Resonating Minds—Emergent Collaboration Through Hierarchical Active Inference

Abstract

Working together on complex collaborative tasks requires agents to coordinate their actions. Doing this explicitly or completely prior to the actual interaction is not always possible nor sufficient. Agents also need to continuously understand the current actions of others and quickly adapt their own behavior appropriately. Here we investigate how efficient, automatic coordination processes at the level of mental states (intentions, goals), which we call belief resonance, can lead to collaborative situated problem-solving. We present a model of hierarchical active inference for collaborative agents (HAICA). It combines efficient Bayesian Theory of Mind processes with a perception–action system based on predictive processing and active inference. Belief resonance is realized by letting the inferred mental states of one agent influence another agent’s predictive beliefs about its own goals and intentions. This way, the inferred mental states influence the agent’s own task behavior without explicit collaborative reasoning. We implement and evaluate this model in the Overcooked domain, in which two agents with varying degrees of belief resonance team up to fulfill meal orders. Our results demonstrate that agents based on HAICA achieve a team performance comparable to recent state-of-the-art approaches, while incurring much lower computational costs. We also show that belief resonance is especially beneficial in settings where the agents have asymmetric knowledge about the environment. The results indicate that belief resonance and active inference allow for quick and efficient agent coordination and thus can serve as a building block for collaborative cognitive agents.

Intertwining the social and the cognitive loops: socially enactive cognition for human-compatible interactive systems

“…Classical work in AI on multi-agent coordination focuses on finding optimal solutions through planning, prior to the actual interaction and from the point of view of either a centralized, omniscient control unit [68] or a decentralized one through dedicated forms of communication between the agents. In recent work [69] we asked whether ‘satisficing’ (in contrast to optimal) collaboration can emerge without planning, only from on-the-fly coordination between socially enactive cognitive agents that rely on situated, prediction-based task behaviour that is affected by their mentalizing about other agents’ intentions. To that end, a model of ‘hierarchical active inference for collaborative agents’ (HAICA; see figure 4, left) was developed.…”

“…The simulations show that two of those socially enactive cognitive agents, when paired together in the environment, are able to collaboratively succeed with this task in many situations and in a very resource-efficient manner. Achieved team performance is comparable to state of the art deep reinforcement learning models [70] and, although not always being optimal, is by orders of magnitude more computationally efficient than full Bayesian planning (figure 4b; see Pöppel et al [69] for details). These results are due to an emergent coordination as the agents dynamically and bidirectionally adapt to each other as well as to the shared environment they simultaneously alter, yet without any explicit coordination through communication or global planning.…”

Interpersonal attunement in social interactions: from collective psychophysiology to inter- personalized psychiatry and beyond

“…First, we dynamically ‘embody’ each other in and through social interaction [ 20 ], enabling interpersonal attunement (e.g. interpersonal belief resonance; [ 29 ]). In other words, by engaging in sensory-motor couplings with others in social interactions, we have our bodily structures mutually transformed beyond the here and now [ 20 , 30 ].…”

On a Possible Basis for Metaphysical Self-development in Natural and Artificial Systems

“…Likewise, G&C review Wanja Wiese model accounting for persistent sense of self in terms of attentional shifts that are accompanied by a pervasive feeling of control over ongoing actions (there is a lot of recent attention to this idea in different areas, e.g. Sennesh et al, 2022, in the context of predictive processing and interoception as allostasis; Kahl et al, 2021, in the context of artificial systems; foundationally, see Sterling, 2012). And, on this model, self becomes evident at highest levels of contextually dependent processing.…”