NASA Use of Generative Artificial Intelligence Technologies.pdf

From: Escalante, Jennifer E. (JSC-BL111) To: Escalante, Jennifer E. (JSC-BL111) Subject: FW: [Contains CUI]: NASA’s Use of Generative Artificial Intelligence (AI) Technologies Date: Thursday, June 29, 2023 4:16:25 PM

From: Agency-IT-Outreach <agency-it-outreach@mail.nasa.gov> Sent: Monday, May 22, 2023 12:14 PM Subject: NASA’s Use of Generative Artificial Intelligence (AI) Technologies

MESSAGE TO All NASA EMPLOYEES FROM THE NASA CHIEF INFORMATION

OFFICER

Point of Contact: Eldora Valentine, Office of the Chief Information Officer, 202-358- 3968, eldora.valentine-1@nasa.gov NASA’s Use of Generative Artificial Intelligence (AI) Technologies and Large Language Models such as OpenAI, ChatGPT, Google Bard, Facebook Llama, or Similar Generative AI – A Door Has Opened With the release of more accessible Generative AI systems like ChatGPT, I wanted to share some thoughts on the use of these technologies in support of the NASA mission. Large language model Generative AI capabilities are gaining popularity worldwide due to their ability to rapidly generate high-quality images, audio, text, code, simulations, and videos. Generative AI can assist humans in:

Rapidly finding information and summarizing knowledge Extracting similar data from disparate formats and multiple sources Writing, documenting, debugging, and translating computer code Generating art Inventing new solutions

While Generative AI technologies have impressive potential and ultimately may be used to great benefit in support of NASA missions, they also today pose significant risks including:

Many of today's Generative AI tools are cloud-based services that may be hosted in non-US locations, potentially exposing data submitted to the system to non-US individuals or those unauthorized to access sensitive NASA data.

NASA sensitive data provided to Generative AI systems may become publicly available information, a risk that is not acceptable for the Agency.

Generative AI results have mixed quality and require human judgement;

some results can appear to be factual and carry a high-level of confidence, while being wholly or partially incorrect.

Today it is unclear how the providers of AI technologies and systems safeguard or may use it for their own purposes in the future.

As with any other system, generative AI systems may have risk of data leaks or hacker penetrations of NASA systems.

Beyond security considerations, use of generative AI systems must be conducted while including ethical and intellectual property considerations;

guidance and care is required in these areas.

Can We Use Generative AI at NASA Today?

The Office of the Chief Information Officer (OCIO) recognizes NASA has a creative, curious, and scientific workforce and many of us are interested in these popular AI technologies. We have conducted an initial level of due diligence through collaboration with other NASA organizations as well as benchmarking with external organizations to identify the maturity of and risks associated with these evolving tools. In view of the potential benefits and risks associated with Generative AI, can/should NASA employees use these technologies today?

Currently, NASA has not authorized the operation of generative AI technologies (e.g., OpenAI, ChatGPT, Google Bard, Facebook Llama, or similar) for widespread use on sensitive NASA data.

NASA follows existing federal requirements and processes to ensure NASA data is appropriately protected. At this time, these evaluations have not been conducted at NASA and/or across the federal government for Generative AI systems.

This means that today, individuals at NASA are not authorized to share, upload, or otherwise expose sensitive Agency data to ChatGPT or any other generative AI capability. Until the Agency formally approves their use, individuals at NASA are not authorized to use them in any way that may implicate NASA data, devices, and/or networks.

Employees may not use NASA emails to register for personal GenAI access.

NASA employees may not download any client-based Generative AI tools, until those tools are authorized.

NASA employees may access generative AI personal accounts via browser-based interfaces, following acceptable use policies, and only use public, non-sensitive data.

NASA employees may also consider careful experimentation with generative AI capabilities using with personal devices provided the do not use NASA non-public or sensitive data with these systems.

If NASA employees use the outputs of Generative AI systems for any purpose, they should thoroughly check and validate the results because quality of outputs for these systems is still uncertain.

What is NASA Doing to Enable the Broader Use of Generative AI?

Today, OCIO is working with a community of potential early adopters across NASA to investigate the use of certain AI technologies, via an authorized cloud environment.

This effort will be governed as a Technology Demonstration activity with scope restricted to non-sensitive data. Participation in these early efforts will be limited in accordance with the cloud environment’s authorization to operate and relevant security controls and will assess quality of results; ease of use; formulation of prompts; security compliance; costs; and supportability, as well as understanding developer-level application programming interface access. In addition, OCIO is coordinating closely with leading industry partners, fellow government organizations including the Federal CIO, Chief Data Officer, and Chief Information Security Officer Councils to understand the significant amount of policy guidance emerging around Generative AI as well as how other organizations are adopting generative AI capabilities. OCIO is also connecting with commercial providers to understand how Generative AI will be integrated into widely available tools, such as the Microsoft 365 suite, visual tools like Adobe Illustrator, and the often-used Google search.

How Can You Find Out More Information or Get Involved?

Contact the OCIO AI Machine Learning Lead, Ed McLarney to contribute to strategies and plans for accelerating scaled adoption of Generative AI at

NASA.

Developer / expert users may contact Martin Garcia for instructions in submitting use cases for early experimentation.

Find out about OCIO Data Science Resources: Data Science Services (sharepoint.com) Find out about OCIO Software Resources: Software Resources (sharepoint.com) Learn about NASA approved/unapproved software:

https://nasa.sharepoint.com/sites/cio/SitePages/Unauthorized-Software-FAQs.aspx Find OCIO Policies: IT Policies & Governance (sharepoint.com)

Jeff Seaton NASA Chief Information Officer