AI Literacy

AI literacy is the working understanding a person needs to use AI tools well and to judge when not to trust them. It is narrower than knowing how models are built and broader than knowing which buttons to press: what these systems are good at, where they fail, what they do with what you give them, and which tasks still need a human to check the result. For most staff the useful curriculum is small. That a model produces plausible text rather than verified fact, and that fluency is not evidence. That confidence in the wording says nothing about accuracy. That anything pasted in may be retained, logged or seen by a vendor, so the data rules apply. That output about a named person or a legal, financial or medical matter needs checking against a source. That a tool which searches company documents can only see what it was given access to, so an absent answer may mean an unindexed file rather than a fact that does not exist. The reason to invest in it is that the alternative is not caution — it is a split between people who avoid the tools entirely and people who over-trust them, and both cost the organisation. The second group ships unverified output; the first group leaves productivity on the table and, more often than expected, quietly uses a personal account because the sanctioned path felt forbidding. Training lands when it is specific to the work. A finance team needs to see what a plausible wrong number looks like in a summarised report; a support team needs to see a confident answer citing a policy that does not exist; engineers need to see generated code that compiles and is subtly wrong. Generic awareness slides age quickly and teach nothing about the failure modes people will actually meet. Treat it as the companion to the usage policy rather than a substitute. The policy states the rules; literacy is what makes them make sense, and it is what people fall back on for the cases the policy did not anticipate — which, in a field that changes this quickly, is most of them.