The Illusion of Agency: How AI Disempowers the Human Mind

The defeat of Lee Sedol by Google DeepMind’s AlphaGo in 2016 was a watershed moment for the game of Go, but the true impact of AI on the game was not the immediate loss of professional dominance. Instead, it was the slow, insidious erosion of human agency. While Chess survived the advent of superhuman engines by pivoting toward human drama and e-sports, Go has become a laboratory for a more concerning phenomenon: gradual disempowerment.

When humans delegate their cognition to an AI that is passably good or superhuman, they often believe they are using a tool to enhance their skill. In reality, they may be automating away the very processes that constitute mastery. This transition is rarely a sudden collapse; it is a series of small, comfortable concessions that eventually leave the practitioner unable to function without the machine.

The Sociology of the AI Cheater

The case of Carlo Metta in the European Team Championship illustrates the social dynamics of AI integration. Metta was accused of using AI in online play, a claim supported by the stark contrast between his superhuman online performance and his stagnant over-the-board (OTB) skill. Despite the evidence, Metta was exonerated due to sloppy prosecution and social pressure.

This created a dangerous precedent. When the cost of accusing a cheater becomes socially higher than the cost of the cheating itself, tournament organizers often succumb to a sense of futility. This environment allows AI use to become endemic, not necessarily driven by malice or a desire for prize money, but by "idle curiosity and laziness."

The Trap of Passive Learning

In educational settings, the temptation to "side-eye" an AI solution during a difficult move is pervasive. Students often justify this by claiming they retain artistic control, arguing that the AI is merely a tool to help them express their latent potential. One student's sentiment echoes a common delusion among AI users:

"I never let Leela choose move. I just decide myself which one is better, for this reason i think i can find my own style with Leela."

This perceived agency is an illusion. There is a fundamental difference between verifying a solution and constructing one. Much like a math student who can follow the atomic steps of a proof but cannot grasp the "big picture" motivations, the AI user registers insights without internalizing them. They are in a state of perpetual preparation for an exam they will never pass, because they have never learned to navigate the void of uncertainty without a prompt.

Beyond the Board: Coding and Creative Work

The patterns observed in Go extend far beyond the game board. In software engineering, the rise of AI coding assistants has shifted the labor from junior-level implementation to staff-level architecture. While this increases productivity, it creates a "liminal moment" for the craft. Junior engineers, prompting their way through a codebase without the hard lessons of manual debugging, may be introducing invisible bugs and architectural chaos that only an experienced engineer can later untangle.

This "cognitive offloading" is visible across various intellectual crafts:

• Writing: Delegating grammar and flow to LLMs can atrophy the writer's ability to think clearly through the act of polishing prose.
• Research: Using models to investigate hypotheses without manually inspecting the data leads to "slop" passing through as legitimate research.
• Coding: Relying on agents to fix bugs often results in a cycle where the AI introduces a problem that it then fails to solve, leaving the user stuck because they lack the foundational understanding to intervene.

The Cost of Comfort

Learning is inherently uncomfortable. It requires languishing in confusion and struggling against resistance. AI removes this friction, offering a "highway to amazing performance" via mimicry. However, this efficiency comes at the cost of exploration. When the AI's move is always available, the incentive to experiment with suboptimal but human-driven strategies vanishes.

As we integrate AI into our professional and creative lives, we face a choice. We can use these tools to bootstrap our understanding—by first noting our own judgment and then using AI to assess it—or we can slide into a state of "brain rot," where we become middle managers between our employers and a model.

Ultimately, the tragedy of the AI-assisted player or programmer is not that they are replaced by a machine, but that they volunteer for their own replacement, trading the satisfaction of genuine mastery for the comfort of a simulated success.