Schools swapped AI bans for supervised failure, and the cost moved to teacher training

The misspelled-map demo works because it fails where a student can check it without asking anyone ^[2]. That is also its expiry date. Garbled country names from a generator are exactly the class of embarrassment a vendor patches, and when it goes, the lesson has to be rebuilt out of whatever the model gets wrong next term. Amanda Bickerstaff of AI For Education recommends the demo for its effect on over-trust ^[3], which is an argument for keeping someone on staff who can go find the next one.

Charleston County went to a third party rather than to a model provider for the policy work, contacting AI For Education a year ago, an organization that had already helped draft guidance for Chicago Public Schools and Houston ^[8]. Phase 1 was the policy itself, written with input from teachers, students and parents ^[9]. There is no settled definition of AI literacy or of how it should be taught ^[19], so a district writing its own guidance is also writing the standard it will later be measured against.

The student course teaches verification through a research-paper scenario in which the tool produces made-up studies and states them with unflinching authority, and it instructs students to check any factual claim it makes ^[13]. A separate section tells them not to enter personal information, since conversations can be stored, used for training and potentially leaked ^[14]. Both controls are instructions to teenagers rather than terms in a contract, which is the cheaper of the two ways to handle that risk.

The demand side is documented. The district's own focus groups found student use ubiquitous but unguided, with teachers and students asking for rules and instruction ^[10], and nationally most teenagers and teachers use AI for schoolwork while teaching themselves, with very few reporting formal training or a technical understanding of how it works ^[11]. Ray Knauer left the summer session with examples of bias and hallucination for his AP Research class ^[17], and described the open problem as finding a balance that neither scares students away from AI nor encourages them to run to it ^[18]. Multiply that preparation across a faculty and you have the actual size of the AI literacy line item.