Research on AI for drug seems to be fast promising – technology already accelerates the development of the drug, Google is using AI to improve its medical advice, and wearable companies are taking advantage of technology for predictive health facilities. Now, the Microsoft is the latest to transfer the target post.
On Monday, the company announced blog post Microsoft AI Diagnostic Orchestrator (MAI-DXO), its medical AI system, successfully diagnosed 85% of cases in the New England Journal of Medicine (NEJM). This rate of diagnosis is four times higher than human physicians. Nejm cases are particularly complex and often require many experts.
Also: Openai’s healthbench shows that AI’s medical advice is improving – but who will listen?
Given how inaccessible, complex and misleading healthcare systems continue, it is no surprise that people are seeking help from technology wherever possible.
“Microsoft’s AI consumer products such as Bing and Copilot, we see more than 50 million health -related sessions every day,” Microsoft said in the announcement. “For the first time, from the query of knee-pain to an immediate-care clinic, the search engine and AI companions are quickly forming a new front line in healthcare.”
how it works
Human physicians have to pass the US Medical Licensing Examination (USMLE) to practice medication, a test that is also used to evaluate how AI systems perform in medical contexts, both models-to-models and compared with humans.
Currently, AI scores well on USMLE-one side effect, Microsoft said, those who remember the model (instead of understanding) answers to many-love questions, which will not produce the most sound therapy analysis. Most industry-standard AI benchmarks have been saturated for a while, meaning that AI model is developing very early for tests that are useful challenging.
To combat the issue, Microsoft created a sequential diagnosis benchmark (SD bench). Sequential diagnosis is a procedure that uses real physician to diagnose patients how their symptoms exist and move forward with questions and tests. The test presents clinical challenges from 304 Nejm cases, which humans and AI models can use to ask questions.
Also: Anthropic states that the cloud helps to support users emotionally – we are not convinced
Microsoft then paired the diagnostic agent, My-DXO with several frontier models, including GPT, Lama, Cloud, Gemini, Grocke and Deepsek, and put the agent in the SD bench test. The Miaai-DXO, which is also using LLM, is “using in a virtual panel with various clinical approaches in the virtual panels of physicians, collaborating to solve clinical matters,” Microsoft explained.
One in Video demoMAI-DXO also shows its argument as it questions the benchmark, develops potential diagnosis, and tracks the cost of each requested test. Once the agent has the necessary information from the benchmark about the case, it changes its diagnosis, asking for different scans and displays a clinical process that is more familiar to human physicians.
The correct diagnosis is low cost
Microsoft’s blog post said, “Mai-DXO promoted the clinical performance of each model we tested.” The company compared the results of 21 physicians from the UK and the US with an experience of five to 20 years, which reached an average accuracy of just 20%.
Also: You should not rely on AI for Therapy – why is it here
Microsoft stated that Mai-DXO is also configured, meaning that it can run within the cost limits set by a user or organization-a feature that allows the agent to run the cost-benefit analysis of some tests, which is highly relevant to astronomical pricing of American medical care and also consider some human doctors and patients.
This feature is also a railing, of type – without it, AI “may be default to order every possible test – the cost, the patient’s discomfort, or careful care,” the blog post explained. MAI-DXO also returned high accuracy and low cost than individual models or human physicians.
Will AI replace your doctor?
Perhaps never soon – although Microsoft’s blog post mentioned that due to its width of knowledge, AI “can overcome clinical argument abilities, which are more than any individual physician, in many aspects of clinical argument.”
The company believes that such systems can “reopen the healthcare” by giving patients the option to check themselves firmly and help doctors with complex cases. Cost savings will be another plus for an industry that prone to continuous high cost and opaque pricing structures.
Also: AI is giving relief to the doctor from burnout. Here is how this mental health is changing
Microsoft admitted that Mai-DXO has only been tested on these special cases, so it is not clear how it will handle everyday tasks. However, this issue cannot be relevant anyway if the agent is not to change human doctors, which Microsoft has also maintained in a blog post.
MAI-DXO is part of a “dedicated consumer health effort”, which was started last year Microsoft AI, the company said. Other AI products are included within that initiative Red-dinoA radiology workflow tool, and Microsoft Dragon CopilotA voice is designed for AI auxiliary medical professionals.
