AUCTC 2026 · Acadia University · June 11

AI as Literacy

Rethinking how we prepare campus communities for an AI-native world.


Doug Langille · Manager, Digital Innovation & Technology
Nova Scotia Community College

doug.langille@nscc.ca · linkedin.com/in/douglangille · digital.douglangille.ca

IDENTITY PRACTICE FLUENCY NSCC
The Problem

You've done the training. Has anything changed?

Lunch-and-learns. Prompt libraries. Completion certificates. Governance task forces.

Participation rates go up. Behavioral change does not follow.

"Six months later, people are doing what they were doing before — except now they have a checkbox marked complete."

Digital fluency is not a competitive advantage. Those not fluent will be left behind — and right now our programs are not building fluency. This is not a training quality problem. It is a category error.

And the programs we're running don't reach everyone who needs them — that gap compounds everything else.

Why Training Reverts

The reversion pattern has evidence.

  • Dutch longitudinal study (222 students, 8 months) — significant AI adoption decline over time. No stable cognitive framework = reversion to prior patterns.
  • Epistemic anxiety — faculty can't audit AI outputs in their domain. Threatening to people who spent 20 years building domain authority. Prompting tips don't touch it.
  • The key finding: professional identity shapes adoption more than technical competency. If students without deep domain identity revert, faculty with 20 years of it have more to lose — and more resistance to address.
Tool adoption programs are not designed to address identity. That is the category error.
44th
of 47 countries in AI training & literacy
Univ. of Melbourne & KPMG, 2025 · n=48,340
3.2/5
NSCC BA faculty confidence guiding students ethically — despite 96% adoption
NSCC internal survey, Spring 2026 · n=25
The Literacy Frame

The word isn't a rebrand. It's a different category.

Gee (1980s–90s): you are literate when you can think in a mode — not just operate its tools. Each time a new cognitive mode emerged, institutions ran tool adoption programs, watched them fail, then recognised fluency was the actual competency.

Reading→ Digital→ Media→ Data→ AI Literacy
  • Long & Magerko (ACM CHI, 2020) — 17 competencies; load-bearing dimension: judgment and disposition, not mechanics
  • UNESCO AI Competency Framework for Teachers (2024) — five dimensions; the one that distinguishes literacy from training: human-centered mindset
  • The measurement gap — instruments test technical understanding. None reliably measures operative disposition
Programs optimize for what they can measure. The load-bearing dimension goes unmeasured — and therefore unaddressed.
A Better Frame

AI is more horse than plough.

A plough is passive. You pick it up, apply force, put it down. A horse has energy you don't — but also its own motion, its own judgment, its own opinion about today's work. You can't just invoke it. You have to show up for it.

Horse + harness + driver + implement = a system that does things neither alone achieves.

  • The harness — the interface between your judgment and the model's power. Most AI failures aren't about the model. They're about the harness being wrong.
  • The driver — has to know the direction. Not "what can this tool do" but "where am I going." That's the skill most training ignores.
  • Institutions are governing AI like a plough — write the policy, define appropriate use, issue the guidance. You don't policy your horse. You train the driver.
The corral is not the point. The work is the point. The corral exists to make the work safe — not to replace the driver's judgment.
The Grimoire Trap

Collecting spells is the beginning, not the end.

Level 1
Incantations — the what
Prompts, templates, mechanics. Necessary. Fits in a workshop. Produces a handout. Feels like training. Where most programs stop.
Level 2
Alchemy — the how
Building prompts from logic, not memory. Iterating with intention rather than desperation.
Level 3
Wisdom — the when
Knowing when to stop, when to go manual, when the tool is in the way. Is this true? Is this me? Is this enough?
Institutions stop at Level 1 not because trainers are lazy — it's the only thing that fits the formats institutions know how to run. That's the structural trap.
The Identity Mechanism

The person who runs keeps running. The person losing weight often doesn't.

Behavior is most durable when tied to identity, not outcome. Every identity-confirming experience is a vote for the self-concept. (James Clear, synthesizing the behavior change literature)

Tool adoption
"Here's how to use Copilot for email summaries." Teaches the incantation. Reverts when attention shifts.
Literacy development
"Use AI to challenge your own analysis — then reconstruct the argument without it." Builds reflective judgment. Confirms the identity.

