S&AS: FND: Norm Processing for Autonomous Social Systems

S

• 批准号: 1723963
• 负责人: Matthias Scheutz
• 金额: $ 60万
• 依托单位: Tufts University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2017
• 资助国家: 美国
• 起止时间: 2017-09-01 至 2021-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1723963&HistoricalAwards=false
• 关键词: S& FND; Norm; Processing; Autonomous

As intelligent systems continue to become more autonomous, it is important to imbue such systems with the types of social and moral norms that are deeply ingrained in human cognition and behavior. Failing to abide by these norms typically causes social reactions from humans, from blame and reprimands, in simple cases, all the way to full-fledged legal consequences. Given the fundamental social expectations humans have of each other, it is very likely that they will extend and apply those to artificial agents as well. Hence, intelligent autonomous systems will need to be endowed with computational mechanisms that ensure their ethical behavior.This project will develop explicit norm representations and planning algorithms for norm-conforming behavior that will address shortcomings of previous approaches, enabling explicit, temporally complex norm representations in stochastic worlds that allow for systematic handling of norm conflicts in ways that maximize the norms an agent can obey at any moment in time. In particular, explicit norm specifications will allow for generalization across unobserved states. In addition, it is important for systems interacting with humans to make decisions that are transparent. To that end, the approach will enable the generation of justifications in cases where an autonomous system is forced to violate some norms because not all can be satisfied simultaneously. The approach and algorithms developed will be demonstrated on a fully autonomous cleaning robot with predefined norms in different scenarios, where conflicting norms are applicable and the robot must determine which norms to follow and which to suspend.

随着智能系统变得越来越自主，重要的是要将这些系统与人类认知和行为中根深蒂固的社会和道德规范相结合。 不遵守这些规范通常会引起人类的社会反应，在简单的情况下，从指责和谴责，一直到全面的法律的后果。 考虑到人类对彼此的基本社会期望，他们很可能会将这些期望扩展并应用于人工智能。 因此，智能自治系统将需要被赋予确保其道德行为的计算机制。本项目将为符合规范的行为开发明确的规范表示和规划算法，以解决以前方法的缺点，随机世界中的时间复杂规范表示，允许系统地处理规范冲突，以最大化代理可以遵守的规范，随时都可以 特别是，明确的规范规范将允许在未观察到的状态的泛化。 此外，与人类交互的系统必须做出透明的决策。 为此，该方法将能够在自治系统被迫违反某些规范的情况下产生正当性，因为并非所有规范都能同时得到满足。 开发的方法和算法将在完全自主的清洁机器人上进行演示，该机器人在不同的场景中具有预定义的规范，其中冲突的规范是适用的，机器人必须确定遵循哪些规范以及暂停哪些规范。

项目成果

Norms, Rewards, and the Intentional Stance: Comparing Machine Learning Approaches to Ethical Training

规范、奖励和意图立场：机器学习方法与道德培训的比较

• DOI: 10.1145/3278721.3278774
• 发表时间: 2018
• 期刊: and Society
• 作者: Kasenberg, Daniel;Arnold, Thomas;Scheutz, Matthias
• 通讯作者: Scheutz, Matthias

Generating Explanations for Temporal Logic Planner Decisions

• DOI: 10.1609/icaps.v30i1.6740
• 发表时间: 2020-06
• 作者: Daniel Kasenberg;Ravenna Thielstrom;matthias. scheutz

A Touching Connection: How Observing Robotic Touch Can Affect Human Trust in a Robot

感人的联系：观察机器人触摸如何影响人类对机器人的信任

• DOI: 10.1007/s12369-020-00729-7
• 发表时间: 2021
• 期刊: International Journal of Social Robotics
• 影响因子: 4.7
• 作者: Law, Theresa;Malle, Bertram F.;Scheutz, Matthias
• 通讯作者: Scheutz, Matthias

Engaging in Dialogue about an Agent’s Norms and Behaviors

参与有关代理人规范和行为的对话

• DOI: 10.18653/v1/w19-8406
• 发表时间: 2019
• 期刊: Proceedings of the 1st Workshop on Interactive Natural Language Technology for Explainable Artificial Intelligence
• 作者: Kasenberg, Daniel;Roque, Antonio;Thielstrom, Ravenna;Scheutz, Matthias
• 通讯作者: Scheutz, Matthias

Generating justifications for norm-related agent decisions

为与规范相关的代理决策提供理由

• DOI: 10.18653/v1/w19-8660
• 发表时间: 2019
• 期刊: proceedings of the 12th international conference on natural language generation
• 作者: Kasenberg, Daniel;Roque, Antonio;Thielstrom, Ravenna;Chita-Tegmark, Meia;Scheutz, Matthias
• 通讯作者: Scheutz, Matthias