AI and programming classes in Guildford

The learner writes each of Carroll's ten premises as an if-then arrow: cat leads to kills mice, kills mice leads to carnivore, and so on, with "kangaroo leads to not suitable as a pet". The arrows form a graph, and a conclusion is simply a path through it. Python's breadth-first search looks for a path from "kangaroo" to "avoided by me". At first it finds nothing, because most of Carroll's arrows point the wrong way for this question.

The missing rule is the contrapositive: if cats kill mice, then anything that does not kill mice is not a cat. Adding each premise's contrapositive gives 20 arrows, and now the search walks from kangaroo, through not a pet, not a moon-gazer, not a night prowler, not a carnivore, not a mouse-killer, not a cat, not in this house, and not one that takes to me, to detested and finally avoided: ten steps, exactly Carroll's printed answer. The same graph also proves, correctly, that every cat in the puzzle is suitable for a pet.

The tempting slip is to add the converse instead, turning "cats kill mice" into "mouse-killers are cats". That makes the graph richer and the program more confident, but wrong: it then "proves" that every pet is a cat, which the premises never say. The learner writes a test with Carroll's shorter problem, where babies, crocodiles and despised persons lead to "Babies cannot manage crocodiles", and a second test that a converse-only chain must never be accepted.

Checking every step of a confident argument is a strong reason for a Guildford teenager to keep learning programming in 2026. The longer argument is in why teenagers should still learn to code in 2026.