"I needed to adopt a radically different mindset about how my assessments work."
— Chen et al., AI Academy (arXiv:2509.11999, 2025) · 25 faculty · 5 sessions · gains in 7 of 9 dimensions

The goal is not to train people to use AI. It is to help them become people who think with AI.
IT's Structural Advantage

IT has the catbird seat. Nobody's using it.

IT sees what no other campus function sees: actual tool adoption patterns, the moment of failure, and the populations invisible to professional development programming.

  • Actual usage data — not survey self-report. IT sees what tools people genuinely use, where they get stuck, what they've quietly abandoned.
  • The just-in-time moment — present when someone can't figure something out. That's the teachability moment. No other campus function consistently has it.
  • The invisible infrastructure people — admin staff keeping the machinery together. Not in professional development cohorts. In IT's world every day. Enormous unmeasured literacy gap.
  • Every sector is a tech sector — every role on campus has an AI literacy need. IT is the only function that sees all of them.
  • The catch — IT has to want the seat. Most departments are still in the service-desk lane, not because they chose it, but because nobody invited them out. That's the other gap to name.
Not territorial. IT brings workflow knowledge and data; CTLs bring pedagogical theory and faculty relationships. The combination is what's missing.
Three Strategies

Lead, don't support.

This isn't IT's job alone — but IT is the function that makes the partnership possible. These strategies are for anyone building programs. IT is the one with the data to make them work.

Strategy 01
Lead with diagnosis
Hire the analyst, not the Oracle. The person who thinks in systems, knows where the bodies are buried, has spent years making complicated things work — that person should be in the diagnostic conversation, not just the deployment queue.
Strategy 02
Help people learn how to learn
The mission isn't teaching AI tools — it's teaching people how to learn technology as it changes. Smaller groups, longer duration, embedded in actual work. Design for Level 2 and 3 — judgment, not just mechanics.
Strategy 03
Reframe the metric
Stop reporting completions and activations. Ask: are people making better decisions with AI than they were before? That question, in a leadership meeting, changes what gets built next.
Policy is the corral. You also need to train the drivers. A corral without drivers is infrastructure for nothing.
The Infrastructure Gap

The access problem is arriving at the same time.

Valley library closures — July 20, 2026: Hantsport, Kentville, Lawrencetown, Port Williams, Middleton. Static provincial funding since 2019–20. $1.2B deficit budget.

Libraries are community digital access infrastructure: public computers, digital help desks, informal literacy support for rural, lower-income, and older adult populations least likely to have other access points.

Literacy programs scoped only to enrolled students and employed faculty miss the populations most at risk.

AI literacy is not why libraries are struggling. But closing community digital access infrastructure while AI literacy becomes a workforce baseline is a compounding gap — not a separate story.
For this room: Kentville is in your backyard. The infrastructure loss and the literacy requirement are arriving simultaneously. That is a policy gap.
The Takeaway

One question.

  1. Does the program address all three levels — incantations, alchemy, and wisdom?
  2. Does it create identity-confirming experiences, not just skill demonstrations?
  3. Can practitioners reconstruct their reasoning without the AI? Ask this one next meeting. If the room goes quiet — that's the gap.
  4. Does measurement track derived value, not raw use?
  5. Is it embedded in ongoing practice, or time-bounded?
The real measure is not adoption. It is: who is this person, and how do they think?

The central argument: most institutions have diagnosed their AI problem as a training problem. The actual problem is a category error — they're applying a tool adoption framework to a literacy challenge. Those require different program designs, different timelines, and different success metrics.

The best predictor of future behaviour is relevant past behaviour. Did your previous training programs — digital literacy, data literacy, privacy awareness — produce durable behavioural change? Probably not. So what's the theory that makes the same approach work for AI? The programs aren't broken; they're aimed at the wrong target. Training transfers procedural knowledge. Literacy changes how someone thinks. The diagnosis — "training isn't working" — may already be consensus in your institution. The harder question is what follows from it. Knowing the problem is a category error doesn't automatically produce a different program design. The rest of this deck is the framework for what a different design actually looks like.

Dutch study: Polyportis (2024), Frontiers in AI, 8-month longitudinal tracking of 222 Dutch higher education students. The decline wasn't uniform — students who developed a stable conceptual framework maintained adoption. Those who didn't reverted. Framework-building is the variable, not exposure or skill level. Epistemic anxiety is the mechanism most programs miss entirely. It's not technophobia. It's a rational response: domain experts who can't audit AI outputs in their own field face a genuine authority threat. No amount of prompting tips addresses that. The fix is developing judgment about AI outputs in domain context — which requires extended practice, not a workshop. The NSCC data is the reversion pattern in a real institution. High adoption, low confidence guiding others. People are using AI for their own work but haven't built the framework to teach it or assess it. The confidence gap between personal use and pedagogical application is exactly where literacy programs need to operate.

