The extreme sensitivity and flexibility of the LOFAR radio telescope is achieved through a large reception area (10,000-1,000,000 m2), wideband (0.2-4 GHz) signal processing and digital beam forming. The trend towards more local autonomy and resource sharing in the infrastructure and on-line real-time distributed calibration provides conditions for emergent behavior. However, scientific instruments need to be dependable and highly accurate. Hence potential undesirable emergent behavior must be controlled. The paper discusses computational intelligence to achieve early detection of emergent behavior. Locally autonomous computing will then validate the instrument stability based on a solid design theory from a consistent combination of non-deterministic models of computation, exact dynamic instrument models, and data-driven system health.