Artificial intelligence has defined the technological landscape in the United States for nearly the past two years. As leaders from the public and private sectors alike grapple with the new AI era, data governance and democratizing AI access have emerged as key priorities, especially for agencies like the Central Intelligence Agency.
“I cannot stress enough how important getting governance right is to us,” said CIA AI Director Lakshmi Raman at the Potomac Officers Club’s 5th Annual AI Summit in March. “We are working hard to ensure our adherence to the President’s executive order on AI that was published in October, governing with trust, safety and security in conjunction with the need to innovate.”
Don’t miss your next opportunity to be part of the conversation with government and industry leaders. Join the Potomac Officers Club on April 17 for the 5th Annual CIO Summit! Register here to save your spot.
In alignment with the agency’s AI governance goals, Raman noted that last year, the CIA established an AI governance council. And currently, the agency is working on a national security memorandum that will address the “unique challenges developing and governing AI in the national security space,” Raman shared.
For the AI democratization component, Raman said that effort “actually involves coordination of a number of different efforts.”
“For example, we need to ensure that our folks have access to models, have the ability to run them and incorporate them within their applications. I also should add that we need to do this in a way that optimizes our resources, especially with respect to compute. My office is actively working this issue,” she said.
Raman shared that last year, the CIA undertook an initiative to create a common platform of shared services, and since then, it has continued to develop and evolve cataloging for models and data. Raman called the initiative “a first effort of its kind specifically for the IC.”
“It was built to address the need for searching, discovering and sharing AI/ML models across the IC, and it’s currently running in production,” Raman explained.
And building on that effort, the CIA is also developing a models-as-a-service platform to support a multi-tenant environment for hosting microservices with the goal of centralizing the deployment and maintenance of AI models.
“With the centralized models-as-a-service platform, our data science teams will have a common approach to deploying and hosting AI models as a service, and application development teams will have a centralized location from which to integrate AI services into their applications. This is under active development as we speak,” she said.
