Intelligence agencies have high hopes for artificial intelligence

With the popularization of the Internet, especially e-commerce, online business and social media, more and more economic and social operations have been digitized, and spy satellites, reconnaissance drones, and surveillance systems are also continuously providing usable videos and images . The intelligence agencies of various countries quickly turned from information hunger to information abundance, but at the same time they also began to face new thorny problems, that is, how to find valuable information from a large amount of unstructured voice, graphics and image data. Intelligence agencies are eager to use artificial intelligence to improve data processing capabilities.

When it comes to artificial intelligence, intelligence agencies have been exploring it longer than any other. During the Cold War, the US National Security Agency (NSA) and the UK's Government Communications Headquarters (GCHQ) explored early AI techniques to transcribe and translate the vast numbers of Soviet phone calls they began intercepting in the 1960s and 1970s.

However, the technology was immature at the time. A former European intelligence official said his agency did not use automated transcription or translation in Afghanistan in the 2000s and still relied on native speakers.

Today, agents expect artificial intelligence technology to get even more powerful. The same trends that have made artificial intelligence more attractive to business in recent years—more data, better algorithms, and greater data processing power—have also inspired intelligence agencies.

On February 24, 2022, the British Government Communications Headquarters published a paper on how artificial intelligence may change the way it works. "Machine-assisted fact-checking" can help it identify fake images, check false information against trusted sources, and identify social media bots spreading it.

AI can even stop cyberattacks by "analyzing patterns of activity on networks and devices" and fight organized crime by spotting suspicious chains of financial transactions.

Several other resource-poor organizations have already demonstrated the potential of AI. The US NGO Nuclear Threat Initiative recently found that applying machine learning to publicly available trade data can uncover previously unknown companies and organizations involved in the illicit trade in nuclear weapons material. And intelligence agencies are not limited to publicly available data.

Some hope that, with their ability to snoop on private information, such humble apps could pave the way for AI-driven giants. On March 1, 2022, the US National Security Council Artificial Intelligence Research Committee issued a report saying: "Artificial intelligence will completely change the way intelligence is obtained."

The report is not lacking in ambition. By 2030, some 17 U.S. intelligence agencies will have established a "joint architecture of continuous learning analysis engines" capable of analyzing everything from human intelligence to satellite imagery to anticipate imminent threats, the report said.

The committee applauded the Pentagon's response to the COVID-19 outbreak, which has integrated dozens of sets of data to identify outbreak hotspots and manage supply and demand.

However, what works in public health does not necessarily work in national security. Western intelligence agencies must grapple with laws about how private data is collected and used.

In any case, the practical constraints faced by artificial intelligence in the intelligence field are as much as the moral constraints. Machine learning is good at recognizing patterns (like the unique patterns of cell phone use), but not so good at predicting individual behavior.

This is especially true in situations where data are scarce, such as in counterterrorism. Predictive policing models process data from thousands of burglaries each year. But terrorist attacks are far less practiced, and thus more difficult to learn from.