Microsoft claims it has developed an AI system that outperforms doctors at diagnosing complex medical conditions.

The new system is four times more accurate than practising doctors according to a blog-post on Microsoft AI’s website.

The company is hailing the development as creating a “path to medical superintelligence”.

AI superintelligence (ASI) is the step beyond Artificial General Intelligence (AGI). The latter is where AI can match human intelligence. ASI is where AI surpasses human experts by a marked factor.

The London-based Microsoft team, led by British AI pioneer Mustafa Suleyman, developed their system, an agentic AI system, as an overlay on top of available AI models. When combined with OpenAI’s o3 AI model, it correctly identified more than of eight of ten medically challenging case studies. By contrast, practising doctors achieved an average success rate of two out of ten.

Mustafa Suleyman, a Londoner who co-founded DeepMind, has worked extensively with the UK health service. In 2016, Suleyman launched DeepMind Health to develop clinician-led technology.

The blogpost also claims that its AI system is significantly cheaper than using doctors in hospitals, saving money especially at ordering the right tests first time. Hospital tests are costly, often involving a range of clinical and technical practitioners. Microsoft’s AI is likely to get the right tests ordered first off, compared to hospital doctors.

However, the company has backed off from claiming that its system should replace human doctors, at least for now. In its blogpost, the company wrote: “Their clinical roles are much broader than simply making a diagnosis. They need to navigate ambiguity and build trust with patients and their families in a way that AI isn’t set up to do.”

The Microsoft system built on complex case studies taken from the New Journal of Medicine.

Suleyman’s team turned over 300 of the studies into “interactive case challenges”.

The Microsoft AI system has been tested by the team on a wide selection of AI models, including OpenAI, Meta, Anthropic, Grok and Google’s Gemini.

The company is submitting their agentic AI model to clinical review.