The next generation of repressive technology will make past efforts to spread propaganda and quash dissent look primitive.
By Richard Fontaine and Kara Frederick, March 15, 2019, Wall Street Journal
Chinese authorities are now using the tools of big data to detect departures from “normal” behavior among Muslims in the country’s Xinjiang region-and then to identify each supposed deviant for further state attention. The Egyptian government plans to relocate from Cairo later this year to a still-unnamed new capital that will have, as the project’s spokesman put it, “cameras and sensors everywhere,” with “a command center to control the entire city.” Moscow already has some 5,000 cameras installed with facial-recognition technology, and it can match faces of interest to the Russian state to photos from passport databases, police files and even VK, the country’s most popular social media platform.
As dystopian and repressive as these efforts sound, just wait. They may soon look like the quaint tactics of yesteryear. A sophisticated new set of technological tools-some of them now maturing, others poised to emerge over the coming decade-seem destined to wind up in the hands of autocrats around the world. They will allow strongmen and police states to bolster their internal grip, undermine basic rights and spread illiberal practices beyond their own borders. China and Russia are poised to take advantage of this new suite of products and capabilities, but they will soon be available for export, so that even second-tier tyrannies will be able to better monitor and mislead their populations.
Many of these advances will give autocrats new ways to spread propaganda, both internally and externally. One key technology is automated microtargeting. Today’s microtargeting relies on personality assessments to tailor content to segments of a population, based on their psychological, demographic or behavioral characteristics. Russia’s Internet Research Agency reportedly conducted this kind of research during the 2016 U.S. presidential race, harvesting data from Facebook to craft specific messages for individual voters based in part on race, ethnicity and identity. The more powerful microtargeting is, the easier it will be for autocracies to influence speech and thought.
Until now, such efforts have been mostly limited to the commercial world and have focused on precision advertising: Facebook itself conducts microtargeting, for instance, and Google labeled users “left-leaning” or “right-leaning” for political advertisers in the 2016 election. But private firms are developing artificial intelligence that can automate this customization for whole populations, and government interest is sure to follow. In an October 2018 discussion at the Council on Foreign Relations, Jason Matheny, the former director of the U.S. government’s Intelligence Advanced Research Projects Activity, cited this kind of “industrialization of propaganda” as one reason to beware of the “exuberance in China and Russia towards AI.”
China is the leader in developing new forms of control and is eager to export them.
AI-driven applications will soon allow authoritarians to analyze patterns in a population’s online activity, identify those most susceptible to a particular message and target them more precisely with propaganda. In a widely viewed TED Talk in 2017, techno-sociologist Zeynep Tufekci described a world where “people in power [use] these algorithms to quietly watch us, to judge us and to nudge us, to predict and identify the troublemakers and the rebels.” The result, she suggests, may be an authoritarianism that transforms our private screens into “persuasion architectures at scale…to manipulate individuals one by one, using their personal, individual weaknesses and vulnerabilities.” This is likely to mean far more effective “influence campaigns,” aimed at either citizens of authoritarian countries or those of democracies abroad.
Emerging technologies will also change the ways that autocrats deliver propaganda. State-controlled online “bots” (automated accounts) already plague social media. During Russia’s 2014 invasion of Crimea and in the months afterward, for example, researchers at New York University found that fully half of the tweets from accounts that focused on Russian politics were bot-generated. The October 2018 murder of Washington Post columnist Jamal Khashoggi prompted a surge in messaging from pro-regime Saudi bots.
But bots will soon be indistinguishable from humans online-capable of denouncing antiregime activists, attacking rivals and amplifying state messaging in alarmingly lifelike ways. Lisa-Marie Neudert, a researcher with Oxford’s Computational Propaganda Project, has warned that “the next generation of bots is preparing for attack. This time around, political bots will leave repetitive, automated tasks behind and instead become intelligent.” The kind of tech advances that fuel Amazon’s Alexa and Apple’s Siri, she told the International Forum for Democratic Studies last October, are also teaching propaganda bots how to talk.
