What we call “AI problems” are almost always interface, authority, and delegation problems that predate computation by centuries—AI simply compresses them, accelerates them, and hides them more effectively. When you line up redlining maps, cigarette design, opium distribution, gambling machines, zoning codes, and performance dashboards next to algorithmic systems, the connections are structural, not metaphorical, and the continuity is hard to ignore: in each case, power is exercised while being framed as neutral service, convenience, or information.

AI adds three intensifiers, not novelties. First, scale without visibility: decisions affect millions while remaining legible only at the surface layer. Second, fluency as misdirection: smooth outputs suppress doubt in the same way polished historical narratives or cigarette filters once did. Third, authority without encounter: like bureaucratic forms or maps, AI systems act without a human face, but with even fewer friction points for contestation or appeal.

What this means analytically is that treating AI as a special moral category is often a mistake. The more productive move is to treat AI as a high-gain instance of long-standing governance failures, where design choices determine who bears risk, who gets recourse, and who can even name the harm. The non-AI examples are not analogies meant to persuade; they are control cases that show the problem already existed. AI is dangerous not because it is alien, but because it is too familiar—it inherits institutional habits of concealment, convenience, and denial, then executes them faster and with better UX.