AI Models Outperform Doctors in Medical Knowledge by Up to 50 Percentage Points, Shifting Focus to Superintelligence

Artificial intelligence (AI) is currently navigating a "jagged AGI phase," exhibiting advanced capabilities in some areas while remaining weak in others, according to Wharton professor Ethan Mollick. This evolving landscape sees AI models achieving superhuman performance in specific tasks, including outperforming human physicians in general medical knowledge tests, even as the industry grapples with significant integration challenges within organizations. Concurrently, leading AI laboratories are reportedly bypassing the pursuit of Artificial General Intelligence (AGI) to directly target superintelligence.

Mollick's concept of "jagged AGI" describes AI's uneven abilities, where systems can excel at complex expert tasks yet falter on seemingly simple, mundane ones. He noted in a recent social media post, "> we're already in a jagged AGI phase, great at some things, weak at others," highlighting the unpredictable nature of current AI progress. This unevenness necessitates human expertise to discern where AI applications are reliable and where they are not.

In a striking demonstration of AI's burgeoning capabilities, recent studies reveal that advanced AI models have significantly surpassed human doctors in medical knowledge assessments. A study published in June 2025 indicated that a GPT-4-turbo virtual assistant outperformed physicians from four European countries in national medical exams, scoring between 72% and 96% accuracy compared to physicians' 46% to 62%. While AI demonstrated superior performance across most domains, human physicians still showed a slight, though not statistically significant, edge in pediatrics.

Despite these impressive advancements, integrating AI into existing organizational structures presents substantial hurdles. Mollick stated that "> integration into orgs is the real challenge," a sentiment echoed by industry reports. Companies face obstacles such as a lack of strategic vision for AI, poor data quality, insufficient skilled talent, compatibility issues with legacy systems, and cultural resistance to change, with some studies indicating that over 90% of organizations struggle with AI integration.

The strategic direction of AI development is also undergoing a significant shift. The tweet noted, "> AI labs are now skipping AGI and aiming directly for superintelligence." Major players like Meta have established new "Superintelligence Labs," explicitly setting goals beyond AGI—defined as AI matching human intellect—to develop systems that far exceed human cognitive abilities. This acceleration towards superintelligence, though debated by experts regarding its feasibility and implications, underscores the intense competition and ambitious long-term visions within the AI research community.