Prospective Memory Degradation

• Prospective memory degradation is the decline in the ability to execute future intentions, impacting both human cognition and AI system performance.
• Empirical research shows that aging, cognitive load, delay intervals, and cue focality critically impair prospective memory, as measured by VR and neuropsychological tests.
• Interventions like VR-based strategy training and context-aware reminder systems have shown promise in mitigating prospective memory failures in diverse settings.

Prospective memory degradation refers to the impaired capacity to remember to carry out delayed intentions in the future, especially when those intentions are embedded within complex, dynamic environments or under conditions of cognitive, perceptual, or contextual stressors. It is an essential construct in cognitive science, aging, neuropsychology, human-computer interaction, and artificial intelligence, as it affects autonomy, functional independence, safety, and system performance across biological and synthetic agents.

1. Conceptual Foundations and Operational Definitions

Prospective memory (PM) is the ability to execute intended actions after a temporal delay, often embedded within ongoing activity streams. Degradation in PM manifests as increased frequency or severity of failures to execute these future intentions—examples include missing a medication dose, not resuming an interrupted task, or forgetting to respond to time- or event-based cues. PM is defined against two principal taxonomies:

• Event-based PM: Triggered by a specific external cue (e.g., "when you see Ana, give her the key").
• Time-based PM: Triggered by an internal schedule (e.g., "call the lab at 14:00"). PM degradation may be observed both in healthy aging and in neurologic or psychiatric impairment, as well as in artificial agents subjected to limited memory or context resets.

A further division is drawn between:

• Semantic PM (knowledge for stable, general intentions; robust to context changes).
• Episodic PM (memory for temporally sequenced, contextually linked intentions; more fragile and context-dependent).

2. Mechanisms and Determinants of Degradation

Empirical and computational research converges on a multidimensional etiology for PM degradation:

Biological and Cognitive Bases

• Aging: Older adults consistently show reduced PM, with prominent deficits in time-based and irregular intentions, greater variability, and increased fatigue in performance contexts (e.g., immersive VR training systems) (Fukumori et al., 2024).
• Cognitive Load: Complexity of ongoing activity (dual-tasking, distractions) exacerbates PM lapses by consuming executive and attentional resources (Hou, 2016).
• Delay Interval: Long delays between intention formation and cue trigger are a primary driver of PM failures. In immersive, ecologically valid VR scenarios, task type effects are secondary to retention interval length (Kourtesis et al., 2021).
• Cue Focality: The semantic relatedness of the triggering cue to the intended action moderates delay effects. Focal, event-based cues are more resistant to delay-associated PM decline than non-focal or time-based cues (Kourtesis et al., 2021).

Computational and Artificial Models

• Memory Decay and Update Dynamics: In sequential decision-making agents, PM degradation results from the balance of memory decay rate ($\mu$) and policy update rate ($\beta$):
  • If $\mu \ll \beta$, agents nearly always perform the optimal future-intended action; if $\mu \gg \beta$, action selection degrades to a randomized distribution, eroding PM capacity (Xu et al., 2019).
• Session Reset and Context Persistence in LLMs: Generative AI systems, such as LLMs, exhibit semantic memory robustness but episodic continuity collapse upon session reset. The Artificial Age Score (AAS), an entropy-based metric, quantifies this structural memory aging by tracking recall decline only in observable agent outputs (Kayadibi, 24 Sep 2025).

Environmental and Technological Factors

• Technological Disruptors: Short-form video streams (e.g., TikTok) with rapid context switching dramatically degrade human PM post-interruption, with a ~40% relative decrease in task performance, mediated by attentional disengagement and working memory overload (Chiossi et al., 2023). Conventional digital reminders often fail to adapt to user context or optimize notification timing, resulting in both omission and annoyance.

3. Empirical Assessment of Prospective Memory Degradation

Standardized Neuropsychological and Experimental Platforms

• Memory for Intentions Screening Test (MIST): Used to baseline and track PM capacity in both healthy and clinical populations (Fukumori et al., 2024).
• Immersive VR Batteries: Systems such as the Virtual Reality Everyday Assessment Lab (VR-EAL) achieve high ecological validity by simulating complex, multitasking-laden, naturalistic environments to measure both event- and time-based PM, allowing fine-grained assessment of delay and cue-type effects (Kourtesis et al., 2021, Kourtesis et al., 2021).

