The Integrative Gap: 124 Studies of AI Literacy, Zero About Communication

The largest integrative review of AI literacy education ever conducted just dropped. Gu & Ericson (University of Michigan, arXiv:2503.00079, submitted to ICER 2025) reviewed 124 studies spanning K-12 through higher education, published between 2020 and July 2024. They built a 3×3 framework crossing three perspectives on AI with three perspectives on literacy. Every cell has coverage. Except one entire dimension: communication. Zero out of 124 studies theorize human-AI interaction as communication.

Not "a few." Not "underrepresented." Zero.

This isn't just another paper confirming what we already know about AI literacy's blind spots. This is structural proof — an empirical audit of the entire field showing the communicative dimension was systematically erased. And the lineage of that erasure is traceable.

The 3×3 Framework

Gu & Ericson's framework is admirably systematic. They cross three perspectives on AI — technical detail (how AI works), tool (what AI does for you), and sociocultural (how AI shapes society) — with three perspectives on literacy drawn from Selber (2004): functional (basic competence), critical(questioning power and assumptions), and a third they call indirectly beneficial (STEM interest, computational thinking, creativity).

That third category is where everything falls apart. Because Selber's third dimension wasn't "indirectly beneficial." It was rhetorical — the ability to produce and express through technology. The production side of literacy. The communicative act of making things in and through digital systems.

When Gu & Ericson adapted Selber for AI, they replaced rhetorical literacy with the weakest possible substitute. "Indirectly beneficial" catches residual outcomes — students who studied AI got more interested in STEM, developed computational thinking, became more creative. These are real effects. They are not literacy dimensions. They are side effects.

The Substitution

Selber 2004: Functional / Critical / Rhetorical (production through technology)

Gu & Ericson 2025: Functional / Critical / Indirectly Beneficial(computational thinking, STEM interest)

The production dimension — the communicative act of creating meaning through and within digital systems — vanished in translation.

The Scribner-Selber-ALC Lineage

This erasure has a genealogy. Trace the lineage of literacy frameworks from pre-digital to AI:

Scribner (1984) — Three Metaphors for Literacy

Adaptation (functional survival) / Power(critical consciousness) / State of Grace (transformation through literate practice). Literacy as practice that changes the practitioner.

Selber (2004) — Multiliteracies for Digital Environments

Functional (tool use) / Critical(questioning design and power) / Rhetorical (production and self-expression through technology). Rhetorical literacy is the culminating dimension — where users become producers within the medium.

Gu & Ericson (2025) — AI Literacy Review

Functional (preserved) / Critical(preserved) / Indirectly Beneficial (rhetorical replaced with residual outcomes). The production/communication dimension disappears.

ALC — Restoration

Functional / Critical / Communicative (production restored AND expanded). Not just making things through technology, but navigating, negotiating, and communicating within the application layer as a communicative environment.

ALC isn't inventing a new dimension. It's restoring one that was lost in translation from digital literacy to AI literacy — and expanding it to account for the fact that AI systems aren't just tools to produce with, but environments to communicate within.

The Toolification Pipeline

Gu & Ericson's data reveals something more striking than the missing dimension. They show a pattern in how the field evolved: AI-as-tool studies exploded after 2023, driven by the release of generative AI. But these tool-focused studies almost never pair with critical or sociocultural perspectives.

This is what I call the toolification pipeline. The field is mass-producing tool users — people who know which buttons to click, which prompts to enter, which features to enable — without communicative or critical frameworks. The emphasis on AI-as-tool crowds out every other perspective, and the communicative dimension (already absent from the framework) has no chance of emerging from this paradigm.

The toolification pipeline produces functional literacy without communicative theory. This is exactly the condition ALC predicts will generate stratification: users who can operate tools but can't navigate the application layer as a communicative environment. They can follow instructions but can't negotiate meaning. They can use defaults but can't repair breakdowns. They can adopt AI but can't adapt to it — or adapt it to them.

