How the CIA Uses AI

How the CIA Uses AI

How the CIA Uses AI is a topic of conversation that should interest US citizens given the agency's history of spying, data mining, and strategic use of technology to drive public opinion.

The CIA has been using AI for a variety of reasons as far back as 2012 according to the agency's AI director, Lakshmi Raman. These uses included AI for translation, transcription, and data analysis, which in situations of delicate time constraint, needed to be instant.

In a recent interview with ex-CIA member Andrew Bustamante, he claimed that the CIA was using AI since before his entry into the agency in 2007 that they acquired from classified contracts.

In the AI arms race, most people would prefer that the United States had superior artificial intelligence than lets say China. However, the prospect of the Central Intelligence Agency being equipped with the strongest large language models and using them for intelligence collection on American citizens, keeps some people up at night.

Generative AI in the CIA

The CIA's artificial intelligence program has leveled up substantially in recent years with the adoption by the intelligence community of generative AI systems into their national security strategies.

The CIA is no different than any other institution when it comes to the appeals of using generative AI. They need help with ideation, writing, research, and analysis just like every other organization. However, the nature of the agency's classified data does force them to have a unique relationship with AI technology.

When using AI solutions with classified data, the agency actually chooses to build its own systems instead of compromising their data. For the clandestine purposes of the CIA, the agency creates its own AI-eco system separate from mainstream AI tools to ensure no cross pollination of data.

For instance, right now thousands of analysts across 18 intelligence agencies are using the CIA-developed generative AI system, Osiris. Although the system is open source, it serves as an example of how intelligence agencies keep their AI in-house.

This means the data the AI systems are trained on and have input into them always remains under the agency's oversight.

As of now, one use case of generative AI by the CIA is to sift through news stories from around the world and then classify events like natural disasters, protests, and cyberattacks based on their importance.

Given many of these news stories are in other languages, generative AI and natural language processing allows the agency to ask the AI questions to understand the events more easily.

The Benefits of Using AI in the Intelligence Community

Before the CIA was using generative AI, one Silicon Valley company was contracted to analyze data Chinese fentanyl trafficking and with the help of AI was able to discover that twice as many companies and 400% more people engaged in criminal activity than the results found by the human analysis.

This is just one example of the benefits AI poses to the intelligence community. That same company, Rhombus Power, was able to predict the Russian invasion of Ukraine four months prior to their attack with 80% certainty.

The predictive capabilities of AI may just be the one aspect of the technology that makes it indispensable for military conflict. In war, AI poses the potential to log military conflicts that can lift the fog of war and mitigate civilian casualties.

Can AI Replace The CIA?

The CIA's advantage over AI job displacement is that being in the CIA requires such a unique skillset that no machine can completely replace their staff. The agency does intend to find human machine teaming solutions to optimize workflows though, placing them in the same boat as many industries today.

The CIA knows the value of human intelligence, especially in the world of geo-politics and spying, where creativity is essential.

Conclusion

The secretive nature of the CIA stops us from gaining any real insight into their AI practices. One can imagine though, the lack of transparency the agency offers into their policies could undermine agreed upon philosophical standpoints in the AI community.

For a breakdown of the CIA's AI policy, look no further than this Fedscoop article featuring quotes from the CIA's AI director. One should take this article with a grain of salt though, because the extroverted aims of the CIA's AI program can't be the same as their internal aims and policies.

The article tells us the CIA uses AI for data analysis of global open-source news stories, but the idea that that's the extent of their AI usage is laughable. For all we know, if the agency was employing AI in 2012, then by now, they might have access to advanced systems we can only imagine.