Microsoft is unveiling details about artificial intelligence systems that outperform human doctors in intricate health assessments, paving a “path to medical closeness.”
The company’s AI division, spearheaded by British engineer Mustafa Suleyman, has created a system that emulates a panel of specialized physicians handling “diagnostically complex and intellectually demanding” cases.
When integrated with OpenAI’s advanced O3 AI model, Microsoft claims its method “solved” more than eight out of ten carefully selected case studies for diagnostic challenges. In contrast, practice physicians with no access to colleagues, textbooks, or chatbots achieved an accuracy rate of only 2 out of 10 on these same case studies.
Microsoft also highlighted that this AI solution could be a more economical alternative to human doctors, as it streamlines the process of ordering tests.
While emphasizing potential cost reductions, Microsoft noted that it envisions AI as a complement to physician roles rather than a replacement.
“The clinical responsibilities of doctors extend beyond merely diagnosing; they must navigate uncertainty in ways that AI is not equipped to handle, and build trust with patients and their families,” the company explained in a blog post announcing the research intended for peer review.
Nevertheless, slogans like “The Road to Overmed Medical” hint at the possibility of transformative changes in the healthcare sector. Artificial General Intelligence (AGI) denotes systems that replicate human cognitive abilities for specific tasks, while superintelligence is a theoretical concept referring to systems that surpass overall human intellectual capacity.
In discussing the rationale for their study, Microsoft raised concerns about AI’s performance on U.S. medical licensing exams, a crucial assessment for acquiring medical licenses in the U.S. The multiple-choice format relies heavily on memorization, which may “exaggerate” AI capabilities compared to in-depth understanding.
Microsoft is working on a system that mimics real-world clinicians by taking step-by-step actions to arrive at a final diagnosis, such as asking targeted questions or requesting diagnostic tests. For instance, patients exhibiting cough or fever symptoms may need blood tests and chest x-rays prior to receiving a pneumonia diagnosis.
This innovative approach by Microsoft employs intricate case studies sourced from the New England Journal of Medicine (NEJM).
Suleyman’s team transformed over 300 of these studies into “interactive case challenges” to evaluate their method. Microsoft’s strategy incorporated existing AI models developed by ChatGPT creators OpenAI, Meta from Mark Zuckerberg, Anthropic, Grok from Elon Musk, and Google’s Gemini.
The company utilized a specific model for determining tests and diagnostics, employing AI systems such as tailored agents known as “diagnostic orchestrators.” These orchestrators effectively simulate a doctor’s panel, aiding in reaching a diagnosis.
Microsoft reported that in conjunction with OpenAI’s advanced O3 model, over eight of the ten NEJM case studies have been “solved.”
Microsoft believes its approach has the potential to encompass multiple medical fields, enabling a broad and in-depth application beyond individual practitioners.
“Enhancing this level of reasoning could potentially reform healthcare. AI can autonomously manage patients with routine care and offer clinicians sophisticated support for complex cases.”
However, Microsoft acknowledges that the technology is not yet ready for clinical implementation, noting that further testing with an “Orchestrator” is necessary to evaluate performance in more prevalent symptoms.
Source: www.theguardian.com
