From Understanding to Creation: A Prerequisite-Free AI Literacy Course with Technical Depth Across Majors

Amarda Shehu

cs.CYMar 19, 2026

From Understanding to Creation: A Prerequisite-Free AI Literacy Course with Technical Depth Across Majors

Authors: Amarda Shehu

Organizations: Professor of Computer Science; Vice President and Chief AI Officer, George Mason University, Fairfax, Virginia, USA

Abstract

Most AI literacy courses for non-technical undergraduates emphasize conceptual breadth over technical depth. This paper describes UNIV 182, a prerequisite-free course at George Mason University that teaches undergraduates across majors to understand, use, evaluate, and build AI systems. The course is organized around five mechanisms: (1) a unifying conceptual pipeline (problem definition, data, model selection, evaluation, reflection) traversed repeatedly at increasing sophistication; (2) concurrent integration of ethical reasoning with the technical progression; (3) AI Studios, structured in-class work sessions with documentation protocols and real-time critique; (4) a cumulative assessment portfolio in which each assignment builds competencies required by the next, culminating in a co-authored field experiment on chatbot reasoning and a final project in which teams build AI-enabled artifacts and defend them before external evaluators; and (5) a custom AI agent providing structured reinforcement outside class. The paper situates this design within a comparative taxonomy of cross-major AI literacy courses and pedagogical traditions. Instructor-coded analysis of student artifacts at four assessment stages documents a progression from descriptive, intuition-based reasoning to technically grounded design with integrated safeguards, reaching the Create level of Bloom's revised taxonomy. To support adoption, the paper identifies which mechanisms are separable, which require institutional infrastructure, and how the design adapts to settings ranging from general AI literacy to discipline-embedded offerings. The course is offered as a documented resource, demonstrating that technical depth and broad accessibility can coexist when scaffolding supports both.

Figures & tables

Panel of Figure 1 — Figure 1: Cross-major AI literacy courses positioned along two dimensions: audience breadth (horizontal axis) and the highest level of Bloom’s revised taxonomy [ 15 ] that students reach through the course’s own assessments (vertical axis). Color depth tracks Bloom’s level, with lighter greens at the Remember and Understand levels deepening to forest green at Create. Existing courses for non-majors cluster at the Understand and Apply levels. The literacy-competency-fluency taxonomy of [ 14 ] is shown for reference but does not span the full Bloom’s progression.

Course / Framework	Institution	Prereqs	Bloom’s ceiling (non-majors)	AI engagement mode	Ethics model	Culminating assessment
Frameworks
Long & Magerko [ 1 ]	–	–	Unspecified	–	Included in competencies	–
AI4K12 [ 2 ]	–	–	Apply	–	Societal impact (big idea 5)	–
Tadimalla & Maher [ 3 ]	UNC Charlotte	None	Apply ∗	Use tools; reflect	Parallel (Pillars 3–4)	AI literacy self-assessment
Courses
CS 309 [ 5 ]	UT Austin	None	Apply	Use sandboxes; solve non-coding problems	Parallel (videos + essays)	Individual written ethics project

Table 1: Cross-major AI literacy courses and frameworks at the university level. Bloom’s ceiling indicates the highest level of Bloom’s revised taxonomy that non-major students are expected to reach through the course’s own assessments. AI engagement mode characterizes the dominant relationship students have with AI systems. Ethics model distinguishes whether ethical reasoning is delivered as a separate module (parallel), woven into technical content through a shared structure (integrated), or treated as the organizing principle of the course (ethics-centered). The final column identifies what the highest-stakes assessment asks students to produce or perform.

