Predictions about AI focus on the technology, sometimes on its economic effects — and stop there. The blind spot is the same one that swallowed the internet and social media: democracy and politics. AI never had the early optimism those technologies enjoyed; it is already unpopular, and that sentiment is widespread, politically unrepresented, and growing. It has every marker of an issue a populist entrepreneur could ride — relatable, opinion-bearing, and met by a political system that looks unresponsive. When that breakthrough comes, as immigration did in 2015, party systems can reshape within a single electoral cycle.

There's a lot of information that needs to pass between today's AI governance literature and the disciplines that have long dealt with takeovers and how to treat them — constitutional law, history, social science, political science. I look at AI takeovers as takeovers, and ask what governance structures the analogy suggests: the designs devised against human-led takeovers. One such design is adversarial institutional design, but many others are possible. The point is that these disciplines need to work together and build a unified framework — if we are to take this risk seriously, both intellectually and in practice.

Ten years ago I published a book arguing that science fiction is not escapism but a laboratory for political thought. That book became four books, a university course, a conference franchise, and a community of 30+ authors. Now I am building the English-language version: a Substack, an international outreach programme, and a growing body of work applying political science to fictional universes — and fictional imagination to political science.