slug: artificial-intelligence-control name: Artificial Intelligence & Control category: Technology & Society type: concept

Artificial Intelligence & Control

Definition & Origins

Artificial intelligence and control is George Soros's late-career thesis that AI — particularly machine learning combined with mass data collection — produces unprecedented instruments of social control, which threaten open societies from within (through manipulation and surveillance capitalism) and arm closed societies with tools of repression no previous dictatorship possessed.

Soros first put the theme on the global agenda at the World Economic Forum in January 2018 (Remarks at Davos 2018). There he warned that the giant IT platforms had become monopolies that "influence how people think and behave without them even being aware of it," with "far-reaching adverse consequences on the functioning of democracy," and that an alliance between authoritarian states and these data-rich platforms could yield "a web of totalitarian control the likes of which not even Aldous Huxley or George Orwell could have imagined."

He sharpened the argument in Davos 2019, calling China "the most dangerous opponent of open societies" precisely because of its lead in applying AI to surveillance; developed it in his 2022 Hoover Institution remarks, where he analyzed Xi Jinping's data-collection regime; and elevated it to first place among global threats in Can Democracy Survive the Polycrisis? (2023): "The main source of the polycrisis afflicting the world today is artificial intelligence."

Core Ideas

AI is not neutral infrastructure; it is an instrument of control. Soros's central claim is that machine learning is exceptionally good at one thing: finding patterns in human behavior that allow prediction and manipulation. Whoever owns the models and the data gains leverage over the people the data describes. That leverage can be commercial (advertising, attention engineering) or political (propaganda, social credit, pre-emptive repression).

The platform business model is itself corrosive. Before AI's current wave, Soros analyzed the attention economy as a control technology: social media companies "deceive their users by manipulating their attention and directing it towards their own commercial purposes. They deliberately engineer addiction to the services they provide" (Davos 2018). The advertising model rewards engagement, engagement rewards outrage, and outrage corrodes the shared reality a democracy needs.

Closed societies gain more from AI than open ones. Repressive regimes face no privacy constraints, no independent courts, and no free press; they can centralize all data in one hand. Xi's China, Soros argues, "collects personal data for the surveillance of his citizens more aggressively than any other ruler in history" (Hoover 2022). AI therefore shifts the open/closed balance of power — the theme that connects this concept to China & authoritarianism.

Yet total control may be self-defeating. Applying his own theory of fallibility, Soros argues that AI-based control carries a hidden flaw: a system that surveils everything but cannot tolerate dissent destroys the feedback it needs to correct errors. In the Hoover remarks he contends that Xi's personalized rule "is bound to be" less successful than Xi believes, because closed systems cannot process the truth about themselves.

The economics reinforce the politics. Soros's market analysis explains why the control problem concentrates: internet platforms are networks with rising marginal returns, so they tend naturally toward monopoly; their "exceptional profitability... is largely a function of their avoiding responsibility for and avoiding paying for the content on their platforms" (Davos 2018). Monopoly profits fund the data accumulation, the data feeds the models, and the models deepen the monopoly — a reflexive loop with political consequences, because the same few companies end up owning the infrastructure of public thought.

Democracy's information immune system is the casualty. In the polycrisis essay (2023), written after ChatGPT's release, Soros frames generative AI as an epistemic threat: when synthetic text, image, and voice become indistinguishable from human production, the error-correction machinery of the open society — journalism, science, courts — loses its grip on shared facts.

Practical Application

Regulation of platforms as public utilities. From Davos 2018 onward Soros has argued that near-monopoly distributors of information are effectively public utilities and "should be subjected to more stringent regulations, aimed at preserving competition, innovation, and fair and open universal access," including responsibility for the content they profit from.

The generative turn. In the 2023 polycrisis essay Soros extends the framework to large language models: ChatGPT's release, he writes, "shocked the world," triggering an arms race among platforms that will deploy synthetic content at scale before anyone has worked out its social effects. His worry is not the technology's existence but its reflexive use: AI-generated persuasion tuned by engagement data is the attention economy's business model perfected — manipulation at industrial scale, priced in milliseconds.

Election integrity. His foundations and advocacy treat information manipulation as a front-line threat to elections — the reason OSF funds disinformation research, independent media, and platform accountability (see political philanthropy).

The China frame. The AI-control thesis structures his China advocacy: oppose Western investment that funds the surveillance state (Investors in Xi's China Face a Rude Awakening, BlackRock's China Blunder), press Europe to recognize the systemic rivalry (Europe Must Recognize China for What It Is), and contest technology transfers (Huawei: Will Trump Sell Out the US on Huawei?).

A test for philanthropies and universities. OSUN, his university network, is partly designed as institutional defense: preserving spaces where independent thought survives both state surveillance and platform monopolies (Davos 2020).

The regulatory lag. Soros observes that democratic institutions move at legislative speed while AI moves at exponential speed. This temporal mismatch means that by the time parliaments understand a technology, its effects are already entrenched — a structural disadvantage for open societies that his philanthropy attempts to counter by funding rapid-response research and advocacy networks. The result is a permanent governance deficit: each new AI capability arrives before its predecessor has been understood, let alone regulated, creating a widening gap between technological power and democratic accountability.

Common Misconceptions

"Soros is anti-technology." He credits the IT platforms with an "innovative and liberating role" (Davos 2018) and does not call for halting AI research. His claim is narrower: the business models and political uses now dominant convert a liberating technology into a control technology, and institutions must be rebuilt accordingly.

"This is standard tech criticism." The distinctive Soros element is the open/closed framework: AI matters most not as a jobs or safety question but as a shift in the balance between open and closed societies — and, reflexively, as a system whose own fallibility (models trained on biased data, regimes blind to their own errors) will produce crises.

"He exaggerates China's capabilities." His argument does not require Chinese AI superiority; it requires only that China faces fewer constraints in applying AI to control. He in fact argues China's system has deep weaknesses — the point is that its surveillance ambitions are unmatched, not that they will necessarily succeed.

"The threat is hypothetical." Soros points to operating systems — Xinjiang's surveillance apparatus, social-credit pilots, platform-driven election manipulation since 2016, and generative AI's demonstrated ability to fabricate reality — as evidence that the instruments are already in use.

Soros's Own Words

Thought Evolution

Key Writings & Related Concepts