Tradition	Core principle	Design problem addressed	Instantiation in UNIV 182
Schema building through repeated traversal [ 24 ]	Robust understanding develops when learners encounter the same conceptual structure in progressively more complex contexts.	Mixed-background students lack a stable frame for organizing new technical material; each topic feels isolated and disposable.	A single conceptual pipeline (problem $\rightarrow$ data $\rightarrow$ approach $\rightarrow$ train/configure $\rightarrow$ metrics $\rightarrow$ evaluate $\rightarrow$ improve $\rightarrow$ reflect) is traversed repeatedly, first with simple classifiers, then with neural architectures, then with ethically-consequential deployments.
Studio-based pedagogy (design education)	Concepts become disciplined action through structured work sessions with real-time critique, documented decisions, and peer accountability.	Non-majors can absorb concepts in lecture but fail to translate them into methodical practice; misconceptions remain latent until high-stakes assessments.	AI Studios at key junctures: building classifiers (no-code, then light code via Google Colab [ 30 ] ), preparing debates, designing experiments, developing prototypes. Protocols require written plans, evidence logs, and rationale before execution.
Assessment as learning [ 25 ]	Assessments designed as discovery tasks produce new understanding rather than measure existing knowledge.	Traditional content-verification exams reward recall and penalize the exploratory reasoning that non-majors need to develop; students with weaker technical backgrounds are disproportionately disadvantaged.	Cumulative portfolio in which each assessment feeds competencies forward: sector briefs supply evidence for debates; CNN ablations supply experimental discipline for the midterm; the midterm’s “fluency $\neq$ correctness” finding supplies evaluative habits for the final project.
Argumentation pedagogy; structured discourse	Holding opposing claims simultaneously, grounded in shared evidence, develops evaluative judgment that neither lectures nor reflection essays can produce alone.	Ethical discussion conducted without technical grounding reduces to opinion exchange; students form positions from headlines rather than from analysis.	“Two Things Can Be True” debates: teams defend opposing positions on AI deployment (facial recognition, AI art, healthcare, defense), with every argument required to follow the shared technical pipeline (problem $\rightarrow$ data $\rightarrow$ approach $\rightarrow$ metric $\rightarrow$ value judgment).

Table 2: Pedagogical foundations informing the design of UNIV 182. Each row identifies a tradition from the learning sciences or professional education, the principle it contributes, the design problem it addresses in cross-major AI literacy instruction, and the course mechanism through which it is realized.

Topics	Activities	Assessments
Foundations: The Pipeline, Data, and First Encounters
Introduction to AI; history from Aristotle to Hinton; AI in everyday life	Lectures; in-class discussion and debate comparing early aspirations with realities	–
How AI works: data, learning paradigms (supervised, unsupervised, reinforcement), evaluation	Lectures; data quality and bias case studies	Quiz (in-class)
When data and models go wrong	Studio : Hands-on with Teachable Machine; high-profile bias cases	HW1 assigned (AI ethics in a chosen domain)
Societal impacts of AI across domains (healthcare, education, climate, policy)	Flipped classroom: small-group discussion seeding HW1 and identifying debate topics	–
Structured Debates, Round 1: Responsible AI Innovation

Table 3: Topic sequence, activities, and assessments for UNIV 182 (Fall 2025). The conceptual pipeline and ethical reasoning run concurrently throughout; what increases is the technical sophistication at which both are practiced. Fig. 2 visualizes this structure.

Panel of Figure 2 — Figure 2: The structure of UNIV 182 as two continuous threads running beneath rising technical sophistication. The conceptual pipeline (orange ribbon) and ethical reasoning (green ribbon) run uninterrupted across the full semester. The blocks represent the technical content students engage with, rising in height to indicate increasing sophistication: from black-box models (Teachable Machine) through perceptrons and MLPs, CNNs and RNNs, transformers and LLMs, generative AI and agents, to the final build-and-defend project. Assessment markers are annotated below. Every assessment intersects both threads: no learning task in the course is purely technical or purely ethical.

