LOFTER-网易轻博

200831 TheComing Revolution in Intelligence Affairs

How Artificial Intelligence andAutonomous Systems Will Transform Espionage

By AnthonyVinci   

August 31, 2020  《F.A.外交事务》

For all ofhuman history, people have spied on one another. To find out what others aredoing or planning to do, people have surveilled, monitored, andeavesdropped-using tools that constantly improved but never displaced theirhuman masters. Artificial intelligence (AI) and autonomous systems are changingall of that. In the future, machines will spy on machines in order to know whatother machines are doing or are planning to do. Intelligence work will stillconsist of stealing and protecting secrets, but how those secrets arecollected, analyzed, and disseminated will be fundamentally different.

Militaryfuturists have recognized a similar sea change, and some have dubbed the riseof AI and autonomous weapons systems a “revolution in military affairs.” Itsanalog in intelligence can be understood as a “revolution in intelligenceaffairs.” Through the coming RIA, machines will become more than just tools forinformation collection and analysis. They will become intelligence consumers,decision-makers, and even targets of other machine intelligence operations. Theultimate concern of these machines will still be the political, social,economic, and military relations of human beings—but machine-drivenintelligence will operate at such speed, scale, and complexity thathuman-driven intelligence will no longer be able to keep up. There is nostopping the RIA. The forces of technological innovation and competition havealready unleashed it on the world. Instead, the U.S. intelligence communitymust embrace the RIA and prepare for a future dominated by AI—or else risklosing its competitive edge.

THE RISE OF MACHINE-DRIVEN INTELLIGENCE

Revolutionsdon’t come out of nowhere. The RIA’s origins date back to the twentiethcentury, when new technologies such as telecommunications and computingbrought intelligence tradecraft new sophistication. Humans remained the agentsof intelligence, but instead of watching with their own eyes, overhearing withtheir own ears, and analyzing and predicting with their own minds, they drew onincreasingly powerful sensors and computing tools to enhance theircapabilities.

Over the last20 years this trend accelerated, leading to an enormous increase in the amountof data available to intelligence agencies. Classified and commercial sensors,ranging from bots on networks to autonomous drones to small satellites inspace, now produce more information than humans could ever comprehend on theirown. In 2017, the National Geospatial-Intelligence Agency predicted that thedata its analysts would have to parse would increase a million-fold within fiveyears. So much data so fast has ginned up fierce competition to make sense ofit all—a dynamic that in turn has driven the adoption of automation, big dataanalytics, and AI. The pressure to keep up is tremendous: nations whoseintelligence agencies can swiftly process vast quantities of complex data willhave an edge over those that can’t.

Revolutionsdon’t come out of nowhere.

The parallelshift to autonomous systems and AI among the world’s militaries has only intensifiedthe competitive pressure: intelligence agencies need to be able to target andsupport advanced warfighting systems. Already, the U.S. military operates morethan 11,000 unmanned aerial systems and even more underwater, space, andterrestrial systems. In addition, U.S. cybersecurity units must dealwith millions of bots on global networks as well as billions of Internetof Things devices that act as sensors. These ever-proliferating systems requiretheir own intelligence to operate, meaning that over time they will become theprimary intelligence customers.

Even morerevolutionary will be the shift to autonomous systems asintelligence targets—that is, when machines begin to spy on and deceiveother machines. A plausible future scenario might involve an AI system that ischarged with analyzing a particular question, such as whether an adversary ispreparing for war. A second system, operated by the adversary, mightpurposefully inject data into the first system in order to impair its analysis.The first system might even become aware of the ruse and account for thefraudulent data, while acting as if it did not—thereby deceiving the deceiver.This sort of spy-versus-spy deception has always been part of intelligence, butit will soon take place among fully autonomous systems. In such a closedinformational loop, intelligence and counterintelligence can happen withouthuman intervention.

To understandwhat is at stake, consider an analogy from the financial world. High-speedquantitative trading systems rely on algorithms that sense changes in global stock markets, analyze hugesums of data to make predictions, and then automatically execute trades in amatter of microseconds. Humans can’t operate at anywhere close to the samespeed and scale. In order to keep up with the competition, even the moststalwart investment firms increasingly rely on quantitative trading systems. Tocompete with one another for secrets, intelligence agencies will increasinglyrequire AI and autonomous systems, too.

