Can education set the agenda for AI? Simulated innovation and pedagogical directionsetting in teacher education

Synthesis: Fajardo-Ramos, Chiappe and Martínez-Pérez ask whether education can set the agenda for AI rather than keep adapting to it, and answer with two concepts rather than data. It is a Conceptual Analysis: no sample, no instrument, no cases, no test. The argument runs from a diagnosis, that educational technology repeatedly produces simulated innovation, defined by four characteristics (symbolic modernization, pedagogical continuity, evidentiary displacement, externalized directionality), to a counter-mechanism, pedagogical primacy, which asks educational communities to define purposes and limits before selecting tools, and then to teacher education as the site where four capacities would form. Generative AI matters for an evidential reason: it produces the essays and feedback that institutions treat as proxies for thinking, loosening the link between product, process, authorship and evidence. It offers a vocabulary and explicit refutation conditions, but cannot show how often simulated innovation occurs or what a pedagogy-led program changes.

Key Findings

No data are reported. These are conceptual claims, tied to the literature the paper assembles.

1. Simulated innovation is defined by four characteristics held together: symbolic modernization, where tool presence signals progress before value is shown; pedagogical continuity, where teaching and assessment formats stay intact; evidentiary displacement, where adoption rates and dashboards stand in for learning; and externalized directionality, where the technology release cycle sets the direction of change.
2. It is separated from four neighbours: technological determinism, which claims technology drives change by its own logic; digital transformation, of which it is one failure mode; technology adoption, which records only uptake; and technological solutionism, of which it is the institutional outcome.
3. Generative AI's distinguishing property is evidential: models produce explanations, arguments and feedback that resemble evidence of understanding, so pressure falls on the relation between product, process, authorship and learning.
4. Pedagogical primacy mirrors the diagnosis: justify use through an explicit learning problem, reconsider roles and tasks before choosing a tool, privilege process evidence such as drafts and oral defense, and deliberate publicly about the futures AI should serve.
5. Pedagogically substantive integration is given four conditions: it answers an explicitly formulated problem, makes reasoning more visible, strengthens Learner Agency rather than dependence on automation, and answers to fairness, privacy and assessment validity.
6. Teacher education is assigned four capacities: diagnostic (does this use address a defined problem), design-oriented, ethical and epistemic (bias and authorship as pedagogical matters), and ecological (shared norms with students and colleagues).

A diagnostic, not a theory of technology

The four characteristics are relational, not absolute: none says technology is bad or adoption fake; each names a gap between what an institution displays and what it changed. Drawing on Foucault and Althusser, the authors add that standardized assessment long predates learning analytics, so technology-led innovation intensifies tendencies schooling already had rather than corrupting a pure field.

Why generative AI is treated as an inflection point

The argument rests less on capability than on fluent output standing in for the artifacts institutions read as intellectual work. The authors decline to call AI a cognitive partner, since the phrase attributes intentionality the systems lack, and use AI-mediated pedagogical support instead: use within a design that keeps the learner responsible for verification and revision. They also insist Scaffolding is conditional, requiring contingent support that is withdrawn, so AI does not scaffold merely by answering. The stakes are epistemic: the same explanation is valuable when students must critique it, and simulated when it replaces the reasoning it should support.

Teacher education as the site, and the ecology that limits it

Teacher education is chosen because it is where purposes become design criteria. A tool-centered model begins with available systems and teaches prompts; a pedagogy-led model begins with the learning, inclusion and authorship it wants to strengthen. AI literacy here is the capacity to decide when use is justified and what evidence would show it worked. Hence the move from episodic workshops to developmental ecosystems combining conceptual work, guided experimentation and classroom inquiry. The program cannot rest on teachers alone: policies rewarding product completion, workload pressure and platform defaults organized around speed defeat well-prepared teachers; teacher judgment reframed as a monitored variable is how that pressure lands.

What would be needed to test it

Nothing here operationalizes the constructs. There are no indicators, thresholds or worked cases; illustrations are hypothetical, and the claim that simulated innovation is recurrent rests on citation rather than a count of instances. A test would need a coding instrument for the characteristics applied to documented institutional cases with coder agreement reported; evidence that the characteristics travel together, since the construct asserts a pattern rather than a trait; and a comparison of pedagogy-led against tool-training programs, with teacher judgment as the outcome. The authors also name one refutation condition: the argument weakens if AI implementations show sustained educational value beyond visibility.

What this means for practice

Instructors should begin from the problem rather than the tool: write down what learning or judgment problem is being addressed. The paper's routines make process visible, requiring students to critique AI output, compare it with disciplinary sources and defend a revision. Faculty developers should treat prompt training as insufficient and build repeated cycles of design, consequence analysis and peer review. Administrators should write policies that define criteria for validity, equity and teacher judgment rather than lists of permitted tools.

Limitations

• There is no empirical component: no cases, no sample, no operational definition, so the claim that simulated innovation is recurrent rests on citation rather than any count.
• The evidentiary base is selective critical scholarship on educational technology and assessment, largely Selwyn, Williamson and Bearman, so rival readings of the same adoption patterns are not weighed.
• The counter-concept is not costed: the paper does not examine what an institution forgoes by refusing a tool, or who bears the added labor of process-sensitive assessment.
• Students appear as intended epistemic agents, but no student data or voice informs the framework, and the examples of substantive versus simulated use are hypothetical.

Citation

Fajardo-Ramos, D. C., Chiappe, A., & Martínez-Pérez, S. (2026). Can education set the agenda for AI? Simulated innovation and pedagogical direction-setting in teacher education. Frontiers in Education, 11, 1897627.