Assessment	Bloom’s Level + Scaffold	Feed-Forward Function
HW1: Sector brief (individual)	Un+An: Map case to pipeline; convert intuition to referenced claim	Evidence pool for debates; team recombination
Debates: “Two Things True” (team, live)	Ev: Studio-prepped argument via shared pipeline across high-stakes deployments	Normalizes complexity of real-world evaluation; reinforces evidence-based reasoning
HW2: CNN + saliency (individual, notebook)	Ap+An: Studio classifiers $\rightarrow$ mechanistic accountability via controlled ablations, saliency interpretation	Experimental discipline (logs, controlled edits) transfers to midterm protocols
Midterm: Chatbot reasoning (team)	An+Ev: Design probes; controlled comparison; score correctness $\neq$ explanation separately	“Fluency $\neq$ correctness” finding; evaluative habits for HW3 and final safeguards
HW3: Create with AI (individual)	Ap $\rightarrow$ Ev: Create artifact, log process, evaluate quality/bias/ownership/privacy/transparency	Ethics as experienced constraints; risk categories carry into final
Final: AI artifact + pitch (team, external)	Cr+Ev: Checkpoint studios $\rightarrow$ timed pitch to external evaluators; defend problem, data, design, safeguards	Transfer test: method and accountability under scrutiny by evaluators external to the course

Table 4: Assessment sequence (chronological). Each assessment builds competencies that feed forward into subsequent assignments. Bloom’s level: Un = Understand, Ap = Apply, An = Analyze, Ev = Evaluate, Cr = Create.

Figure 3: Assessment submission rates across the semester for all enrolled students (solid line, $N=38$ ) and course completers (dashed line, $N=33$ ). Assessments are ordered chronologically. For all enrolled students, submission rates held steady near 89% through the first three assessments, rose to 95% at the midterm, dipped to 79% at HW3 (an individual generative AI assignment in the final quarter), recovered to 100% for Debates Round 2 (an in-class team activity), and settled at 84% for the final project. The completers line shows that among students who finished the course, participation remained above 88% at every assessment point, with the same HW3 trough and Debates Round 2 recovery visible at a smaller magnitude.

Item	Mean	Median	SD	Min
Environment facilitated engagement with content	4.58	5.00	0.51	4
Technologies/tools increased engagement	4.58	5.00	0.51	4
Encouraged expression of diverse perspectives	4.58	5.00	0.67	3
Learned through variety of opportunities	4.45	5.00	0.69	3
Gained understanding of main concepts	4.42	4.00	0.51	4

Table 5: Selected items from institutional course evaluations, Fall 2025. All items scored on a 1–5 Likert scale. Items were selected for relevance to the course design claims; the full instrument contained 14 items.

Figure 4: Joint distribution of technical description mode and ethical reasoning mode across HW1 submissions. Each submission was coded along two dimensions: how the student described the AI technology (Descriptive, Pipeline-Mapped, or Mechanistic) and how the student reasoned about ethical consequences (General Concern, Pipeline-Referenced, or Causal Chain). The dashed border highlights the modal cell (42%): students who organized their technical analysis around the information management pipeline but whose ethical reasoning remained at the surface level, referencing the framework without tracing causal paths from design choices to downstream consequences. Only 10% of submissions contained causal chain reasoning at this stage.

Figure 5: Distribution of argument grounding levels across all distinct arguments coded from Debate Round 1. Arguments were classified as general societal concern (claims from intuition or media framing), pipeline-grounded (tracing at least two steps of the technical pipeline), or technically-specific (referencing model architecture, training methodology, or failure modes). Pipeline-grounded reasoning appeared in 39% of arguments, compared with 10% of HW1 submissions that contained causal chain reasoning. Technically-specific reasoning was essentially absent, consistent with the course design: the debate preceded the deep learning modules that would supply architectural knowledge.

Figure 6: Prevalence of three evaluative capabilities across HW3 submissions. Each submission was coded for the presence or absence of empirical testing (controlled variation or systematic probing of AI output), fluency-correctness distinction (separating surface quality from validity or originality), and provenance tracing (attempting to identify where output elements originated in training data or prior works). All three capabilities were essentially absent in HW1 submissions. Their emergence in HW3 is consistent with the scaffolding objective: the AI Studios built experimental discipline, the midterm supplied the fluency-correctness finding, and the cumulative portfolio extended into independent creative evaluation.