Assessment Method	Features	Outcome Measures
MIST	Standardized, clinical PM battery	Accuracy/success in executed intentions
VR-EAL	Real-life, multitasking VR protocols	PM via functioning/ability, cue-type × delay × performance
Drift Diffusion Models	Psychometric and cognitive process models	Drift rate, decision bound, variance post-interruption (Chiossi et al., 2023)

Data Streams and Measurement in Assistive Robotics

• Wizard-of-Oz Paradigms: Human-in-the-loop simulation for SARs allows direct manipulation of context sensitivity and real-time corrective intervention, facilitating the logging and classification of both user and system-origin PM failures during shared activities such as music-listening and tea-making (Li et al., 2023).

4. Cognitive, Computational, and Environmental Predictors

Cognitive Processes

• Delayed Recognition: Dominant for event-based PM; robust predictor of real-world PM failures (Kourtesis et al., 2021).
• Planning and Executive Function: Critical for time-based PM, especially in the context of competing tasks or when intention must be retained without salient external cues.
• Visuospatial Attention: Both speed and accuracy modulate PM, especially for complex or visually demanding environments.
• Imagery Ability: Strongly correlates with improvement in prospective remembering, as shown in VR-based multimodal training (Fukumori et al., 2024).

Artificial and Hybrid Systems

• Shared and Situated Memory in SARs: Implementation of shared, context-sensitive, and state-aware PM models allows robots to adaptively generate or suppress reminders, mitigating PM failures more effectively than rigid, schedule-based prompts (Li et al., 2023).
• Entropy-based Aging Metrics (AAS): Task-independent, formally bounded scores quantify AI memory aging by linkage to behavioral recall metrics—robust against internal state opacity (Kayadibi, 24 Sep 2025).

5. Interventions, Mitigation, and Design Considerations

Remediation and Training

• VR-Based Strategy Training: Systematic progression through increasingly complex, ecologically valid PM tasks, especially with visualization strategies, can measurably improve PM even in older adults, though fatigue may rise (Fukumori et al., 2024).
• Optimal Reminder Systems: Psychologically informed, algorithmically optimized reminder platforms dynamically compute the number, schedule, and modality of reminders, adapting to user-specific factors (task complexity, age, motivation), minimizing both omission and annoyance (Hou, 2016).

Human-AI System Design

• Context-Aware Assistive Robotics: Resilient SAR designs use multimodal shared memory and context tracking to deliver just-in-time, non-redundant reminders, supporting user autonomy and reducing both unnecessary prompts and missed intentions (Li et al., 2023).
• Media Platform Architecture: Explicit design minimization of rapid, continuous context switching (as in short-form video feeds) is necessary to preserve users’ prospective memory in both clinical and everyday contexts (Chiossi et al., 2023).

6. Quantitative Characterization and Key Findings

Factor	Quantitative Effect/Key Result
Delay (VR-EAL assessments)	Longer delays (>45 min) sharply amplify PM failures, with focal cues providing resilience (Kourtesis et al., 2021)
Task Type	Time-based and non-focal event-based tasks are disproportionately vulnerable (Kourtesis et al., 2021)
Technology Interruption (TikTok)	~40% decrease in recall of PM intentions after short-form video interruption (Chiossi et al., 2023)
Cognitive Correlates	Correlation (r) between imagery ability and task achievement > 0.8 in most PM task types (Fukumori et al., 2024)
AI Memory Aging (AAS)	Session resets cause episodic memory collapse, sharply increasing AAS; semantic memory preserved (Kayadibi, 24 Sep 2025)
Age	Elderly participants demonstrate reduced overall PM achievement and increased fatigue in VR testing (Fukumori et al., 2024)

7. Implications and Open Directions

The convergent evidence demonstrates that prospective memory degradation is a multideterminant phenomenon influenced by cognitive, contextual, and system-level factors. In both biological and artificial agents, preservation of PM depends critically on memory span, context continuity, executive control, and environmental design. Theoretical and empirical models (including closed-form solutions for memory/action dynamics, VR-based variance decomposition, and entropy-driven structural aging diagnostics for AI) provide actionable pathways for diagnosis, remediation, and next-generation assistive systems.

Persistent open questions concern optimal adaptation/learning algorithms for individualized PM support, quantification of redundancy effects in AI memory aging, and generalization of ecological validity from simulation to daily-life environments in both human and artificial populations.