A Technique for Emulating Human Recall Timing in Artificial Intelligence

Abstract

Human recall of items from a specified semantic category, such as breeds of dogs, follows a characteristic pattern in which early responses tend to occur quickly, whereas later responses occur more slowly. Artificial intelligence (AI) systems, by contrast, typically produce items with relatively uniform latency. This paper presents the AI Recall Timing Model, a computational technique for applying human-like recall timing to AI-generated lists of items from a specific category without altering their supplied order. The model repeatedly samples numeric timing placeholders and rejects those that have already been selected. As more unique placeholders are obtained, duplicates become increasingly likely, requiring more attempts to produce each new item and naturally generating progressively longer interresponse times. The model also supports semantic clustering by allowing a contiguous group of related items to share one timing placeholder, with the group members emitted in their supplied order without separate sampling attempts. Across repeated runs, the model’s average per-item attempts curve converges toward the uniform coupon collector per-item expectation. This convergence provides a mathematical explanation for why repeated sampling with rejection produces the characteristic timing shape observed in averaged human recall. The technique allows timing and order to remain separate: the input supplies the item sequence, while the model supplies the timing pattern. v39: Restructured the paper to present the human recall results and AI Human Recall Model before the mathematical expectation and reworded some parts of the paper for improved readability.