Figure 7: Prevalence and depth of four design integration competencies across final project team checkpoint documentation. Each team was coded for data pipeline skepticism (identifying specific bias sources in their own pipeline), causal ethics reasoning (tracing design choices to downstream consequences with evidence), failure mode design (designing around limitations rather than merely disclosing them), and constraint specification (defining what the system should not do). Bars show the proportion of teams exhibiting each competency with specificity (dark), at a surface level (medium), or not at all (light). The near-universal presence of these competencies, which were essentially absent in HW1, is consistent with the claim that the scaffolding mechanisms accumulated across the semester.

Assessment	Stage-specific code	Bloom’s level	Justification (cognitive process)
HW1	Descriptive (technical) or General Concern (ethical)	Remember / Understand	Recall or surface-level characterization of AI systems; no decomposition or application of a framework
	Pipeline-Mapped (technical) + Pipeline-Referenced (ethical)	Apply	Use of a provided framework (the information management pipeline) to organize analysis, without independent decomposition
	Mechanistic (technical) or Causal Chain (ethical)	Analyze	Tracing relationships between technical components and downstream consequences; differentiating causes from effects
Debate Round 1	General societal concern	Apply	Constructing claims from intuition or media framing; assembling an argument without decomposing it through the pipeline
	Pipeline-grounded	Analyze	Tracing at least two steps of the technical pipeline to derive a conclusion; differentiating data, model, and outcome
	Technically-specific	Evaluate	Referencing architecture or training methodology to judge system behavior against an independent technical criterion

Table 6: Mapping of assessment-specific codes to Bloom’s revised taxonomy levels. Each row shows the assessment stage, the code assigned during stage-specific analysis, the Bloom’s level to which the code was mapped, and the cognitive process that justifies the mapping. Cognitive process definitions follow from work in [ 38 ] : Remember/Understand involves retrieving or constructing meaning from information; Apply involves carrying out a procedure in a given situation; Analyze involves breaking material into parts and determining how parts relate to one another; Evaluate involves making judgments based on criteria and standards; Create involves putting elements together to form a novel, coherent whole.

Figure 8: Distribution of coded reasoning across Bloom’s revised taxonomy levels at four assessment stages. At each stage, the proportion of reasoning at each taxonomic level was estimated from the stage-specific codings reported in the preceding subsections. In HW1, reasoning was concentrated at Remember/Understand and Apply (90% combined). By the final project, 95% of coded reasoning operated at the Analyze, Evaluate, or Create levels, with Create appearing at substantial scale (40%) for the first time. The upward trajectory of the distribution across the semester is consistent with the claim that the course’s scaffolding mechanisms produced cumulative growth in higher-order competencies among non-major undergraduates.

AI Literacy

AI literacy develops people's ability to understand, use, evaluate, and shape AI systems responsibly across everyday, educational, and professional settings.

See all 43 papers

Apr 28, 2026

Beyond Tool Adoption: A Practical Five-Stage Developmental Continuum for AI Literacy in Higher Education

Artificial intelligence (AI) literacy is increasingly recognized as a foundational competency for all university graduates. Yet students' engagement with AI tools often clusters at two extremes: avoidance driven by fear, mistrust, ethical concern, or lack of access, and uncritical reliance that produces fluent output while masking misunderstanding. Existing AI literacy frameworks provide valuable competency definitions, but most offer limited guidance for diagnosing where learners begin and how they progress toward responsible, critical engagement. This paper proposes a five-stage AI Literacy Continuum: 0) Not Yet Engaged, 1) Uncritical Use, 2) Informed Use, 3) Critical Evaluation, and 4) Improvement --that describes developmental orientations toward AI use in higher education. The continuum complements dimensional frameworks by providing educators with a practical diagnostic and instructional pathway aligned with international frameworks, including UNESCO and OECD. We present a design-based implementation case from North Carolina State University, where credit-bearing courses and intensive hands-on workshops engaged more than 330 participants between Fall 2024 and Spring 2026. Because the implementation did not use a validated pre/post instrument or comparison group, we frame the findings as observational and practice-based: participants exhibited behaviors consistent with movement from non-engagement or uncritical use toward informed engagement, while sustained and discipline-embedded experiences produced stronger evidence of critical evaluation and improvement-oriented practice. We discuss curricular pathways, opportunity considerations, assessment strategies, and argue that AI literacy should be understood not as tool adoption alone but as a developmental capacity to understand, evaluate, and responsibly apply AI systems in disciplinary and societal contexts.