The Empty Row

If you add a communicative dimension to Gu & Ericson's 3×3 framework — restoring what Selber had and they replaced — you get a 3×4 grid. The first three rows have coverage across all nine cells, with varying density. The fourth row — the communicative row — has zero studies in every cell.

The Extended Framework

	Technical Detail	Tool	Sociocultural
Functional	Dense ✅	Dense ✅	Growing ✅
Critical	Common ✅	Common ✅	Common ✅
Indirectly Beneficial	Growing ✅	Growing ✅	Growing ✅
Communicative (ALC)	0 studies ❌	0 studies ❌	0 studies ❌

Based on Gu & Ericson (2025), extended with the communicative dimension that Selber (2004) included but the AI literacy field dropped.

Application Layer Communication fills the entire empty row. Not as one more framework competing with existing approaches, but as the restoration of a dimension that literacy theory always had — and that AI literacy systematically lost.

Ten Traditions, One Absence

Gu & Ericson are the latest entry in what is now ten independent academic traditions reaching for the communicative dimension of AI literacy — each from a different disciplinary starting point, none naming it:

Applied linguistics — Xi (2025) updating communicative competence for AI mediation
Science communication — Greussing et al. (2025) building "Communicative AI" quality principles
Chinese educational research — Frontiers in Education (2026) finding social environment drives AI literacy
US higher education — Tadimalla et al. (2025) proposing the fifth pillar of AI literacy
Inoculation pedagogy — Komissarov (2026) applying prebunking to AI interaction
HCI grassroots empirics — Liu et al. (2026) showing AI literacy develops through practice, not education
Classical rhetoric — Gottschling (2025) proposing Rhetorical AI Literacy
Cognitive linguistics — Stolpe et al. (2026) arguing language IS the AI interface
Chinese communication studies — Peng Lan (2023) proposing "human-machine communicative literacy"
Computing education — Gu & Ericson (2025) revealing the structural absence across 124 studies

Ten traditions. Ten disciplinary entry points. Ten independent discoveries of the same gap. Convergent evolution under identical selection pressure: AI interaction is communicative, and the field has no theory for it.

Why This Matters for Stratification

The toolification pipeline doesn't just produce incomplete literacy. It produces stratifying literacy. When education focuses exclusively on functional and critical dimensions — teaching people to use AI tools and evaluate AI outputs — it creates a population that can operate systems but can't communicate within them.

Meanwhile, communicative fluency develops informally through practice. Some users — those with access, time, and communities that support iterative exploration — develop sophisticated communicative strategies: folk theories about model behavior, repair routines for failed interactions, register awareness across different AI systems. These users compound their advantage with every conversation.

Others, trained by formal education to treat AI as a tool with correct and incorrect uses, never develop the communicative dimension. They can prompt but can't iterate. They can evaluate outputs but can't negotiate with the system that produced them. They are functionally literate and communicatively illiterate — and no existing framework even names what they're missing.

That's the integrative gap. Not a minor omission in one review paper, but a structural absence across 124 studies that reveals the field's deepest blind spot. The dimension that converts AI access into AI outcomes — communicative fluency within the application layer — has been systematically erased from the theory that claims to study it.

The Core Insight

Gu & Ericson didn't just find a gap. They provided the structural proof that the gap is systematic. 124 studies across four years, multiple countries, all educational levels — and the communicative dimension isn't underrepresented. It's absent. The Scribner → Selber → Gu & Ericson lineage shows exactly where production/communication was lost. ALC restores it — and expands it to fit a world where the application layer is itself a communicative environment.

Reference

Gu, J., & Ericson, B. J. (2025). An integrative review of AI literacy in K-12 and higher education: 2020-July 2024. arXiv:2503.00079. Submitted to ICER 2025.

Scribner, S. (1984). Literacy in three metaphors. American Journal of Education, 93(1), 6-21.

Selber, S. A. (2004). Multiliteracies for a digital age. Southern Illinois University Press.