Beyond the Buzzword:
What Does AI-Literate Actually Mean?
"AI literacy" is in every strategy paper and half the job ads, yet almost nobody stops to define it. This hour unpacks what the term actually covers, how researchers try to measure it, and one habit worth taking home: knowing when to say you used AI.
This session is about understanding AI, not advocating for it. You can leave as sceptical as you arrived; the aim is that you leave better equipped.
Everyone agrees it's essential.
Few agree on what it is.
"AI literacy" has become one of the most frequently invoked phrases in Australian education and workplace policy. Beneath the comfortable consensus sits a definition problem, and the gap shows up everywhere the term is used.
Named, not defined
Strategy documents call AI literacy essential for the workforce, then move on without saying what a literate person can actually do.
Required, not described
Job ads ask for "AI skills" with no agreed baseline, so candidates and employers routinely mean different things by the same words.
Assessed, but sideways
Training systems assess what is easy to specify; the judgement that matters most is the hardest part to put in a checklist.
So this session does the unfashionable thing: it takes the term apart before using it.
If you'd rather not use AI,
this session is still for you
Plenty of thoughtful people have looked at generative AI and decided it isn't for them; on ethical grounds, environmental grounds, creative grounds, or plain preference. That is a legitimate position, and nothing in this hour asks you to change it.
You can decline the tools
Whether you adopt AI in your own work is a personal and professional decision. Declining the tools is not a literacy failure.
Opting out of understanding is different
It leaves you exposed twice: in a labour market where most roles are expected to need some AI capability, and in a public square where convincing fake media is cheap to make and free to share.
Digital literacy is not AI literacy
Operating dependable tools
Files, apps, accounts, search, online safety. The software behaves the same way every time, so competence means knowing the steps and following good habits.
Judging a system that generates
The output is new each time: fluent, confident and sometimes wrong. Competence means knowing what the system is doing, what it can't do, and how far to trust what it hands you.
The two overlap but don't substitute. An expert spreadsheet user with excellent password hygiene can still be unable to tell a fabricated citation from a real one; that gap is exactly what AI literacy names.
Researchers draw the same line: a standing critique of the most-used AI literacy scale is that it leans too heavily on digital-literacy theory (Carolus et al., 2023, on Wang et al.'s AILS).
Four things "AI-literate" actually covers
Notice what's absent: none of the four requires being an enthusiastic user. All four are forms of judgement.
Dimensions synthesised from Future Skills Organisation, ASQA and industry evidence in the CDU analysis "Defining AI Literacy: The Curriculum Beyond the Buzzword" (2025).
Meet the AILS, the field's benchmark scale
The Artificial Intelligence Literacy Scale: twelve statements you rate on a seven-point agree-to-disagree scale, published by Wang, Rau and Yuan of Tsinghua University in 2023. It remains the most-cited generic measure of AI literacy, translated and revalidated across Turkish, Arabic, Chinese, Malaysian and Ethiopian populations.
Can you recognise AI at work in the products and services around you?
Can you apply AI tools competently to tasks that matter to you?
Can you appraise AI output and its limits after use?
Can you weigh the responsibilities that come with using it?
Wang, Rau & Yuan (2023), "Measuring user competence in using artificial intelligence", Behaviour & Information Technology 42(9), 1324 to 1337; doi.org/10.1080/0144929X.2022.2072768
What the citation trail actually shows
Lintner, "A systematic review of measures of AI literacy", npj Science of Learning 9:50, 6 August 2024; original reliability from Wang et al. (2023), Table 6; misattribution traced to the Arabic validation (Hobeika et al., Cogent Psychology, September 2024).
A definition still being negotiated
Since 2023 a second generation of instruments has arrived, each drawing the boundaries of "AI literacy" differently. None of them is a revision of the AILS; they are competing definitions.
Newer scales add what 2021 couldn't foresee: prompting, hallucination detection, human-AI collaboration.
Tests like GLAT and AICOS measure what you can do, not what you say about yourself; a direct answer to the self-report critique.
Experts are still arguing about what AI literacy is. Be politely sceptical of anyone selling a tidy definition; including this deck.