J. Paul Liu, Rachel Levy

1. Dept of Marine, Earth, and Atmospheric Sciences; 2. AI Hub for Science; 3. Center of Geospatial Analytics; 4. Data Science and AI Academy; 5. Department of Mathematics North Carolina State University

Higher Education AI in Education AI Literacy Responsible AI

Ask PDF

May 14, 2026

Mapping AI Programs in the U.S: A Status Report from Early 2026 and an Analysis of AI Majors and Minors

We present a report on the status of undergraduate Artificial Intelligence (AI) programs in the United States in Spring 2026. In so doing, we 1) describe our scraping and mapping tools, which dynamically update to track the state of AI education in the U.S., and 2) create a historic record at a time of great upheaval. The tool we developed, available at https://cicmap.ai, detects, scrapes, and displays data from more than 350 undergraduate AI programs--majors, minors, concentrations, and certificates--at 4-year universities. Our tool searched over 560 institutions to locate these programs, a sample that represents 86% of all undergraduate Computer Science (CS) graduates in the U.S. This tool allows prospective students, guidance counselors, administrators, and faculty to easily access AI program requirements and is designed to continually update as new programs emerge. To the best of our knowledge, this survey represents the most comprehensive snapshot of the state of AI programs in the U.S. to date. With this work we offer three important contributions: 1) a record of AI programs in the U.S. at a time of great upheaval; 2) a tool to explore AI programs and their requirements; and 3) an analysis of the courses required for 66 AI majors and 87 AI minors. Our analysis of majors and minors shows great variability in the size and the requirements of these degrees, but we note two takeaways. First, not all majors require a general AI course, but if they don't, they do require a Machine Learning (ML) course. Second, while more than a third of majors require an Ethics in AI course, just under a quarter of AI minors do.

Felix Muzny, Carolyn Jones, Carter Ithier +3

Center for Inclusive Computing Khoury College of Computer Sciences Northeastern University Boston, Massachusetts, United States

AI in Education

Ask PDF

Mar 5, 2026

The Trilingual Triad Framework: Integrating Design, AI, and Domain Knowledge in No-code AI Smart City Course

This paper introduces the "Trilingual Triad" framework, a model that explains how students learn to design with generative artificial intelligence (AI) through the integration of Design, AI, and Domain Knowledge. As generative AI rapidly enters higher education, students often engage with these systems as passive users of generated outputs rather than active creators of AI-enabled knowledge tools. This study investigates how students can transition from using AI as a tool to designing AI as a collaborative teammate. The research examines a graduate course, Creating the Frontier of No-code Smart Cities at the Singapore University of Technology and Design (SUTD), in which students developed domain-specific custom GPT systems without coding. Using a qualitative multi-case study approach, three projects - the Interview Companion GPT, the Urban Observer GPT, and Buddy Buddy - were analyzed across three dimensions: design, AI architecture, and domain expertise. The findings show that effective human-AI collaboration emerges when these three "languages" are orchestrated together: domain knowledge structures the AI's logic, design mediates human-AI interaction, and AI extends learners' cognitive capacity. The Trilingual Triad framework highlights how building AI systems can serve as a constructionist learning process that strengthens AI literacy, metacognition, and learner agency.

Qian Huang, King Wang Poon

Lee Kuan Yew Center for Innovative Cities, Singapore University of Technology and Design

Generative AI in Education Human-AI Collaboration Educational Technology AI Literacy

Ask PDF

Abstract

AI Literacy

Explore similar work

Beyond Tool Adoption: A Practical Five-Stage Developmental Continuum for AI Literacy in Higher Education

Mapping AI Programs in the U.S: A Status Report from Early 2026 and an Analysis of AI Majors and Minors

The Trilingual Triad Framework: Integrating Design, AI, and Domain Knowledge in No-code AI Smart City Course