James Paul Gee (New Literacy Studies, 1980s–90s) argued that literacy is never just a technical skill — it's membership in a discourse community with its own values, ways of knowing, and identity practices. You become literate by participating, not by completing a course. Long & Magerko's 17 competencies (ACM CHI 2020) span five themes: understanding AI, using and applying AI, evaluating AI, AI ethics, and AI design. Most institutional programs address competencies 1–2. The load-bearing ones for durable adoption are 3–5. The UNESCO AI Competency Framework for Teachers (2024) defines five dimensions: human-centred mindset, ethics of AI, AI foundations and applications, AI pedagogy, and AI for professional learning. Note: the companion UNESCO AI Competency Framework for Students (2024) has four dimensions — a different document. When citing in conversation, specify "for Teachers" to avoid confusion. The measurement gap is the policy problem. Current instruments (SNAIL, AILQ, AICOS) measure technical understanding reliably. None reliably measures operative disposition — whether someone actually engages with AI as a thinking partner. Until that's measurable, programs will keep optimizing for the wrong outcome. Falsifiability condition: The training-vs-literacy distinction would collapse if a well-designed training program — run over 8+ weeks, embedded in real work, with small cohorts and reflective feedback — produced the same durable adoption as a literacy-framed program. What "literacy" adds is the identity and disposition layer. If those can be built without the reframe, the category distinction is rhetorical, not structural. That's worth naming.

From the blog post "AI Is More Horse Than Plough" (Digital Doug, 2026). The plough framing is passive: invoke, apply, put down. The horse framing is relational: show up, calibrate, direct. Those are different stances toward the technology, and they predict different adoption trajectories. The harness is the interface between your judgment and the model's capability. Most AI failures aren't model failures — they're harness failures. Same model, same task, radically different output depending on framing and context. The corral — governing where the horse can go — is necessary. But the corral is governance and policy, not driver training. You can have a perfect corral and no drivers. That's where most institutions are. Shelf life caveat: the horse metaphor is calibrated to current LLMs — tools that require active human direction. Agentic AI systems will stress-test the harness-and-driver frame. The underlying principle (judgment and disposition matter more than mechanics) survives that shift. The specific metaphor has a shorter shelf life than the literacy argument itself.

The grimoire is a book of spells — the witch's prompt library. The metaphor names a real failure mode precisely because it looks like literacy development. People leave with artifacts (prompt libraries, templates, checklists) that feel like capability. They're not — they're recipes. Recipes produce the dish when followed correctly; they don't develop the cook. The three levels: Level 1 fits a 90-minute workshop. Level 2 requires iteration and feedback over weeks. Level 3 requires reflective practice embedded in real work over months. Institutions don't have formats for that. Building those formats is the actual program design challenge. The three-question audit (Is this true? Is this me? Is this enough?) is a Level 3 practice — it can be taught as a habit, but only if the program includes repeated practice with feedback, not a one-time introduction.

James Clear's identity framework (Atomic Habits, 2018, synthesizing Duhigg, Fogg, and the broader behavior change literature): identity-based habits are more durable than outcome-based habits because maintenance is identity confirmation rather than continued motivation. Applied to AI adoption: the difference between "someone who uses AI tools" and "someone who thinks with AI" isn't just framing — it predicts what happens under pressure, time constraints, or when a tool fails. Identity-based practitioners adapt; tool-based practitioners revert. AI Academy study: Chen, Tang, Cheng, Chawla, Ambrose & Metoyer (arXiv:2509.11999, Univ. of Notre Dame, 2025). Five sessions over 8 weeks, 25 faculty, gains across seven of nine competency dimensions (six highly significant: assessing AI outputs, contextual knowledge, continual learning, ethical implications, prompt engineering, tool use skills; one moderate: legal knowledge). The critical finding was qualitative: the faculty who showed the most durable change described a shift in how they understood their own role, not a gain in technical skill. That's the identity mechanism operating in the wild.

