The challenge of integration of virtual agents and co-robots into the human society requires for these future artifacts to become human-compatible, in addition to being generally intelligent and capable of providing specific expertise. Specifically, a human-compatible agent should be able to induce in humans a sense of co-presence of a potentially equal mind capable of mutual understanding and human-like learning, long-term personal relationships and generous initiative. Achieving these qualities requires social emotional intelligence. In order to be able to implement equivalents of social emotions in artifacts, it is vital to better understand social interactions in small groups based on cognitive architectures like eBICA (Samsonovich, 2013). The present work makes one step further in this direction, continuing the development and computational exploration of the new framework for emotional cognitive architectures. In the focus are relationships of trust and mutual respect, leadership and allegiance, and related to them social and complex emotions.