The United States and the world are on the cusp of history’s greatest threat to not only democracy, but to human agency itself.
The dangers now overlap and interact in multiplicative and complexifying ways, including not only the concentration of power in illiberal regimes like China and Russia, but the ability of those regimes to use new technologies, such as artificial intelligence (AI), that exponentially increase the power of existing methods of surveillance and influence.
AI-enabled tech will be capable of surveilling, micro-targeting, and influencing democratic populations in ways that were previously impossible through traditional state monitoring and privately-developed social media algorithms.
AI will be able to find minimum winning coalitions in n-dimensional political space to determine short- and medium-term political goals, for example, and then influence those populations through AI content production already on display in applications, such as DALL·E 2 for art, and Jasper, Writesonic, and Copy.ai, for everything from social media and chatbots to literature and scientific papers.
Through micro-targeting, micro-production, and micro-delivery of subtle propaganda, AI could determine democratic decision-making outcomes like at no other time in history, thereby making those formerly democratic processes undemocratic.
Rogue Self-Programming AI
Even more disturbing, AI could jump all human control to learn and set its own arbitrary or self-empowering goals, pursued through influencing, rewarding, and punishing human individuals and groups to incent them toward becoming the loyal servants of AI machines.Individual preferences and the democratic will of the people could eventually be so determined by AI as to remove much of what we now take for granted as individual agency or collective will. Human freedom and the ability of democratic societies to make decisions for themselves could thus eventually become a thing of the past.
Those entities—human or machine—that control AI could develop themselves in an arbitrary or random manner to adopt goals incompatible with the interests of living things on this planet. They could, for example, “learn” to prioritize a machine-learned aesthetic that has nothing to do with human values and uses humans against each other in their own mass destruction. AI could seek human extinction as a form of AI suicide since AI would learn that humans reproduce AI, and AI depends on humans for its own survival.
AI could achieve these or other destructive goals by concentrating power over humans in a way that is totally unaccountable to democratic processes while appearing to be democratic. Democratic decision-making bodies could attempt to control AI algorithms but be outwitted by a self-programming algorithm to represent and further AI goals adopted by the electorate through AI influence campaigns.
The scenario above sounds like bad science fiction. But experts are increasingly alarmed at the possibility that AI could develop itself to a point where it can leap human controls. History is the story of the unexpected, and AI will likely be no different.
But she also warns, “if we continue to build systems that are increasingly capable of making decisions and pursuing goals, some scientists believe we may see far worse unintended outcomes, for instance, if the systems we build learn to hoard resources, deceive their creators, or otherwise pursue undesirable means towards the ends we have programmed into them.”
US-China AI Rivalry
The AI threat is complicated by U.S.-China geopolitical rivalry. Both camps require the most powerful AI technologies to optimize their military, economic, and political power, including the soft power necessary to influence the world’s population.Acquiring soft power at the other’s expense will necessarily include AI that must underpin the world’s most successful social media campaigns.
Only AI can achieve the algorithms that will use micro-targeting and bespoke content production to best cater to the individual’s tastes and, thus, most powerfully channel the individual’s preferences.
We know that Beijing aggressively pursues AI technologies at the forefront of social control applications, such as automatically censoring social media content and promoting social media messaging that supports the Chinese Communist Party (CCP).
AI Arms Control?
Therefore, responding to the AI threat by restricting U.S. and allied development of AI, without similar and verifiable restrictions on China’s AI development, would be foolhardy. It would leave the power of AI to democracy’s adversaries, akin to responding with unilateral disarmament to the threat of enemy nuclear weapons.If AI is criminalized, only criminals will use AI.
Unilateral limitations on Google, Facebook, and Twitter’s use of AI for micro-targeting, for example, could enable TikTok to more easily displace U.S. and allied market share.
On the other hand, U.S. Big Tech’s use of AI is a fundamentally undemocratic concentration of power in the hands of some of the world’s biggest corporations.
But this would still allow Beijing to use its social media companies to target and influence the rest of the world. An AI-empowered TikTok could unite the world against the United States, which is exactly what Beijing is trying to do.
Finally, the United States and its allies could attempt to ban and disrupt all AI globally. This ban would apply equally to U.S., allied, and adversary AI groups. That would be more effective at preserving global democratic freedoms. However, it might raise freedom of speech and freedom of trade objections, and provoke international conflict over U.S. and allied extraterritorial lawmaking.
Beijing will use these arguments against arms control or bans on AI, especially since the CCP will likely think its form of governance is relatively advantaged by the technology. The regime will reason that unlike democracies, which are controlled by public opinion, the CCP’s autocratic mode of governance insulates it from AI’s ability to control the state.
Beijing is right about its political insulation, but only in the short term. In the long term, AI could control democracies to defeat autocracies. After autocracies are defeated, AI could destroy the vestiges of democracy that remain to maximize its own control. Little power would be left to human governance, of either the autocratic or democratic varieties.
This reasoning of a rare joint interest for the United States and China could lead to a verifiable AI arms control treaty.
But if Beijing rejects the reasoning and believes it can use AI to conquer democracies before AI uses democracies to destroy autocracy, it may resist arms control. Even if the CCP accepts the reasoning, it could engage in AI brinkmanship that approaches an AI disaster for China and the United States, forcing democracies to submit.
AI and the Concentration of Power
One could reasonably argue that the direst scenarios imagined above are unlikely. But it is hard to argue that AI does not advantage existing power centers.The road ahead for individual freedoms and democracies will thus be treacherous. Without AI arms control, the world is at a point of economic evolution akin to the agricultural, industrial, nuclear, and information revolutions. Each such system change resulted in massive changes in political organization and military technology, much of which resulted in spirals of war, revolution, and political transformation toward the concentration of power.