ADAPTING THE INTELLIGENCE COMMUNITY

As machinesbecome the primary collectors, analysts, consumers, and targets ofintelligence, the entire U.S. intelligence community will need to evolve. Thisevolution must start with enormous investments in AI and autonomizationtechnology as well as changes to concepts of operations that enable agencies toboth process huge volumes of data and channel the resulting intelligencedirectly to autonomous machines. As practically everything becomes connectedvia networks that produce some form of electromagnetic signature or data,signals intelligence in particular will need to be a locus of AI evolution. So willgeospatial intelligence. As satellites and other sensors proliferate,everything on earth will soon be visible at all times from above, a state thatthe federal research and development center Aerospace has called the “GEOINTSingularity.” To keep up with all this data, geospatial intelligence, likesignals intelligence, will need to radically enhance its AI capabilities.

The U.S.intelligence community is currently split up into different functions thatcollect and analyze discrete types of intelligence, such as signals orgeospatial intelligence. The RIA may force the intelligence community toreassess whether these divisions still make sense. Electromagnetic informationis electromagnetic information, whether it comes from a satellite or anInternet of Things device. The distinction in origin matters little if no humanever looks at the raw data, and an AI system can recognize patterns in all ofthe data at once. The division between civilian and military intelligence willbe similarly eroded, since civilian infrastructure, such as telecommunicationssystems, will be just as valuable to military objectives as militarycommunications systems. Given these realities, separating intelligencefunctions may impede rather than aid intelligence operations.

The RIA mayrequire the creation of new organizations as well. If the individual human wasonce the basic building block of intelligence, that distinction now belongs tothe device—a piece of software, say, or a sensor, or an autonomous drone.Intelligence operations will increasingly hone in on these devices, which willmean spying not only on the devices themselves but also on the designers,developers, and supply chains that produce them. In the near future,understanding AI and autonomous systems technology, supply chains, and venturecapital will be just as important as understanding Islamic fundamentalistideology was in the past. The United States may need new organizations to studythese areas. At a minimum, it will need to expand existing economic andtechnological intelligence directorates, just as it expanded counterterrorism units after 9/11.

If theindividual human was once the basic building block of intelligence, thatdistinction now belongs to the device.

At the sametime as the U.S. intelligence community evolves to embrace the RIA, it willneed to limit the ability of its adversaries to do the same—in particular byslowing and stopping their ability to master machine-driven intelligence.Impeding the design and development of adversaries’ AI and autonomous systemswill become increasingly important. This will require covert action—some of itundertaken by machines. For example, the United States can inject false datainto an adversary’s machine learning system, in order to deceive or deny theadversary’s larger AI system.

But just as theUnited States will target its adversaries’ AI and autonomous systems,adversarial intelligence services will target U.S. systems. As a result, theUnited States will need to erect new defenses and embrace new forms ofcounterintelligence. To compete, counterintelligence officers will need thesame eye for deceit that they have always relied on, but they will also needmore economic and technical expertise than ever before. All in all, the RIAwill force change at every level of intelligence, including organizations,training, technology, concepts of operation, and counterintelligence.

THE REVOLUTION BEGINS WITH PEOPLE

Intelligencegathering and analysis may cease to be an exclusively, or even primarily, humanendeavor, but its ultimate objective will still be to understand human-ledgovernments, societies, and militaries. Moreover, humans will bring creativity,empathy, comprehension, and strategic thinking to intelligence that machinesare unlikely to match anytime soon. As a result, division chiefs, caseofficers, and analysts will still have important roles to play long into thefuture—though the nature of their jobs may change.

Nonetheless,the RIA is coming, and the intelligence community will have to adapt andembrace it. Resistance to change has caused disasters in the past—for instance,when the U.S. Navy refused toreplace battleships with aircraft carriers before World War II. The navy wasled by sailors who didn’t grasp the awesome advances in airpower that wouldallow Japan to launch the devastating attack on Pearl Harbor. Similarly, theintelligence community is primarily run by human spies who may not always see(or accept) the inevitability of machine-driven intelligence. Intelligenceagencies must break down cultural barriers, invest in technology, and dedicateentire offices to AI and automation-driven intelligence. If the United Statesrefuses to evolve, it risks giving China or some other adversary atechnological edge that Washington won’t be able to overcome.