But even after all of that, one critical source of insight remains indispensable: The user.

No randomized clinical trial can fully replicate the infinite combinations of genetics, medications, supplements, stress levels, microbiome differences, lifestyle patterns, and food combinations that shape real-world human outcomes.

The future of nutrition and wellness will not come from abandoning evidence-based science.

It will come from combining rigorous evidence with personalized observation.

That is why we are designing our system as a conversation, one that combines evidence-based insights with the lived experiences, responses, and observations of individuals themselves.

Because the best health outcomes will likely emerge not from rigid certainty, but from intelligently combining broad scientific evidence with highly personalized human feedback.