In this keynote, I will present the challenges facing research on software architecture to address the building of artificial collective intelligence in systems-of-systems, while enforcing safety. I will discuss the discriminating characteristics of systems-of-systems when compared with single (even very large) systems from the software architecture perspective. Especially, I will analyze why novel architectural approaches are needed to handle the complexity of software-intensive systems-of-systems in particular regarding the architectural challenges implied by self-organization and emergent behavior, the two key features for supporting collective intelligence. I will survey novel solutions for architecting systems-of-systems based on supervenience for achieving intelligent collective behavior and illustrate their applications in the field of driving automation for connected and automated vehicles. I will conclude by identifying the guarantees of correctness of the proposed approaches for architecting safe systems-of-systems, based on formal description and verification techniques dealing with dynamicity and uncertainty.