IT's structural advantage is a vantage point, not a mandate. The argument isn't that IT should own literacy development — it's that IT sees things no other campus function sees: actual tool adoption patterns, the moment of failure and teachability, and the populations who are invisible to professional development programming. The invisible infrastructure people are the concrete test case. Admin assistants, facilities coordinators, lab technicians — enormous AI literacy needs, zero access to programs designed for faculty. IT encounters them daily at the moment of need. That's an untapped literacy intervention at scale. Why IT stays transactional: Most IT departments didn't choose the service-desk lane — they got put there. The catbird seat has been empty not because IT lacks the vantage point, but because the formats that would use it don't exist yet. Building them is the move. What "lead" actually requires: IT leads the diagnostic work — usage data, failure patterns, population mapping. The CTL leads pedagogical design — faculty relationships, program expertise, learning theory. Neither can do this alone. The combination requires someone senior enough to force the partnership and protect it through the first budget cycle. Without that person named in advance, "IT should lead" stays a conference-talk aspiration. Ironic footnote: this deck argues for centering the people being developed — and the students who will actually sit inside these programs never appear. The next slide gets closest. Make of that what you will.

Strategy 1 — Analyst not Oracle: "Oracle" here is a job description — the all-knowing expert brought in to train everyone. That model fails because it treats AI literacy as a knowledge transfer problem. The analyst model treats it as a diagnostic and design problem — someone who understands the institution's actual workflows, failure points, and capacity constraints. That person is usually already inside IT. Strategy 2 — Driver training design principles: Smaller cohorts (8–15 people) allow for the reflective discussion that deeper learning requires. Longer duration (6–10 weeks minimum) allows for iteration between sessions. Embedding in actual work means participants practice on real problems, not contrived scenarios. Cross-functional cohorts (IT + faculty + operations) produce knowledge transfer as a side effect. What kills Strategy 2: The first scheduling conflict. The first budget cycle that needs the headcount back. The director who can't justify six weeks of staff time for a program with no completion certificate at the end. Strategy 2 isn't just a program design choice; it requires someone with enough organizational standing to protect the time and defend the format. Name that person before you start. Strategy 3 — The metric reframe in practice: "What percentage of staff have completed AI training" is a compliance metric. "What are our most effective AI users doing differently, and can we make that legible to others?" is a literacy metric. The second question changes what leadership funds, what IT builds, and what counts as success.

The library closure story is the equity dimension of the literacy argument, not a separate one. If AI literacy becomes a workforce baseline and community digital access infrastructure contracts simultaneously, the gap in who reaches literacy doesn't close — it compounds. The populations losing library branches are the same populations least likely to have other access points: rural households, lower-income adults, older workers navigating career transitions. The Nova Scotia context: Five Annapolis Valley branches closing July 20, 2026. Static provincial funding since 2019–20, $1.2B deficit budget (2026-27), no bridge funding increase. For a post-secondary audience in the Valley, this is local and immediate — not a distant policy abstraction. Note: specific closure data is volatile — the durable argument survives any outcome; the numbers may need updating before delivery. Libraries and literacy broadly: Libraries aren't just digital access points — they're the institution society built specifically to develop literacy at population scale, for free, without enrollment requirements. Closing them while AI literacy becomes a workforce baseline isn't just an access problem — it's dismantling the only infrastructure that ever tried to reach everyone. The design implication: Literacy programs scoped only to enrolled students and employed faculty will systematically miss the populations most at risk. Open sessions, community partnerships, and drop-in formats aren't add-ons — they're what it looks like to take the literacy gap seriously at its actual scale.

These five questions are an evaluation framework, not a checklist. Use them to audit existing programs or stress-test proposals before they're funded. Q1 (three levels): A program that only covers Level 1 will produce compliance metrics and reversion. Ask what the program does after the workshop ends. Q2 (identity-confirming experiences): Skill demonstrations show people what AI can do. Identity-confirming experiences require people to use AI in their own domain, reflect on the output against their own judgment, and articulate the difference. Those are designed differently. Q3 (reconstruct without AI): This is the cognitive sovereignty test. If practitioners can only produce AI-quality work with AI, they've developed dependency, not capability. Programs that never require reconstruction without the tool are building a fragile skill — one that fails under scrutiny, in assessments, or when the tool is unavailable. Q4 (derived value): Activation counts measure reach. "Are people making better decisions?" measures impact. The second question is harder to answer — which is exactly why it rarely gets asked. Q5 (ongoing practice): Literacy is not an event. Any program that ends at the workshop boundary will produce reversion. The design question is what the ongoing practice looks like — communities of practice, embedded feedback, periodic reflection sessions.