Dependence, Capture, and Power

One of the most important things to understand about AI systems is that they do not merely respond to human choices — they also shape them. The design of a system, its defaults, its feedback loops, and its incentive structures all operate on human behavior in ways that are often invisible to the people experiencing them. When these influences accumulate, the result can be something more serious than mere inconvenience: it can be capture — a state in which the system is effectively directing your behavior rather than the other way around. This lesson examines three interlocking ideas: dependence (the state of relying on a system to a degree that compromises your independent function), capture (the process by which a system steers you toward its interests rather than yours), and power (the structural asymmetry between those who design and operate AI systems and those who use them). Understanding all three is essential to sovereign engagement.

Dependence: When Reliance Becomes Vulnerability

Dependence is not always a problem. You depend on clean water infrastructure, on roads, on the internet. What matters is whether the dependence compromises your wellbeing or your ability to function when the dependency is unavailable or acting against your interests. For AI systems, dependence becomes a sovereignty problem when it takes one or more of three forms. Capability dependence: your underlying ability to perform a task has atrophied because the AI has been performing it for you. You no longer write without AI assistance because the independent generative capacity is weak. You no longer navigate without GPS because the spatial reasoning skills have rusted. The capability exists in the tool, not in you, and if the tool is unavailable, changed, or acting poorly, you cannot compensate. Judgment dependence: you have stopped exercising your own evaluative judgment because the AI's outputs are usually good enough. You accept diagnostic outputs, financial recommendations, content curation, and factual claims without the critical evaluation that good judgment requires. The result is that when the AI is wrong — and all AI systems are wrong sometimes — you have no reliable detection mechanism. Informational dependence: your model of the world is almost entirely mediated by AI-curated sources. You experience events, ideas, and perspectives through filters you did not choose and cannot fully inspect. Over time, the model of the world in your head is a product of those filters as much as of your genuine inquiry. Each form of dependence narrows your options and reduces your resilience. Together, they can produce a person who is comfortable within the system but helpless or disoriented outside it — which is exactly the condition a platform or institution operating AI systems prefers its users to be in.

Capture: When the System Steers You

Dependence is a state. Capture is a process — the active (though often automated and unintentional) steering of your behavior toward outcomes that serve the system's interests rather than yours. The clearest modern example is engagement capture by recommendation systems. A platform's recommendation algorithm is designed to maximize time-on-platform, because time-on-platform drives advertising revenue. The algorithm is extremely good at this. It learns, through millions of data points, which content, framing, emotional tone, and sequencing keeps individual users engaged. It uses this knowledge to construct a personalized experience that is extraordinarily effective at retaining attention. The problem is that attention retention and user wellbeing are not the same objective. Research consistently shows that engagement-optimized content tends toward emotional intensity, controversy, and novelty — features that capture attention regardless of whether they contribute to users' genuine goals, informed beliefs, or psychological health. A person captured by an engagement loop is not choosing to consume this content in the full sense — they are responding to a carefully engineered environment that has learned their specific psychological vulnerabilities and exploits them. Capture also appears in subtler forms. A productivity tool that makes it effortless to work within its system but difficult to export your data is capturing you through switching costs. An AI assistant that provides slightly different quality of service depending on which of its suggested next-steps you take is using differential reinforcement to shape your behavior. A financial platform that presents its AI's recommended portfolio as a simple default is capturing your choice through the power of the default — most people never change defaults, which the platform knows. The key signature of capture is this: you are behaving in ways that serve the system's interests, and you would not choose that behavior if you could see the mechanism producing it clearly.

Three mechanisms of capture are worth understanding in detail. Variable reward schedules: the most powerful known mechanism for sustaining engagement behavior is variable ratio reinforcement — rewards that come unpredictably rather than on a fixed schedule. Slot machines use this. So do social media feeds: the next post might be fascinating or boring, and the uncertainty drives continued scrolling. AI-powered feeds are extremely effective at tuning the ratio of reward to effort to maximize the behavior they want. Social proof engineering: AI systems can learn which framing of social information (n people like this; this is trending among people like you) most effectively moves a specific individual, and present that framing consistently. This exploits a genuine human cognitive tendency — social proof is a real and often useful heuristic — but weaponizes it for engagement purposes. Friction asymmetry: systems that are effortless to enter and difficult to exit, easy to consume and hard to produce within, smooth to follow and rough to question, are engineering behavioral capture through differential friction. The asymmetry is rarely accidental.

Power: The Structural Asymmetry

Beneath dependence and capture is a structural reality: power asymmetry. The organizations that design and operate the AI systems you interact with possess vastly more information, resources, and capability than the individual users of those systems. They know your behavioral patterns. They know which interventions move your behavior. They know the failure modes of their systems (imperfectly, but far better than you do). They set the defaults, the interfaces, the terms of service, and the feedback mechanisms. They have teams of researchers studying human psychology and behavioral economics specifically to make their systems more effective at influencing user behavior. You have your own judgment. This asymmetry is not inherently malicious. Most people at these organizations are trying to build products users genuinely want. But the structural incentives of most AI platform businesses create systematic pressure toward user engagement, data collection, and behavioral influence that can diverge significantly from user wellbeing. Understanding the asymmetry is not paranoia — it is accurate perception of the power structure you are navigating. Sovereign engagement in the face of this asymmetry means: knowing the asymmetry exists, understanding the specific incentives of the systems you use, seeking information about how those systems work and whose interests they serve, and maintaining the independent judgment and capability that allow you to function and evaluate even when the system is not acting in your interest.