Вештачка интелигенција

The Pentagon Introduces Responsible Artificial Intelligence Toolkit

Summary

The Department of Defense (DoD) Pentagon has unveiled a toolkit for responsible artificial intelligence (AI) to continue promoting the ethical use of this technology across the defense sector and explore new applications for it, while mitigating potential harm. The “Responsible […]

The Pentagon Introduces Responsible Artificial Intelligence Toolkit

The Department of Defense (DoD) Pentagon has unveiled a toolkit for responsible artificial intelligence (AI) to continue promoting the ethical use of this technology across the defense sector and explore new applications for it, while mitigating potential harm.

The “Responsible Artificial Intelligence Toolkit” provides a centralized process that identifies, monitors, and enhances the compliance of AI projects with DoD’s best practices and ethical principles, while also leveraging opportunities for innovation, as stated in the DoD’s report released last week. The technology also enables users to incorporate transparency and verification concepts throughout the AI development cycle.

As part of this toolkit, the Department of Defense will have an AI incident repository and an AI incident database. The repository, which is still under development, will collect incidents and failures related to AI to further enhance AI development in the future. The database will index the collective history of damages or potential damages in the real world due to the application of AI. This database aims to help the DoD learn from past experiences and prevent or mitigate harmful consequences.

As part of DoD’s ongoing efforts to reduce risks, this AI toolkit also includes a risk documentation tool that identifies and mitigates risks associated with the use of AI. It also provides a guide for addressing questions that arise during the creation of AI systems.

“A responsible approach to artificial intelligence means innovating at a pace that can surpass existing and emerging threats, and with a level of efficiency that provides justified security in the technology and its use,” the background information of the toolkit states. “Responsible AI means ensuring that our technology aligns with our values. This enables our nation to lead in technological innovation while remaining committed advocates of our democratic principles. To achieve this goal, RAI work within DoD translates ethical principles into concrete objectives for each use case. The toolkit provides resources to accomplish this task and is designed to be accessible, modular, and adaptable.”

Moreover, the DoD aims to minimize harm. One of the 70 tools in the toolkit is an audit tool that helps DoD examine, report, and mitigate discrimination and bias in machine learning models by providing metrics for AI tools. Additionally, there is a set of tools to reduce human bias in AI projects. The Responsible Artificial Intelligence Toolkit provides a framework for mitigating AI and machine learning threats.

The toolkit includes tools that assist throughout the project. For example, there is a tool for setting responsible AI goals for a project and a guide for senior leadership to assess AI project managers. There is also a tool that defines and establishes roles and responsibilities for AI projects. Another tool provides examples of AI usage. This use case application tool is utilized after the Pentagon created Project LIMA to explore the application of generative AI.

The toolkit was released last week, following the adoption of ethical principles for AI by the DoD in 2020, and the adoption of the RAI strategy and implementation path in 2022, which defined over 60 DoD efforts to support its ethical principles. The Responsible Artificial Intelligence Toolkit helps the DoD implement its strategy and implementation plan. It also comes after the DoD’s new AI strategy, which was published earlier this month, focusing on agile implementation of AI across the entire DoD. The toolkit simultaneously aligns with the efforts of the Office of the DoD Chief Digital and Artificial Intelligence Officer to enhance AI knowledge throughout the department via a digital learning platform.

FAQ: