Symposium

Symposium 4: Advancing Cognition Through Adversarial Collaboration: The Case of Working and Long-Term Memory (Leading Edge Workshop) 
Saturday, November 6, 10 AM - 12 PM CT

Chair: Robert H. Logie (University of Edinburgh, UK)
One of the long-standing debates in cognition concerns how temporary representations and ongoing mental activity are related to the formation and retrieval of more lasting records of events and the accumulation of knowledge. This debate continues and was highlighted in the 2021 Psychonomic Society Leading Edge Workshop on how adversarial collaboration in research between opponents in debate might lead to resolution of those debates rather than perpetuating debate indefinitely. This symposium highlights key areas of research that offer pathways to progress in understanding the relationship between temporary and longer-lasting representations, both as the psychological constructs, working memory and long-term memory, and in terms of neurobiology.

Integrated Cognition from Multiple Components for Mental Workspace and Stored Knowledge

Presenter: Robert H. Logie (University of Edinburgh)
The assumption that working memory (WM) comprises specialized components, distinct from long-term memory (LTM) has been supported by a wide range of data. But, it faces contrary evidence, and questions about how WM components interact with each other and with LTM. Also, WM contents are widely assumed to be interpreted from stored knowledge, not raw sensory images, so WM cannot act as a gateway between sensory input and LTM. A possible solution is that WM is currently activated LTM with rapid learning from the focus of attention. But this faces challenges from studies that demonstrate neuropsychological dissociations, limited impact on performance of dual task demands, and lack of learning from repeated presentation of a stimulus array. Examples will be presented of the latter evidence, and of participants changing how they perform the same task under different experimental conditions. The talk will conclude by arguing that multiple components of cognition, including components of WM and LTM, continuously interact as a flexible integrated system, that different combinations of components may be deployed for any given experimental condition, and that control arises from local interactions between components.

Embedded Processes and the Route from Working-Memory To Long-Term Memory 

Presenter: Nelson Cowan (University of Missouri)
In this talk I clarify concepts of long-term memory, its formation, and its activation within the embedded-processes view of information processing. Working memory is said to comprise activated elements of long-term memory and, embedded within them, the focus of attention. Making that description workable, activated long-term memory can include rapidly-learned information added during an immediate memory trial and used on that same trial, before the new information has ever been inactive. The process of learning involves associations between objects or concepts held concurrently in the focus of attention. In support of this approach, I will discuss research indicating that the instantaneous content of working memory of an array is a good predictor of how much information will be retrievable later. I will also discuss research indicating that when two list items occupy the focus of attention at once, an enduring association forms between them. Unresolved issues will be discussed.

Three Controversies in Event Perception 

Presenter: Jeffrey M. Zacks (Washington University in St. Louis)
It is well established that people form stable representations of current ongoing events—event models—and update those representations when events change. The updating of event models has been linked to the conscious experience of event segmentation, and to changes of availability of information in memory. However, there are several current controversies about the mechanisms of event model updating. What determines when event models are updated? Is event model updating an all-or-none proposition, or is updating incremental? Are some event boundaries more salient or stronger than others, or is being an event boundary an all-or-nothing proposition? In this talk I will introduce each of these issues, describe the current state of the data, and try to highlight how these constructive debates are furthering our understanding of perception and memory.

Prediction Errors Disrupt Hippocampal Representations and Update Episodic Memories 

Presenter: Morgan D. Barense (Rotman Research Institute)
How does the brain link past, present, and future? The concept of predictive coding provides a framework that bridges memory and perception. We draw on past experience to make predictions, and then compare those predictions to present perceptual input. This comparison process allows the brain to segment continuous experience, learn from error, and adaptively integrate new information into memory. Past studies have demonstrated that the hippocampus signals prediction error, or surprise, but have not linked this neural signal to memory updating. In my talk, I will provide evidence for this missing connection. Using fMRI, we elicited prediction errors by interrupting familiar narrative videos immediately before the expected endings. We found that the same amount of hippocampal activity exerted opposing effects on memory: hippocampal activity preserved memories after expected endings, but updated memories after prediction errors. We examined the mechanisms of this processing shift, showing that prediction errors disrupt the temporal continuity of hippocampal patterns. We conclude that prediction errors create conditions that favor memory updating, prompting the hippocampus to abandon ongoing predictions and render memories malleable.

A Computational Systems Neuroscience Perspective on Interacting Memory Systems 

Presenter: Randall C. O'Reilly (University of California, Davis)
Neocortex is densely interconnected. While studies have identified coactivated networks of brain areas, it is difficult to argue that any significant cognitive function is supported exclusively by any one area or network. This presents a challenge for scientific theory: how do these networks interact, and in what ways are they specialized? Biologically-based computational modeling has addressed this problem by developing models based on specific anatomical features of different areas, and their interconnectivity, e.g., we can identify hippocampal properties that are critical to rapidly form new episodic memories, and can understand the nature of these memories in terms of separate dorsal and ventral pathways feeding into the hippocampus. Prefrontal cortex has neural specializations supporting active maintenance, long associated with the psychological construct of working memory, but hippocampal episodic memory can support similar cognitive demands. We are developing large-scale systems neuroscience models that learn ground-up through predictive learning, in ecologically-based foraging-like environments, to gain insight into how such systems might interact more organically to support a range of basic cognitive functions.