AI and democracy: the right to resist optimization

Taiwan’s cyber-ambassador Audrey Tang reflects on the real threat posed by AI, not its ability to mimic human behavior, but its capacity to transform how humans interact with technology. Tang argues that the danger lies in humans adapting to machines, rather than machines mimicking humans. As digital systems have learned to sort attention on a massive scale, people have been trained to behave like components within ranking systems, always visible, reactive, and measurable.

This transformation began long before AI could write essays, compose songs, or simulate conversations, as individuals had already been conditioned to function within these systems.

Tang believes the crucial question of the century is not whether machines will become more human, but whether humans will retain the ability to remain more than what machines can easily evaluate. She has witnessed both possibilities: AI assisting in conversations, making them possible, and then fading into the background; and AI systems that do the opposite, training individuals to live at the pace of feeds, compressing themselves for platform legibility, and mistaking constant reaction for genuine participation.

Tang emphasizes that the issue is not AI talking like humans, but institutions rewarding humans for talking like machines.

A humane technological future, Tang suggests, will depend on designing tools, institutions, and norms that protect distinctly human qualities in public life, such as the capacity to interpret others across differences, make commitments over time, revise oneself without humiliation, hold power accountable, and care for consequences that no benchmark can fully capture.

Tang criticizes the instinct to defend human dignity by preserving every manual task, arguing that humanism cannot be reduced to the preservation of manual effort. She believes that many human tasks are not inherently sacred but serve as scaffolding for more important aspects.

The crucial distinction lies between expression and obligation. While a model can help find words, it cannot stand behind the meaning those words convey. Similarly, a system can assist in preparing for a conversation but cannot assume responsibility for the choices made during that exchange. Tang emphasizes that tools, like glasses, subtitles, maps, hearing aids, translation tools, and search engines, extend human capacities without diminishing humanity.

The same is true for AI systems; their best uses are those that enhance understanding without replacing human agency.

Tang highlights the importance of distinguishing between tools that extend participation and those that replace it. A proxy that books a room or organizes a calendar may be useful, but a proxy that gradually becomes one's social presence or substitute conscience represents a more significant threat. As AI systems begin to shape the standards by which participation is evaluated, humans may adapt to the proxy, learning to write in machine-friendly ways, structure work for model visibility, and optimize emotional expression for algorithmic uptake.

This dehumanization represents the true danger of AI, as people start arranging themselves around what the system can process.