Overview:
Scientists have developed computer algorithms to forecast daily occurrences, demonstrating that addressing uncertainty—rather than solely focusing on prediction mistakes—enhances comprehension.
This finding questions the notion that unexpected events are the sole catalysts for understanding and indicates that the brain might employ two strategies.
Concurrently, memory research indicates that recognizing event boundaries can improve recall, particularly among older individuals. Current investigations are focused on enhancing memory by aiding people in better identifying these boundaries.
The results could pave the way for strategies to combat age-related memory decline and deepen our insight into cognitive mechanisms. This research underscores the complex relationship between how we segment events and store memories.
Essential Information:
Simulations indicate that uncertainty enhances our understanding of daily occurrences. Recognizing the boundaries of events is a strong indicator of how well memories are retained.
Many older individuals face challenges in processing events, which is associated with a decrease in memory function.
Life consists of countless minor moments: brewing coffee in the morning, taking the dog outside, booting up a laptop, and then bringing the dog back inside. When you tally these moments together, they form an entire day.
According to Jeff Zacks, the Edgar James Swift Professor in Arts & Sciences and chair of the Department of Psychological & Brain Sciences, our minds are dedicated to noticing and interpreting the happenings that shape our everyday existence. Zacks emphasizes that grasping where events start and finish is essential for comprehending our surroundings.
In two recent publications, Zacks and colleagues from the Arts & Sciences and McKelvey School of Engineering investigate a crucial aspect of human thought processes.
Zacks directed a research project that involved training computer models on over 25 hours of footage showing individuals engaged in basic daily activities, like tidying up a kitchen or preparing food, to forecast subsequent actions.
The research yielded an unexpected finding: the computer models demonstrated their highest accuracy when they addressed uncertainty. When faced with significant uncertainty about future events, the models would reset and reevaluate the situation, which enhanced their overall understanding.
The study’s co-authors, set to be featured in an upcoming issue of PNAS Nexus, include Tan Nguyen, a graduate student at Zacks’s Dynamic Cognition Laboratory; Matt Bezdek, a senior scientist in the same lab; Aaron Bobick, dean of the McKelvey School of Engineering and James M. McKelvey Professor; Todd Braver, who holds the William R. Stuckenberg Professorship in Human Values and Moral Development; and Samuel Gershman from Harvard University’s neuroscience department.
Zacks had earlier posited that human brains are particularly sensitive to minor surprises encountered in daily life. He suggested that individuals reassess their environment whenever they notice something unexpected—a concept referred to as prediction error.
However, this new discovery that effective computer models prioritize uncertainty over prediction errors challenges that previous theory. Science is about revising theories based on fresh evidence, Zacks remarked.
Nguyen added that while surprises remain significant and there is no need to discard prediction error entirely, researchers are beginning to consider that the brain might utilize both strategies rather than choosing one over the other. Each model offers distinct insights into our understanding of human cognition.
Maverick Smith, a postdoctoral researcher at the Dynamic Cognition Lab, is exploring the relationship between event understanding and memory retention.
Collaborating with Heather Bailey, a former postdoc from WashU who is currently an associate professor at Kansas State University, Smith co-authored a review published in Nature Reviews Psychology.
This article compiles increasing evidence that long-term memory is closely linked to one’s ability to accurately recognize the boundaries of events. According to Smith, there are significant individual variations in recognizing when events begin and end, which can be strong indicators of how well people remember those events later on.
He expressed hopes of developing an intervention aimed at enhancing memory by assisting individuals in segmenting events more effectively. Similar to Zacks’ approach, Smith utilizes video clips to gain insights into how the brain interprets events; however, his footage features everyday activities like shopping or setting up a printer rather than cooking or cleaning.
In his experiments, participants indicate the start and finish of specific events by pressing buttons while their subsequent recall is assessed through written questions about the clips.
Smith has observed that older adults often struggle more with processing events—a challenge that may contribute to age-related memory decline. He suggested that there might be strategies we could implement to help them retain memories of important life occurrences more effectively.
Zacks, Nguyen, Smith, and their colleagues in the Department of Psychological & Brain Sciences have set ambitious goals to deepen their knowledge of how the brain processes and retains memories of events.
Zacks’ team is utilizing fMRI brain imaging to observe the real-time reactions of 45 participants as they watch videos depicting everyday situations. We are investigating the actual neural dynamics involved in these cognitive functions, Zacks stated.
Additionally, another study focuses on monitoring eye movements to gain fresh insights into our perception of the world. Zacks noted that when individuals view common activities, they tend to concentrate significantly on people’s hands.
Meanwhile, Smith is conducting video-based experiments aimed at enhancing memory in participants—including older adults and those with Alzheimer’s—by clarifying event boundaries. His ultimate goal is to comprehend how observations of events are encoded and preserved in long-term memory.
Some individuals excel at breaking down events into meaningful segments, Smith remarked. Can we enhance this skill, and will it lead to better memory retention? These are the questions we continue to explore.
Understanding human activities on a human scale:
Forecasting, partitioning, and classification Humans create models of events that represent their current circumstances to anticipate the progression of activities.
Various theories exist regarding how the cognitive system identifies when to break down behaviors into segments and transition between different active event models.
In this study, we developed a computational model that acquires knowledge about different event types (event schemas) by integrating recurrent neural networks for short-term behavior dynamics with Bayesian inference for transitions between events.
This framework utilizes event schemas to create a range of event models. It was trained on a single pass through 18 hours of real-life human activities, while an additional 3.5 hours were allocated for testing each variant’s alignment with human segmentation and categorization.
The architecture demonstrated the ability to predict human activities, achieving segmentation and categorization levels that were nearly comparable to those of humans.
We then evaluated two variations of this framework aimed at more accurately mimicking human event segmentation: one variant transitioned when the active event model exhibited high uncertainty in its predictions, while the other transitioned in response to significant prediction errors.
Both variants effectively learned to segment and categorize events; however, the uncertainty-based variant showed a closer resemblance to human segmentation and categorization, despite lacking feedback on these processes.
These findings indicate that transitioning between event models based on prediction uncertainty or error can replicate key aspects of how humans understand events.