Collective Reasoning Breakthrough: Multi-Agent System Outperforms Human Expert Panels

On January 7, 2028, the National University of Singapore AI Laboratory published a paper in Nature Machine Intelligence presenting full experimental data for the Nexus-7 collective reasoning framework. The system, comprising seven specialized AI agents working in concert, has surpassed human expert panels in multiple high-stakes professional tasks for the first time.

The core innovation of Nexus-7 lies in its "divergence-driven deepening" mechanism. When multiple agents arrive at different conclusions about the same problem, the system does not simply take a majority vote. Instead, it treats the points of divergence as new reasoning starting points, generating more granular sub-questions for deeper analysis. Professor Chen Weiming, the project lead, explains that this mechanism gives the system a particular edge when dealing with ambiguous information.

In medical diagnosis testing, Nexus-7 achieved 94.3% accuracy on 2,000 complex cases, compared to 89.6% for a panel of five senior specialists. In legal contract analysis, the system identified 97.1% of contractual vulnerabilities, versus 91.8% for the lawyer control group.

The research team acknowledges current limitations. Nexus-7's average reasoning time is 3.2 times that of human expert panels, with a single inference costing approximately $47. Professor Chen noted the team is developing a lightweight version targeting costs below $5.

Reactions from the industry have been mixed. DeepMind researcher Sarah Chen described the achievement as a paradigm shift from individual to collective AI intelligence. However, MIT Media Lab professor Kate Crawford cautioned that collective reasoning could amplify biases present in individual agents, calling for more rigorous fairness auditing frameworks.

The Singapore government has designated Nexus-7 as a national AI priority project, committing S$120 million over the next two years for deployment in healthcare and legal sectors.