In the complex tapestry of human interactions, understanding actions goes beyond mere recognition of objects, faces, or scenes. We possess a remarkable ability to discern the actions of others, whether it involves something as mundane as walking, reaching, or eating, or more nuanced actions like chasing and following. This understanding is crucial not only for predicting others’ behaviors but also for inferring deeper insights into their attitudes and mental states, such as intention, agency, deception, confidence, belief, preference, and value.
Actions, regardless of their underlying motives, typically serve a common purpose—they are instrumental or pragmatic, aimed at achieving specific physical outcomes in the environment. However, human actions are not limited to instrumental goals. We also engage in actions to communicate, signal characteristics, or even purely for creative expression. Among these broader categories of actions lies a lesser-explored domain—epistemic actions—actions driven by the pursuit of knowledge or information.
Epistemic actions are woven intricately into our daily lives. Whether it’s checking the temperature of water by dipping a toe into a pool, pressing on a door to determine if it’s locked, or shaking a wrapped-up gift to guess its contents, these actions are fundamentally driven by the quest for knowledge. What’s fascinating about epistemic actions is that the nature of the information sought can influence how one carries out the action. For instance, shaking a box to discern the number of objects inside is different from shaking it to determine their shape, texture, or weight.
Despite their prevalence, epistemic actions have received less scientific attention compared to other types of actions. This prompts the question: Can observers decipher an individual’s information-seeking intent by simply observing their body movements?
The Present Experiments: Can You See What I Want to Know?
To delve into the realm of epistemic action understanding, a case study focused on manipulating objects to learn about them was explored. The experiments comprised two phases, offering a unique perspective on the recognition of epistemic actions.
Phase one involved filming volunteers engaged in a “physics game.” In this game, objects were concealed within an opaque box, and participants had to guess what was inside by shaking it. This game had two distinct rounds: the first round required guessing the number of objects in the box, while the second round involved guessing the shapes of the hidden objects. Previous research had suggested that participants should perform well in both rounds, meaning they could accurately determine the number and shapes of the concealed objects based on their shaking.
The focus of the present study lay in the second phase. Independent participants were tasked with watching videos of the physics game and determining which videos corresponded to each round—discerning who was shaking for number and who was shaking for shape. This seemingly straightforward task concealed a web of complexities.
Successfully identifying the intentions behind the actions demanded several layers of physical and psychological reasoning. Participants needed to grasp which properties could be deduced from various interactions, comprehend the strategies behind box-shaking that revealed such information, and ascertain whether the players themselves understood these dependencies. If participants succeeded in this endeavor, it would signify that casual observers could, in fact, recognize an agent’s epistemic intent merely from the kinematics of their actions.
Top: Players were filmed trying to determine the contents of a box (specifically, the number or shape of the objects inside), only by shaking it. Later experiments vary the box’s contents. Bottom: Observers watched these videos and judged which came from which round: Who was shaking for number and who was shaking for shape?
The present study delved into the intriguing realm of epistemic action understanding, shedding light on the remarkable ability of observers to infer what information another person seeks solely through the observation of their motor behavior directed towards a box. This phenomenon held true across a broad range of participants, encompassing hundreds of individuals and various experimental conditions. The key findings of this investigation are discussed below, along with their implications and avenues for future research.
Robust Results: One of the central findings of this study is the robustness of the observed effects. Participants, both those who correctly guessed the box’s contents and those who made incorrect guesses, exhibited systematic differences in their shaking motions across rounds (number vs. shape). Furthermore, the success of observers in determining which shaking motions corresponded to specific epistemic goals was consistent across variations, including situations with limited information about the players’ task and when the box’s contents were equalized. These findings underscore the reliability and consistency of the ability to discern epistemic intent from observable actions.
Visual Recognition of Epistemic Goals: While previous research has demonstrated the ability of observers to infer instrumental or pragmatic goals from behavior, this study extends these findings by showcasing that epistemic goals can also be inferred through visual observation. This goes beyond mere recognition of what someone wants to do and extends into the realm of understanding what someone wants to know. Additionally, the observers in this study displayed sensitivity to the nuanced content of these goals, surpassing simple perceptual knowledge attribution. These results complement existing research on various action categories, such as communicative and affiliative actions, as well as actions performed purely for creative expression.
Future Directions: The present study paves the way for future research on epistemic action understanding. One intriguing avenue is the investigation of the strategic differences in actions observed during different rounds, such as the up-and-down shaking for Number and side-to-side tilting for Shape. Precisely characterizing these patterns, including their stability across variables like the material of the objects and box, presents an interesting challenge for computer vision systems operating on kinematic data. Developing models that formalize how observers use their physical knowledge to infer epistemic intent could also inform computational work related to intuitive mentalizing.
Future research might delve into the developmental trajectory of these abilities, exploring when epistemic action understanding emerges during development and whether it coincides with the development of other mentalizing abilities. Additionally, the study could be extended to explore other forms of epistemic actions, as mentioned in the Introduction, such as inferring someone’s search intentions based on the size and shape of containers they investigate. Finally, the focus of the present study was on sensitivity to one epistemic goal over another (determining number vs. shape), but there is potential to investigate sensitivity to epistemic goals as opposed to pragmatic goals, such as navigating an environment for exploration versus reaching a specific destination.
Cognitive Implications: Beyond its specific contributions to the field of epistemic action understanding, this study highlights the significance of exploring what someone is attempting to know, rather than just what someone is attempting to do, across various cognitive domains. It emphasizes the need to consider the broader landscape of human cognition, which includes not only actions and intentions but also the pursuit of knowledge and information-seeking behaviors. Understanding these dimensions of human behavior can have far-reaching implications for our comprehension of social interaction, decision-making, and communication in diverse contexts.
In conclusion, the present work advances our understanding of how individuals perceive and interpret the actions of others in the context of seeking information. The findings underscore the capacity of observers to decode epistemic intent from observable actions and offer a platform for further research into the intricacies of epistemic action understanding, its developmental trajectory, and its implications for computational modeling and real-world applications.
reference link : https://www.pnas.org/doi/10.1073/pnas.2303162120