Artificial intelligence incomparably better than a doctor. Surprising research results in the USA
This material is only informative and does not constitute medical advice or incentive to independently use artificial intelligence systems in diagnostics, treatment or making clinical decisions. In case of any health problems, consult with authorized medical staff.
The latest publication of the Harvard Medical School and Stanford University team shows that Artificial intelligence is doing very well in the medical sector. The authors checked the O1-Review (OpenAI) model in six separate experiments-from classic cases published in New England Journal of Medicine (Nejm) for real duties at the emergency room in Boston. In each test, starting with generating a differential list of recognition and ending with decisions regarding further treatment, the algorithm achieved results that exceeded certified internists.
Researchers They compared, among others 143 cases from the Clinical and Patological Conference of NEJM. The model aptly placed the right disease in its five -point list in 78 percent. situationwhile doctors in historical research usually did not exceed 65 percent.
Even more spectacular was the first contact with the patient in the emergency room when clinical data is the least. The algorithm indicated a precise or very close diagnosis in 66 percent. cases. It's a dozen or so points more than on duty internists.
Read also at Business Insider
Check also: Is it still worth studying? Artificial intelligence and the future of mental work
Artificial intelligence as a diagnostician
The differences are not limited to “guessing diseases”. Wherever thinking counts step by step, LLM presents explanations that meet-and often exceeding-a strict, ten-point R-IDEA scale, created to measure the quality of human clinical reasoning.
The openai model was also more reliable in planning additional tests and therapeutic decisions, which in practice means a shorter diagnostic path and a smaller number of unnecessary procedures.
Where does this superhuman advantage come from? The algorithm did not sleep, for example, in pathophysiology classes – he studied all over the available Internet, including millions of scientific publications. However, the key is in the architecture of the O1 series itself. This model “stops” to repeatedly prescribe its own reasoning before it answers.
This internal recursion, supported by the gigantic capacity of working memory, allows him simulate the clinical grinding process of hypotheseswhich in people is limited to what will fit in short -term memory and a few minutes of consultation. This approach, named by OpenAI More time thinkingtranslates into a better quality of arguments and a lower variance of the results.
The second pillar of the advantage is immediate access to a wide epidemiological database and all known recommendations. A specialist doctor can be an expert in several hundred disease entities. The AI model without effort juggles thousandsupdating knowledge at the pace of scientific publications published. As a result, it is less susceptible to heuristic errors (e.g. availability effect) and less often omits rare but key diagnoses.
Read also: From a billion dollars to bankruptcy. What does the fall of the “star” of AI support us
Why this is not the end of work for doctors
Despite the advantage in paper and retrospective tests, the model does not receive the role of the role of final decision -makers. First of all, his statistical knowledge is still similar to people. In probabilistic tasks (e.g. risk assessment after the test result) did not pierce previous versions of the GPT-4, which shows that where the data is uncertain or dependent on the local context, human intuition is still needed.
Secondly, The system requires infrastructure – from private clouds with HIPAA certificate to interfaces enabling tracking logic of decisions. This means multi -million investments of hospitals and insurers, but also new markets for providers of computing equipment and specialized platforms integrating medical solutions with the API of artificial intelligence models.
Industry Health-Tech He reads the results of the study as a signal that the transition from pilotage to mass implementation is a matter of user regulation and experience, and no longer computing power.
See also: In 10 years the Internet will be beyond recognition. Better start preparing now [OPINIA]
Patients will finally decide
Finally, the trust of patients and staff counts. The author's vision does not match the clinical realities in which relationship and empathy are needed. It is still a man who explains the risk, sets the patient's life priorities and takes responsibility for many things.
Researchers now suggest using the “double look” model. That is? AI offers hypotheses, and the doctor – equipped with audit tools – verifies and adapts them to a specific case.
If the results of Harvard and Stanford are confirmed in subsequent clinical studies, A double economic effect is expected. First of all, a decrease in diagnostic costs (less unnecessary tests, faster diagnosis), and secondly – postponement of demand from human work towards subscription models and computing infrastructure.
This, in turn, may lead businesses – from public systems to Big Techy – to accelerate acquisitions and partnerships to secure the advantage of scale at an early stage of adoption.
For pharmaceutical companies and equipment manufacturers, the possibility of integration of algorithms is opened directly with a contrast syringe or an ultrasound apparatus. Where the time from the scan to conclusions is critical.
From the perspective of innovation, this means Return towards clinical co-pilot. Instead of providing only administrative tools, start-ups will build systems that understand the doctor's note in real time, propose five best documented diseases, indicate missing data, or automatically fulfill diagnostic orders. We are regulating with a bit virgin area, which is why pioneers should prepare strategies of compliance with the upcoming regulatory requirements and European AI ACT.
The medical sector will change quickly
The Harvard -Stanford study puts a seal on what we've seen in licensing tests and Big Techów reports for two years. Mianowice: Algorithmic collaborator not only catches up, but begins to surpass a specialist in a key area of medicine.
Faster than later the diagnosis will become the result of dialogue between man and the machine. And it is there that the most interesting part of the digital health revolution will take place in the interface of this dialogue.
Author: Grzegorz Kubera, Business Insider Polska journalist
