Ontologies for the Physical Turing Test

物理图灵测试本体

• 批准号: RGPIN-2020-05781
• 负责人: Gruninger, Michael
• 金额: $ 2.99万
• 依托单位: University of Toronto
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2022
• 资助国家: 加拿大
• 起止时间: 2022-01-01 至 2023-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=759653
• 关键词: Ontologies; Physical; Turing; Test

The Turing Test was originally proposed as a way of evaluating how well a software system represents human intelligence, based on natural language conversations.  It has recently been recognized that a different approach is needed that incorporates capabilities required for general intelligence. The Visual Turing Test extends the Turing Test to include perception -- it specifies an open domain task of question answering based on corpora of images depicting real-world scenes.  The Physical Turing Test extends this further by specifying questions that require the integration of perception, reasoning, and action. To capture the commonsense knowledge required to perform the Physical Turing Test, existing approaches to computer vision and robotics have begun to include simple knowledge representation schemes such as WordNet.  However, such approaches cannot fully specify the intended semantics of the terminology,nor can they support automated reasoning.  What is needed is an ontology, that is, a computer interpretable specification that is used by an agent, application, or other information resource to declare what terms it uses, and what the terms mean. In this project, we will design new ontologies in first-order logic that can represent objects in the physical world, the activities in which these objects can possibly participate, and how to recognize these objects and the spatial relationships between them within images. (We will refer to these as the PRAxIS Ontologies, for Perception, Reasoning, and Action across Intelligent Systems) These ontologies will be integrated to support reasoning about physical objects, their behaviours (e.g. the ways in which they can change), and their interactions (e.g.manipulation of solid physical objects). Ontologies for perception will be tested using question-answering based on real-world images, such as detecting and localizing instances of objects and relationships among the objects in a scene. Ontologies for action will be tested through the representation of instructions for assembling, using, and maintaining physical objects, including devices and everyday objects such as furniture To support the question-answering for the Physical Turing Test, this project will also investigate reasoning (theorem proving and constraint satisfaction) with first-order ontologies.

图灵测试最初是作为一种基于自然语言对话来评估软件系统代表人类智能的方式而提出的，最近人​​们认识到需要一种不同的方法来整合通用智能所需的功能。图灵测试包括感知——它指定了基于描绘现实世界场景的图像语料库的开放域问答任务，物理图灵测试通过指定需要集成感知、推理和行动的问题来进一步扩展这一任务。为了捕捉执行物理图灵测试所需的常识知识，现有的计算机视觉和机器人技术已经开始包括简单的知识表示方案，例如WordNet，但是这些方法不能完全指定术语的预期语义，也不能支持自动推理。我们需要的是一个本体，即代理、应用程序或其他信息资源使用的计算机可解释的规范，以声明它使用的术语以及这些术语的含义。在这个项目中，我们将设计新的本体。一阶逻辑可以表示物理世界中的对象，这些对象可能参与的活动，以及如何识别这些对象以及它们在图像中的空间关系（我们将这些称为 PRAxIS 本体论，用于感知、推理、和跨智能系统的行动）这些本体将被集成以支持对物理对象、它们的行为（例如它们可以改变的方式）以及它们的交互（例如用于感知的实体本体的操纵）的推理。将使用基于真实世界图像的问答来进行测试，例如检测和定位场景中的对象实例以及对象之间的关系将通过组装、使用和维护物理指令的表示来进行测试。为了支持物理图灵测试的问答，该项目还将研究一阶本体的推理（定理证明和约束满足）。