Скачать или смотреть The Cognitive Bottleneck: Solving Decision Conflict in AI systems

This transcript explains how human minds and AI architectures manage the cognitive bottleneck, a limitation that allows for only one conscious action at a time. To solve this, systems like ACT-R utilize "specialists" known as production rules that compete to perform specific tasks. Decision-making is governed by utility scores, which represent a rule's calculated value based on past success, penalties, and a touch of random noise. These scores are processed through a softmax function to determine the probability of an action rather than a guaranteed outcome. Because these values update via reinforcement learning, the system constantly adapts to prioritize more successful strategies. Ultimately, this framework illustrates how rational behavior emerges from a competitive, probabilistic process of trial and error.