Cognitive-Level Salience for Explainable Artificial Intelligence
Cognitive-Level Salience for Explainable Artificial Intelligence
复制标题
可解释人工智能的认知层面显着性
DOI：
--
复制
发表时间：
--
期刊：
影响因子：
--
通讯作者：
Robert Thomson
中科院分区：
文献类型：
--
作者：
Sterling Somers;Robert Thomson
我们提出了一种通用的方法，用于确定人工智能代理的动作决策中的特征显著程度。我们的方法不依赖于人工智能的特定fic实现(例如，深度学习、符号人工智能)。该方法也适用于不同抽象级别的特征。我们给出了我们的显著技术的三个实现：两个针对可解释的人工智能(深度强化学习代理)，第三个针对风险评估。
We present a general-purpose method for determining the salience of features in action decisions of artificial intelligent agents. Our method does not rely on a specific implementation of an AI (e.g. deep-learning, symbolic AI). The method is also amenable to features at different levels of abstraction. We present three implementations of our salience technique: two directed at explainable artificial intelligence (deep reinforcement learning agents), and a third directed at risk assessment.