Interrogating the machine about the law

Instead of arguing about the reform from hearsay, they put an AI voice assistant on speaker and run the longest live consultation in the diary. The transcript doubles as a manual for AI-assisted civic literacy, four moves deep.

Catch the contradiction. The AI says the surcharge rises “poco a poco”, having just said it started at 80%. He pounces: “¿cómo me dice que poco a poco, si empezó en el 80?” ▸ 21:21, and the machine apologizes and restates correctly. A fluent voice is not a verified fact; press it.

Ask for the experiment. “¿Hicieron algún piloto, consiguieron datos reales, o fue una discusión de ‘yo pienso que esto es así’?” ▸ 23:27. Answer: no pilot; studies and other countries’ experience. That single question converts a policy debate into an epistemology check.

Trade anecdotes for averages. When the burger simulation starts with invented numbers, he stops it: “vamos a utilizar cifras más reales, busca el ticket promedio, el personal promedio, las ventas promedio” ▸ 40:34, and the model fetches real figures before recomputing.

Let it kill your own theory. Julia’s family-business thesis, relatives work free, so family shops win, goes into the simulator and dies: the law applies to family workers too, verbal arrangements included; only actual owners are exempt. Verdict, delivered against their own side: “baila la idea de los negocios familiares” ▸ 35:38, survivors are solo carts and the spouse-child triangle, and the realistic prediction is a rise in informality ▸ 37:17.

la IA no es el oráculo; es el laboratorio →

None of this required trusting the machine. It required treating it as a laboratory where claims, the government’s, the AI’s, and their own, get stressed until something breaks. Forty minutes, zero credentials, one phone: civic analysis is now a conversation anyone can hold, provided they hold it like an interrogation…