PI Brief Update

Why Medical Diagnosis Is a Dead End for AI Doctors - Dr Mark Burgin

26/05/25. Dr Mark Burgin explains how generative AI models will change the way that medical information is used to help patients.

The biomedical (medical) model has been dominant for 2000 years. The way that medical students are trained still follows the basic structures of Hippocrates. Diagnoses are linked with scientific explanations which predict the effectiveness of treatment. The prediction or prognosis has allowed doctors to offer timely interventions.

The window of opportunity has always been a weakness for the medical model. A treatment offered too early might cause over treatment, too late and it will not be effective. Cardiac risk factors such as raised cholesterol and smoking are treated like diseases. Pre-diabetes and pre-hypertension have been invented to justify medicalise the population.

Public health has been undermined by lack of trust in the medical model. Immunisation and infection control have had extraordinary impacts on population health. It is worth noting that TB was largely controlled by social changes (e.g. hygiene) before drugs and immunisation. Population change occurs when the individuals work together and there is not a pill for vaccine hesitancy.

Big data is overwhelming the medical model, doctors struggle to find explanations for their patient’s phenomena. Changes seen on scans can take attention away from those whose illnesses are more complex. Rounds of endless follow up without improving care can be disappointing. Health care funding spirals with minimal increases in productivity.

Alternatives to the Biomedical (Medical) Model

• Public health is based upon the idea that the richest have always had access to better health. Even in Roman times it was not unusual for an emperor to live to 70 years old. Protecting everyone from infections, poor nutrition and pollution gives everyone a chance to live longer. Population engagement has been limited by overuse of one-size-fits-all and heavy handed mandates.
• The biopsychosocial model (BPSM) sees the person as a complex being part of their community. This approach allows the GP to see psychosocial steps that will improve the person’s health. Managers have been unable to understand why the BPSM is effective leading to micromanagement and underfunding. Despite the 100 times greater effectiveness of the BPSM it remains largely used.
• The Disability model considers life expectancy as less important than quality of life. Identifying the impairments and how they impact the functioning has led to an insight. Many of the adverse effects of illness are due to way that society is constructed. For instance a person with mobility problems is disabled by the social barriers they face. Reasonable adjustments will help everyone live a better life.

Political barriers to each of these alternatives means that progress will be slow. The other models are less attractive or easy to sell leading to public resistance. Challenges such as chaotic GP leadership, the lack of a disability professional and public health in retreat make things worse. It would take an extraordinary health secretary to guide the health service back to balance. 

The AI model of health

Many AI models of health rely upon the medical model. They attempt to take a scan or a medical presentation to find the answer. This approach has led to a number of difficulties from hallucinations to differences of opinion between experts as to the right answer. This has made further developments slow as increasing amounts of training data has been required.

Frustratingly the vast amount of medical data available through apps and medical monitoring has been largely ignored. Avoiding the complexity of medical data seems to be the opposite of what the AI doctor should be doing. The large language model LLM based AI models store their knowledge as weights. There is an alternative type of AI model which might be more useful.

Weather models such as GenCast are based on a Diffusion Model (like generating video). This allows them to show the likely outcomes as a series of predictions. The advantage of these models is that they do not need to make any assumptions or know the science. They would be able to suggest the results of tests or treatments before they are tried.

Think for a moment how having a way of modelling the weather of illness would change clinical practice. The model may suggest a treatment or test that does not appear to make any sense based upon the suspected diagnosis. Like a futuristic Sherlock Holmes the AI doctor would see patterns that others cannot and find impossible answers.

Conclusions

Creating a virtual model of the patient has advantages over an AI LLM based problem solver. Including psychosocial information and understanding of disability would enhance the functioning. The model would be able to predict the patient’s psychosocial responses as well as the medical effects of the treatment. It would be able to combine all the information about the patient.

Medical Diagnosis is likely to remain as a shorthand to help patients feel engaged with their care. The choice of treatments and investigations is likely be more fluid and less based upon protocols. Modelling the weather of illness will adapt to changes in the clinical patterns. Doctors should resist changing the patient’s treatment based upon every suggestion and instead support the patient’s choice.

Current AI models that are based on finding a diagnosis will hit a dead end. Medical diagnosis was always a guess, a place holder whilst waiting for more information. For some diseases understanding the science has led to successful treatments. For many diseases we still do not understand the science behind them any better than the weather.

Diffusion models can predict the weather better than current models and using a fraction of the computing power. They can incorporate complex inputs without needing to have them explained. As this technology is applied to medicine and law it will allow the question to move from ‘what is the problem’ to ‘what are the next steps’.

Doctor Mark Burgin, BM BCh (oxon) MRCGP is a Disability Analyst and is on the General Practitioner Specialist Register.

Dr. Burgin can be contacted on This email address is being protected from spambots. You need JavaScript enabled to view it. and 0845 331 3304 websites drmarkburgin.co.uk and gecko-alligator-babx.squarespace.com

This is part of a series of articles by Dr. Mark Burgin. The opinions expressed in this article are the author's own, not those of Law Brief Publishing Ltd, and are not necessarily commensurate with general legal or medico-legal expert consensus of opinion and/or literature. Any medical content is not exhaustive but at a level for the non-medical reader to understand.

Image ©iStockphoto.com/Just_Super