Symposium on Interactive Cognition

Interactive cognition refers to a theoretical perspective in cognitive science that conceptualizes cognitive processes as emerging from continuous interactions between individuals and their physical, digital, and social environments. Rather than treating cognition as confined to internal mental representations and processes, this view emphasizes the dynamic coupling of perception, action, memory, and external artifacts. Cognition is not only driven by endogenous intentions and mechanisms but also by exogenous tasks, environmental constraints or affordances, and feedback loops; it is mediated by tools such as external symbols, interfaces, and devices; it is often the result of complex interactions between cooperating and competing agents. The talks in this symposium will emphasize the central role of interaction, either interpersonal, human-machine, or human-environment, in cognition. They will contribute to expanding the scope of the theoretical framework of integrated cognition (a.k.a., cognitive architecture) to include mind-environment interactions as explanatory factors. Presenters will showcase empirical data, theories, concepts and models that demonstrate improved characterization of cognitive phenomena when situational and relational factors are considered. For example, individual cognition and behavior will be significantly moderated by time pressure, risk, tool availability, and contexts of cooperation or competition between agents. The talks will each address the following questions: • What is the added value of adopting an Interactive cognition approach in your research? • What are core principles of Interactive Cognition that are addressed? • How can these principles be transferred to other situations or tasks? • How should the topic of "Interactive Cognition" be developed in the future?

This is an in-person presentation on July 20, 2026 (09:00 ~ 09:20 EDT).

Some recent advances in creating and testing models of interacting have theoretical and practical implications. We have had a simulated eye and hand (VisiTor) that works with an existing ACT-R model of a 14-min spreadsheet task to use two different, uninstrumented interfaces and compare task time on the two spreadsheet systems. The model uses oN-the-shelf vision algorithms to tie models to interfaces, and extends the knowledge used, such as visual icons, through its own simple interface. The model shows that the commercial system is faster and requires less visual, procedural, and declarative knowledge. A related system (VisiGaze) has been created to track eye movements in dynamic environments. These results suggest that engineering design based on user models is increasingly possible through modeling and testing (including high-level behavior representation languages) because VisiTor and VisiGaze may provide design patterns for this work. This work provides new types of insights (e.g., visual icons might need to be measured and taught), and that several new areas can be explored including error generation, error correction, and visual search can be more directly, and easily, and appropriately modeled. This approach also lets situations in which “milliseconds matter” to be more directly studied and predicted because now knowledge in the world (vs. in the head) will be more easily and more often available. All of which provides new topics for interactive cognition.

This is an in-person presentation on July 20, 2026 (09:20 ~ 09:40 EDT).

Agents interacting in dynamic environments face uncertainty about environmental states. This poses a fundamental challenge to cognition. In interactions with other agents, this uncertainty extends beyond sensor noise and changing environments to include the beliefs and preferences of those agents, which are not directly observable. Therefore, agents require an approach to model the uncertainty of these diverse beliefs and utilize them in cognition. This talk presents approaches to modeling uncertainty and explores their implications across physical, social, and introspective dimensions. From a physical perspective, quantifying uncertainty allows agents to evaluate potential actions by balancing pragmatic utility with epistemic information gain. This allows them to build a more accurate mental representation of their environment. Socially, modeling the unobservable beliefs and preferences of interaction partners fosters behavioral alignment and shared understanding. Introspectively, representing uncertainty regarding the agents goals and preferences can promote flexible and potentially safer behavior. To ground these theoretical concepts, we outline a planned application involving a humanoid robot in a dynamic factory setting. We will explore how integrating uncertainty modeling concepts could support the robot with real-time spatial reasoning, collaborative workflows, and safe task execution alongside human workers.

This is an in-person presentation on July 20, 2026 (09:40 ~ 10:00 EDT).

In dialogic interactions between humans and cognitive agents - typically implemented using large language models (LLMs) - the stateless nature of these models prevents the agent from building up experience and establishing a common ground with the human partner. Statelessness here means that these models do not store memories of previous conversations; there is no awareness of prior interactions. This raises the question of how experiences, knowledge, and insights gained through ongoing interactions between individuals and cognitive agents in their physical and social environments can be persistently stored by the agent, retrieved in a contextually appropriate manner during subsequent interactions, and effectively utilized as memories. The Adaptive Control of Thought-Rational (ACT-R) cognitive architecture, with its symbolic and sub-symbolic capabilities for retrieving information from declarative memory, enables the realization of human-like memory functions. The LLM used for the agent’s conversational contributions generates keywords related to the current conversation content as well as the actual recollection phrase. This is stored in a chunk of declarative memory and the recollection can be recalled associatively via the keywords. The persistent storage of memory chunks - including the activations of individual chunks generated through base-level learning - enables the formation of long-term memory and learning through preferred memories with higher activation over extended periods, as well as the ability to temporarily shut down the agent, restart it later, and continue to utilize the memories. Such shared experiences, made possible by shared memories and common ground, foster a deeper interaction between humans and machines, in which humans feel more perceived, thereby enriching human-robot interaction with a sense of connection and togetherness.

This is an in-person presentation on July 20, 2026 (10:00 ~ 10:20 EDT).

Interactive cognition refers to the dynamic co-evolution of thought and action within continuous exchanges between an agent and its physical, digital and social environment. Unlike traditional models that treat cognition as an internal, representational process, interactive cognition emphasizes the situated, embodied, and temporally extended nature of thinking. This perspective foregrounds two interrelated principles. The sense of control in continuous interaction compares expected outcome with sensed outcome on different levels, e.g. it emerges from ongoing sensorimotor loops but also driven by high level intentions – on both levels sense of control is influencing behavior. Constructive cognition highlights that understanding and problem-solving are actively built through interaction—agents shape their cognitive landscapes by selecting, comprehending, and projecting environmental information in real time. Together, these principles frame cognition as an emergent, adaptive process unfolding through participation rather than computation alone. Developing systems that embody these principles—whether human, artificial, or hybrid—requires rethinking representation, agency, and knowledge as inherently interactive phenomena.

This is an in-person presentation on July 20, 2026 (10:40 ~ 11:00 EDT).

Divergent thinking (DT) is the ability to generate multiple, varied, and novel ideas in response to an open-ended problem. DT has been shown to improve with practice. The goal of the current study was to determine whether peer-assisted learning (PAL) is superior to individual learning (IL) in improving DT. Sixty-six participants were randomly assigned to complete a modified version of the Divergent Association Task (Olson et al., 2021) either alone (IL condition) or with a partner in a dyad (PAL condition). Over 10 rounds, participants in the IL condition generated multiple words as they compiled lists of words that were as unrelated to each other as possible, whereas PAL participants generated half the words with the other half supplied by their partners. Overall, IL participants achieved higher DT scores than PAL participants and both groups improved their scores with practice; however, PAL participants improved at a faster rate than IL participants, starting lower but reaching the IL level and slightly exceeding it in the last three rounds. A behavioral measure of trust was used to explain PAL’s initial disadvantage and its faster learning rate. Initially, PAL participants selected self-generated words much more frequently than peer-generated words, even though the peer-generated words were more divergent and would have improved their scores. However, with practice, their trust increased allowing them to benefit from more divergent peer-generated words. This study contributes to advancing the science of interactive cognition by demonstrating that cognitive and interpersonal processes are dynamically coupled.

This is an in-person presentation on July 20, 2026 (11:00 ~ 11:20 EDT).