Forget AI, Microsoft is working on "superintelligence" - with medical diagnosis the first area of interest (www.techradar.com)

Microsoft says it’s pivoting from an open-ended race for general-purpose AGI to a narrower, “Humanist Superintelligence” (HSI) approach meant to serve defined human goals under strict oversight. The company highlights medicine and education as first use cases: its diagnostic system MAI‑DxO reportedly achieves an 85% success rate on complex diagnostic challenges—surpassing expert performance—and Microsoft envisions personalized AI companions to assist teachers. The move is pitched as balancing powerful automation with human control, but raises familiar questions about validation, regulation, privacy, clinical integration, and the risk of overdependence on algorithmic decision-making. HSI’s promises rest on heavy technical and infrastructure bets. Microsoft acknowledges HSI will require massive, CPU‑intensive data centers and expects electricity demand from AI to grow (projected >30% by 2050), even as the same tech is proposed to optimize renewables and batteries. Mustafa Suleyman emphasizes HSI must never be autonomous or self‑improving, yet the company’s containment and control mechanisms remain largely untested—leaving open how limits would be enforced if models acquire self‑modification capabilities. For AI/ML researchers and policymakers, Microsoft’s proposal is significant: it reframes ambitions toward purpose-driven systems but surfaces urgent technical, governance, and environmental trade-offs that must be resolved before widescale deployment.