Enabling efficient communication in artificial agents brings us closer to machines that can cooperate with each other and with human partners. Hand-engineered approaches have substantial limitations, leading to increased interest in methods for communication to emerge autonomously between artificial agents. Most of the research in the field explores unsituated communication in one-step referential tasks. The tasks are not temporally interactive and lack time pressures typically present in natural communication and language learning. In these settings, agents can successfully learn what to communicate but not when or whether to communicate. Here, we extend the literature by assessing emergence of communication between reinforcement learning agents in a temporally interactive, cooperative task of navigating a gridworld environment. We show that, through multi-step interactions, agents develop just-in-time messaging protocols that enable them to successfully solve the task. With memory—which provides flexibility around message timing—agent pairs converge to a look-ahead communication protocol, finding an optimal solution to the task more quickly than without memory. Lastly, we explore situated communication, enabling the acting agent to choose when and whether to communicate. With the opportunity cost of forgoing an action to communicate, the acting agent learns to solicit information sparingly, in line with the Gricean Maxim of quantity. Our results point towards the importance of studying language emergence through situated communication in multi-step interactions.

During every waking moment, we must engage with our environments, the people around us, the tools we use, and even our own bodies to perform actions and achieve our intentions. There is a spectrum of control that we have over our surroundings that spans the extremes from full to negligible. When the outcomes of our actions do not align with our goals, we have a tremendous capacity to displace blame and frustration on external factors while forgiving ourselves. This is especially true when we cooperate with machines; they are rarely afforded the level of forgiveness we provide our bodies and often bear much of our blame. Yet, our brain readily engages with autonomous processes in controlling our bodies to coordinate complex patterns of muscle contractions, make postural adjustments, adapt to external perturbations, among many others. This acceptance of biological autonomy may provide avenues to promote more forgiving human-machine partnerships. In this perspectives paper, we argue that striving for machine embodiment is a pathway to achieving effective and forgiving human-machine relationships. We discuss the mechanisms that help us identify ourselves and our bodies as separate from our environments and we describe their roles in achieving embodied cooperation. Using a representative selection of examples in neurally interfaced prosthetic limbs and intelligent mechatronics, we describe techniques to engage these same mechanisms when designing autonomous systems and their potential bidirectional interfaces.