Earlier this week I mentioned that while there’s all sorts of demand for “AI literacy,” there doesn’t seem to be a shared understanding of what that means. After floating a few possibilities, I asked my wise and worldly readers for their definitions.

One reader responded with Ohio University’s definition, which includes components of “effective practices,” “ethical considerations,” “subject knowledge” and “rhetorical awareness.” The idea is that students should be able to develop and deploy useful prompts, weigh the inherent biases in data, understand how user expectations shape the product, and have sufficient subject-matter knowledge to judge the output.

Those all strike me as good goals, but they stand in some tension with each other. If you have serious moral or ethical objections to the use of AI, for example, then becoming good at using it is like becoming a more effective criminal. And it’s not clear that students coming up now, having used AI routinely for years, will have had the chance to develop the subject-matter knowledge to discern either subtle biases or small hallucinations.

Another reader offered a useful metaphor:

I like this quite a bit. It makes AI much less threatening and suggests that it can be useful in certain cases. (Cory Doctorow sometimes compares it to spell-check, making a similar point.) And the example of someone who doesn’t know anything beyond microwaved food gets at my anxiety around folks who’ve used it for much of their educational journey.

Part of the role of formal education, in this model, might be to provide counterexamples that would help ground judgments of when and how AI use makes sense. Melt butter in the microwave? Sure. Cook a turkey in it? Um, no …

A variation on the same theme:

This one contains an assessment mechanism, which is helpful. If a student can show effective application of critical judgment, then the tech is beside the point. Still, I remain concerned that effective critical judgment—as opposed to snarky nihilism—relies on relatively broad context knowledge, which is exactly what AI seems to threaten.

A thoughtful reader suggested building on the American Library Association’s definition of information literacy:

As longtime readers know, I’m a huge fan of libraries and librarians. Still, this definition strikes me as much too narrow to just substitute “AI” for “information.” AI takes much of the search work out of the process, but the search work is where much of the learning happens. I suspect the librarians will come up with something more tailored to the new technology soon.

Finally, Annette Vee volunteers her version of “critical AI literacy,” outlined in a thoughtful column. She suggests, “Critical AI literacy is the ability to understand, apply and assess AI operations, uses and outputs.” Her entire column is well worth the read, not least for its opening metaphor about the uses and abuses of composition courses. Her focus on “reflection” may provide a useful way to jump-start critical thought when subject matter knowledge is shaky. The distinction she implies between AI literacy and critical AI literacy helps sell it for me, and it may distinguish what many employers want from what many educators want. It even comes with some useful classroom exercises! I can’t really do her piece justice in a paragraph or two, so I’ll just recommend it highly.

Thanks to everyone who answered the call! AI is a rapidly moving target, but we owe it to our students to try to get this right. If we’re going to have meaningful discussions about AI policy, we should probably have a clear sense of what we’re trying to achieve.