For years, the Chinese government has employed what’s known as the “50 Cent Army”-thousands of fake, paid commenters-to post online messages favorable to Beijing and to distract online critics. In the future, bots will do the work of the current legions of regime-paid desk workers.
These increasingly insidious bots will work together with other new tools to let dictatorships spread disinformation, including “deep fakes”-digital forgeries impossible to distinguish from authentic audio, video or images. Audio fakeries are already getting good enough to fool many listeners: Speech-synthesis systems made by companies such as Lyrebird (which says it creates “the most realistic artificial voices in the world”) require as little as one minute of original voice recording to generate seemingly authentic audio of the target speaker.
Video is soon to follow. On YouTube, one can already see an unnerving mashup of actors Steve Buscemi and Jennifer Lawrence and a far-from-perfect video made by the Chinese company iFlytek showing both Donald Trump and Barack Obama “speaking” in fluent Mandarin. Soon, such fakes will be chillingly convincing. That will leave those playing defense “outgunned,” according to Dartmouth computer science professor Hany Farid. There are probably 100 to 1,000 times “more people developing the technology to manipulate content than there is to detect [it],” he told Pew in January. “Suddenly there’ll be the ability to claim that anything is fake. And how are we going to believe anything?”
The kind of advances in AI that fuel Alexa and Siri are also teaching propaganda ‘bots’ how to talk.
New tools will also make it possible for dictators to conduct surveillance as never before, both online and in the real world. Humans are training computers to identify and interpret emotional context within blocks of text using natural language processing (an application of machine learning). Facebook now uses similar techniques to examine linguistic nuances in posts that might flag users who are contemplating suicide. Smaller companies are working to score individual social-media posts based on attitude, emotion and intent.
The California-based AI startup Predictim scoured the text of Twitter , Facebook and Instagram to develop risk ratings for (of all things) wannabe babysitters. Based solely on the language in potential babysitters’ social-media postings, the app provided automated assessments of their propensity to bully, be disrespectful or use drugs. The startup’s efforts triggered a swift backlash last year, but China, Russia and other autocracies won’t share such scruples. Jack Clark, who directs policy for the research firm OpenAI, warns that “we currently aren’t-at a national or international level-assessing or measuring the rate of progress of AI capabilities and the ease with which given capabilities can be modified for malicious purposes.” This, he adds, “is equivalent to flying blind into a tornado-eventually, something’s going to hit you.”
The next generation of natural language processing tools will become more sophisticated as advances in machine learning accelerate. Applied by the wrong regime, they can be combined with other data to assess an individual’s trustworthiness, patriotism and likelihood of dissenting.
Such applications do not yet exist, but an early move in that direction can be seen in China’s public statements. As The Wall Street Journal has reported, “By 2020, the government hopes to implement a national ‘social credit’ system that would assign every citizen a rating based on how they behave at work, in public venues and in their financial dealings.” Local governments across China are already keeping digital records of citizens’ behavior and docking them for jaywalking, breaking family-planning rules or paying bills late. Those who end up on the blacklist lose out, unable to buy high-speed train tickets, obtain government subsidies, purchase real estate or even get hired. According to a plan issued by Beijing’s municipal government, by 2021, the capital’s blacklisted citizens will be “unable to move even a single step.”
Venezuela has introduced its own “carnet de la patria” (fatherland card), a smart-chip-based piece of identification that citizens need to get access to government services such as health care and subsidized food. Human Rights Watch reports that the card may capture voting history as well. The data that this system generates is stored by the Chinese company ZTE, which has also reportedly deployed a team of experts within Venezuela’s state-run telecommunications company Cantv to help run the program, according to a 2018 investigation by Reuters.
