Education DCL: EAGER: Developing Experiential Cybersecurity and Privacy Training for AI Practitioners

人工智能(AI)和人工智能支持的工具在开发和使用方面都获得了发展势头，并在工作场所变得越来越普遍。然而，许多人工智能从业者没有意识到与构建基于人工智能的系统相关的网络安全和隐私风险，例如对机器学习模型的对抗性攻击，或者与使用基于人工智能的系统围绕社会问题进行决策相关的隐私和道德风险。该项目的目标是通过开发和评估一个全面的12个研讨会的体验式培训计划，提高人工智能工作者的安全和隐私风险意识。研讨会系列将提供构建人工智能系统所需的知识和技能，这些系统不仅从人工智能的角度来看在技术上是健全的，而且是安全的、道德的和隐私保护的。评估后改进的材料版本将向更广泛的社区提供，使该项目有可能广泛提高人工智能员工的技术知识和网络安全意识。研讨会将遵循体验式学习模式，并将由专家设计、组织和提供，以实现学习目标。这些目标围绕五个主要模块进行分组，旨在涵盖围绕人工智能模型的广泛安全和隐私问题：基础和威胁；对手攻击和稳健性；隐私、道德和信任；安全发展和数据治理；以及案例研究。每个单元将在两到三个两小时的工作坊课程中讲述。每个研讨会以一个小时的网络研讨会或专家小组辩论或讨论研讨会的关键主题开始，然后是一个小时的体验式学习部分。第二个部分将包括使用真实世界示例的演示或要求参与者完成实验练习的动手练习，该练习建立在第一个小时涵盖的材料的基础上。然后，参与者将向其他参与者演示他们的实验工作，或者写下他们所学到的反思。项目团队将运行三次研讨会系列，每个系列之后都有一个评估和迭代周期，以改进材料。总之，这项工作将使人们更好地理解如何构建更值得信赖的基于人工智能的系统，以及如何将安全、隐私和道德培训作为技术课程的一部分。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Artificial Intelligence (AI) and AI-powered tools have gained momentum in both development and usage and are becoming increasingly prevalent in the workplace. However, many AI practitioners are not aware of the cybersecurity and privacy risks associated with building AI-based systems, such as adversarial attacks on machine learning models, or privacy and ethics risks associated with using AI-based systems for decision-making around social issues. This project's goal is to raise AI workers' awareness of security and privacy risks by developing and evaluating a comprehensive 12-workshop experiential training program. The workshop series will provide knowledge and skills needed to build AI systems that are not only technically sound from an AI perspective but also secure, ethical, and privacy-preserving. Versions of the materials that have been improved after the evaluation will be made available to the wider community, giving the project the potential to widely increase the AI workforce's technical knowledge and cybersecurity awareness.The workshops will follow an experiential learning model and will be designed, organized, and delivered by experts to achieve the learning objectives. These objectives are grouped around five main modules designed to cover a wide space of security and privacy concerns around AI models: Fundamentals and Threats; Adversarial Attacks and Robustness; Privacy, Ethics, and Trust; Secure Development and Data Governance; and Case Studies. Each module will be covered in two to three two-hour workshop sessions. Each workshop starts with one hour of a webinar or a panel of experts debating or discussing the workshop's key topics, followed by one hour of experiential learning component. That second component will consist of either a demo using real-world examples or a hands-on activity asking participants to complete a lab activity, which builds on the materials covered in the first hour. Participants will then present a demonstration of their lab work to other participants or write a reflection on what they learned. The project team will run the workshop series three times, with an evaluation and iteration cycle after each series to improve the materials. Together, the work will lead to a better understanding of both building more trustworthy AI-based systems and how to incorporate security, privacy, and ethics training as part of technical curricula.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.