Since the seminal FAIR principles for research data publication were introduced in 2016, considerable effort has been devoted to making datasets and data repositories findable, accessible, interoperable, and reusable. In 2020, the FAIR envelope was broadened to include research software. However, in a practical sense, there is still a large gap between the first half of the acronym, FA (findability and accessibility), and the last half, IR (interoperability and reusability). While data and software repositories are used daily by researchers, providing findability and accessibility to proliferating research data and software, interoperability and reusability depend on a critical additional factor. This element is often referred to as the run-time environment (RTE), which comprises the operating system, applications, software dependencies, and hardware drivers. The RTE provides a compatible environment to couple the data and software components. To build a model or execute an analysis, a reusable, interoperable system must integrate data and software in the appropriate runtime environment.

Data, software, and container (pre-packaged RTE snapshots) repositories are standard and, with varying degrees of success, address the Findable and Accessible aspects of the FAIR principles. However, the complex relationships that allow the Interoperability and Reusability principles are much more challenging to accomplish. Nowhere is this more apparent than with AI data and software. The RTE, which involves building AI models or comparing new data to AI-generated models, is a crucial ingredient often overlooked in the AI ecosystem. This includes considering rapidly evolving processors such as GPUs, TPUs, and FPGAs, which are critical to many AI applications.

While findability and accessibility solutions (some more FAIR than others) have been proposed and are in place in some systems, represented by the vertices of Figure 1. Interoperability and reusability, however, happen along the edges. To achieve interoperability and, ultimately, reusability, there are three requirements: a) software must be paired with data it understands and can use; b) the runtime environment must meet the software’s dependencies; and c) the runtime environment must have access to the data and be executable on hardware available to the user. This presents challenges, especially for AI-dependent research.

The FAIR-RTE working group will coordinate community-led discussions on how to define and effectively apply FAIR principles to runtime environments. To accomplish this, the proposed working group will:

• Provide a community-developed document defining the FAIR principles for Runtime Environments
• Provide guidelines on how to apply the FAIR principles to Runtime Environments
• Provide a document describing the implementation guidelines and adoption examples for FAIR Runtime Environments.

Since the seminal FAIR principles for research data publication were introduced in 2016, considerable effort has been devoted to making datasets and data repositories findable, accessible, interoperable, and reusable. In 2020, the FAIR envelope was broadened to include research software. However, in a practical sense, there is still a large gap between the first half of the acronym, FA (findability and accessibility), and the last half, IR (interoperability and reusability). While data and software repositories are used daily by researchers, providing findability and accessibility to proliferating research data and software, interoperability and reusability depend on a critical additional factor. This element is often referred to as the run-time environment (RTE), which comprises the operating system, applications, software dependencies, and hardware drivers. The RTE provides a compatible environment to couple the data and software components. To build a model or execute an analysis, a reusable, interoperable system must integrate data and software in the appropriate runtime environment.

Data, software, and container (pre-packaged RTE snapshots) repositories are standard and, with varying degrees of success, address the Findable and Accessible aspects of the FAIR principles. However, the complex relationships that allow the Interoperability and Reusability principles are much more challenging to accomplish. Nowhere is this more apparent than with AI data and software. The RTE, which involves building AI models or comparing new data to AI-generated models, is a crucial ingredient often overlooked in the AI ecosystem. This includes considering rapidly evolving processors such as GPUs, TPUs, and FPGAs, which are critical to many AI applications.

While findability and accessibility solutions (some more FAIR than others) have been proposed and are in place in some systems, represented by the vertices of Figure 1. Interoperability and reusability, however, happen along the edges. To achieve interoperability and, ultimately, reusability, there are three requirements: a) software must be paired with data it understands and can use; b) the runtime environment must meet the software’s dependencies; and c) the runtime environment must have access to the data and be executable on hardware available to the user. This presents challenges, especially for AI-dependent research.

The FAIR-RTE working group will coordinate community-led discussions on how to define and effectively apply FAIR principles to runtime environments. To accomplish this, the proposed working group will:

• Provide a community-developed document defining the FAIR principles for Runtime Environments
• Provide guidelines on how to apply the FAIR principles to Runtime Environments
• Provide a document describing the implementation guidelines and adoption examples for FAIR Runtime Environments.