Yoshua Bengio, a computer scientist known as one of the three “godfathers” of deep learning in AI, recently described to Bloomberg his concerns about the growing use of technology for political control. “This is the 1984 Big Brother Scenario,” he said, “I think it’s becoming more and more scary.”
Autocrats’ ability to spy on their citizens will be further enhanced by advances in artificial intelligence that make sense of enormous data sets. In both the U.S. and China, companies are optimizing new chips to support neural networks-an algorithmic approach loosely inspired by human brain function. China’s Ministry of Industry and Information Technology recently said that it hoped to mass-produce neural-network optimized chips by 2020. They will allow oppressive regimes to more efficiently collect information on their population’s speech and behavior, sift through massive data sets and quickly exploit the information.
One particular application of AI-facial recognition-could be as ubiquitous in a decade as smartphone cameras are today. The technology has been used by the U.S. Department of Homeland Security, San Diego’s Police Department and others to enhance security at large events like the Super Bowl. In the hands of autocrats, however, the technology has great potential for repressive use. Chinese police deployed facial-recognition glasses in early 2018, and Beijing-based LLVision Technology Co. sells basic versions to countries in Africa and Europe. Such glasses can be used to help identify criminals like thieves and drug dealers-or to hunt human-rights activists and pro-democracy protesters.
A political dissident in Harare may soon have as much to fear as a heroin smuggler in Zhengzhou: The Chinese AI firm CloudWalk Technology has sold Zimbabwe’s government a mass facial-recognition system. It will send facial data on millions of Zimbabweans back to the company in China, allowing it to refine its algorithms and perfect the system for further export. Business is also booming for other companies. The global client list of the Chinese surveillance firm Tiandy, a CCTV camera manufacturer and “smart security solution provider,” includes more than 60 countries.
The rise of new “smart cities” around the world could also mean trouble. Autocratic regimes will be able to weave diverse data streams into a grid of social control. China plans to build more smart cities like Yinchuan, where commuters can use a positive facial ID to board a bus, or Hangzhou, where facial data can be used to buy a meal at KFC. Planned megacities like Xiong’an New Area, a development southwest of Beijing, suggest the shape of future panopticons. These cities of the future could use centralized systems of control across financial, criminal and government records, drawing on websites, visual imagery, phone applications and sensors-all of it propelled by 5G data transmission.
Until quite recently, it was easy to see the digital revolution as a great liberalizer, a way to transmit ideas faster than any would-be censor could react. The reality is turning out to be far more complicated.
The internet dispersed data, but new technological advances can concentrate its power in the hands of a few. With more than 30 billion devices expected to be connected to the internet by 2020, each one generating new data, those who can control, process and exploit the information rush will have a major advantage. A regime bent on stability may feel virtually compelled to do so.
But we shouldn’t assume that the benefits will accrue only to repressive governments. When dictatorships sought in recent years to monitor their citizens’ online communications, the U.S. State Department and others sponsored encryption tools that allowed would-be dissenters to safely communicate. When regimes censored information and blocked access to key websites, circumvention tools cropped up to allow unfettered access.
That is the right idea. Open societies will need to marshal an array of responses in the contest ahead. Democracies will need to slap sanctions on the individuals and groups using new tools for repressive ends, inflict higher costs on technology companies complicit in gross human-rights abuses, invest in countermeasures and harden their own systems against external intrusions. Free governments will also have to differentiate between using new technologies for legitimate purposes (such as traditional law enforcement) and using them to solidify single-party control, curtail basic rights and meddle in democracies abroad.
Dictators from Caracas to Pyongyang will seek to exploit the enormous potential for political misuse inherent in the emerging technologies, just as they have over the decades with radio, television and the internet itself. Democracies will need to be ready to fight back.
Mr. Fontaine is the CEO of the Center for a New American Security in Washington, D.C. Ms. Frederick is an associate fellow in the center’s technology and security program and worked previously for Facebook, the U.S. Naval Special Warfare Command and the Department of Defense.