We propose an agent model consisting of perceptual and pro-prioceptive pathways. It actively samples a sequence of per-cepts from its environment using the perception-action loop.The model predicts to complete the partial percept and propri-ocept sequences observed till each sampling instant, and learnswhere and what to sample from the prediction error, withoutsupervision or reinforcement. The model is exposed to twokinds of stimuli: images of fully-formed handwritten numer-als/alphabets, and videos of gradual formation of numerals.For each object class, the model learns a set of salient locationsto attend to in images and a policy consisting of a sequence ofeye fixations in videos. Behaviorally, the same model givesrise to saccades while observing images and tracking whileobserving videos. The proposed agent is the first of its kindto interact with and learn end-to-end from static and dynamicenvironments to generate realistic handwriting with state-of-the-art performance.

Distributed cognition and decision making has been a topic ofintense research in the recent years. In this paper, a computa-tional model of distributed decision making using a commu-nity of predictive coding agents is developed. The agents areembodied multimodal entities and situated in a shared envi-ronment. They have different visibility of the environment dueto unique sensory and generative models. We show that com-munication between agents helps each of them reach a shareddecision in a way that cannot be reached by brain processes ina single agent. Using a simulated environment, we show thatsensory limitations may lead to incorrect or delayed causal in-ferences giving rise to conflicts in the mind of a predictive cod-ing agent, and communication helps to resolve such conflictsand overcome the limitations.