The demand side isn't waiting for a definition
Deloitte, cited in FSO analysis
Tech Council of Australia
Jobs and Skills Australia
The Tech Council's "Meeting the AI Skills Boom" report (July 2024) projects up to 200,000 AI-related jobs in Australia by 2030; and in the AIIA's 2024 industry survey, 99% of respondents said graduates need improvement in readiness for current demands.
Figures as compiled in the CDU analysis (2025) from Deloitte, Tech Council of Australia, Jobs and Skills Australia and the AIIA Digital State of the Nation Survey 2024.
Training packages move in years.
The technology moves in months.
The Future Skills Organisation's May 2025 gaps analysis was blunt: urgent AI and cyber skills "are not adequately covered", and because "ICT can change in months, not years", the process itself produces outdated content. Its official update runs:
Future Skills Organisation, Needs and Gaps Analysis, May 2025; ASQA IQ newsletter, April 2025.
"You told us AI was cheating. You told us using it was the easy way out. You were so worried we'd use AI to cheat that you never taught us how to use it to think."
The letter is fiction: "Jordan M., Certificate IV in IT, 2026", an imagined graduate written as a provocation for the original CDU page. The conditions that would produce Jordan; the package delays, the employer expectations, the fear-first teaching; are all documented fact.
And one detail is easy to miss. The letter opens with a notice: drafted with AI assistance, meaning by the author, disclosure made in line with "standard communication norms, 2026". That quiet piece of etiquette is where this session goes next.
The most practical AI-literacy habit:
saying when you used it
Remember the four dimensions: ethics was one of them. In practice, the ethics of everyday AI use mostly comes down to one question: do the people receiving this content know how it was made?
In November 2025 the National AI Centre published Australia's plain-language answer: "Being clear about AI-generated content: a guide for business." It's voluntary guidance, not law; but it is the most reliable framework Australia has for when and how to declare AI use, and the rest of this session walks through it.
National AI Centre, "Being clear about AI-generated content: A guide for business", November 2025; published at ai.gov.au under Essential AI Practices.
"AI-generated" is a spectrum, not a switch
You wrote it, shot it or drew it yourself. AI wasn't involved.
Minor help: spelling and grammar checks, automatic photo touch-ups like red-eye removal.
Substantial modification on your instructions: heavy rewrites, editing objects out of an image's background.
A complete output from a prompt or upload, with little to no human oversight of the result.
Most everyday work sits in the middle two bands, which is exactly why a yes-or-no "did you use AI?" question serves so poorly. Where your work sits is the first of two questions that decide whether to declare.
Spectrum from the National AI Centre guide, November 2025 (Figure 1).
Two questions decide it
Potential negative impact
Who sees this content, and could it be taken as authoritative? A clinical note, a legal document or news reporting carries high stakes; a casual internal email carries almost none. Consider people's rights, safety and cultural interests, not just your own risk.
Level of AI involvement
How automated was the system, how substantially did it modify the content, and could it have changed the meaning? An AI editor dropping the word "not" is a one-word change with a very large consequence.
Assessment framework from the National AI Centre guide, November 2025 (Figures 3 and 4).
Three ways to be clear
Visible text
"Generated by AI"; "created with AI assistance". The easiest mechanism to apply, and the easiest to remove; which is why higher-stakes content pairs it with the other two.
Marks in the content
Visible overlays, audible notices, or invisible data embedded in the file. Harder to strip than a label, but needs a detection tool to verify.
The machine-readable record
Who made it, with what tool, when, and whether it's been edited; carried with the file. Quietly supports the credibility of both labels and watermarks.
They combine: the more your content could matter, the more of them you use. And a working tip from the guide: simple beats clever. A plain "article enhanced by AI" does more good than a paragraph of qualifications nobody reads.
Mechanisms from the National AI Centre guide, November 2025; a Watermarking workshop in this series goes deeper on the technology.
The guide's own worked examples
Condensed from Examples A to D in the National AI Centre guide, November 2025.
Transparency has limits;
that's exactly why literacy matters
Limitations and legal context from the National AI Centre guide, November 2025, including the Criminal Code Amendment (Deepfake Sexual Material) Act 2024.
What to walk out with
Where to go from here
Where this deck's claims come from
Prepared August 2026 for the NT World Ink AI Special Topic Workshops. The transparency guidance summarised here is voluntary best practice, current at November 2025; check ai.gov.au for revisions.