The Pedagogy of AI Mistakes: Fostering Higher-Order Thinking

Hadi Hosseini · 2026

Read The Pedagogy of AI Mistakes: Fostering Higher-Order Thinking by Hadi Hosseini (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-05472.txt.

Introduction

Hadi Hosseini turns a common complaint about AI in the classroom into a teaching tool. Students in his database design course worked with a syllabus that deliberately used AI mistakes. The model played the role of a learning companion whose flawed output students had to analyze, judge, and fix.

The study used mixed methods to track what happened. Structured encounters with AI errors supported reflection on one's own thinking and kept subject standards firm. Hosseini connects the gains to Bloom's taxonomy and to students' sense of their own AI literacy and subject skill.

Catalog record

CREATORHadi Hosseini
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE25 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.05472

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

The Pedagogy of AI Mistakes: Fostering Higher-Order Thinking Hadi Hosseini Abstract As generative AI becomes increasingly integrated into higher education, its frequent errors and hallucinations, often seen as limitations, offer a unique pedagogical opportunity. By framing AI as a “learning companion” whose imperfect outputs prompt analysis, evaluation, and reflection, we argue that instructors can engage students in the fundamental processes of higher-order thinking. This paper presents a design-oriented study in which an AI-integrated syllabus in a database design course deliberately leverages AI’s limitations to foster critical thinking and higher-order cognitive skills aligned with Bloom’s taxonomy of learning. Using a mixed-methods approach, we examine how structured interaction with AI-generated errors supports metacognitive engagement, reinforces disciplinary rigor, and relates to students’ perceived AI literacy and subject-matter competency. Keywords : Learning Theory, Database Design, AI, Scaffolding, Higher-Order Thinking 1 Introduction Generative Artificial Intelligence (AI), and in particular Large Language Models (LLMs), has sparked significant debate about its role in education. Recent works demonstrate both the growing presence of AI tools in learning contexts and their effectiveness in supporting student task completion [ 2 , 17 , 7 ] . In computer science education, LLMs are increasingly introduced from the first day of instruction in introductory programming courses [ 16 ] and are being explored across a range of paradigms, including collaborative and pair-programming settings [ 4 , 10 ] . At the same time, prior studies caution that AI-generated code can obscure underlying concepts, as students may successfully produce solutions they do not fully understand [ 10 ] . These developments have amplified debates within the education community, with some advocating for restriction or prohibition of student AI use [ 12 , 5 ] . However, such approaches risk overlooking both the inevitability of AI’s integration into educational practice and its potential to serve as a pedagogical companion when thoughtfully designed. While over-reliance on AI can impede the development of core cognitive skills [ 19 ] , its scaffolded use can position AI tools as “learning companions” that support deeper conceptual understanding and critical thinking [ 13 , 18 ] . Because LLMs can produce errors, hallucinations, or context-insensitive responses, effective instruction must extend beyond domain knowledge to include the ability to evaluate, contextualize, and revise AI-generated outputs. Guided by Bloom’s taxonomy [ 11 ] , we argue for a shift in syllabus design and classroom practice, from the foundational cognitive levels of knowledge, comprehension, and application toward the higher - order processes of analysis, evaluation, and synthesis. 1 1 1 Other frameworks such as Webb’s Depth of Knowledge [ 21 ] and Marzano’s taxonomy [ 14 ] also encompass higher-order cognitive skills, including analysis and evaluation. In this sense, the fallibility of LLMs becomes pedagogically productive : by requiring students to interrogate and validate AI-generated responses, instructional activities can actively engage learners in higher-order reasoning and critical judgment, aligning AI use with established learning theory. Students learn more effectively by actively critiquing and revising solutions than by passively receiving feedback [ 20 , 9 , 8 ] . Thoughtfully integrated AI, particularly through its imperfect or error-prone outputs, can further foster reflective engagement with course content, supporting deeper mastery. In this course-redesign study, we integrate AI-centric activities such as critical evaluation of model outputs, AI-guided problem decomposition, and structured reflection on AI use within core instructional tasks. We ask the following research questions: (i) How can generative AI be meaningfully integrated into syllabus design to promote higher-order thinking, as guided by Bloom’s taxonomy? (ii) What role do students’ prior AI and domain knowledge (self-reported and actual) play in shaping their engagement with AI-integrated activities? 1.1 Contributions We anchor our investigation in the context of an undergraduate database design course, using it as a concrete framework to explore AI - integrated pedagogy. We introduce a course design that integrates generative AI activities, e.g., prompt engineering, error - analysis exercises, and themed case studies, into each module to foster analysis, evaluation, and creation. In particular, we make the following contributions: 1. We present a redesign of an undergraduate database course that integrates generative AI as a pedagogical companion to explicitly promote metacognitive skills, including self-monitoring, error detection, and critical evaluation of AI-generated solutions, thereby shifting learning emphasis toward higher-order cognitive processes. 2. We propose a mixed-methods evaluation framework that combines self-reported literacy , objective AI competency measures, and subject-matter performance assessments, enabling a nuanced examination of how students engage with and learn alongside generative AI. 3. Through pre/post assessments and correlational analyses, we demonstrate gains in database competency and identify systematic misalignment between students’ perceived AI abilities and their objectively measured performance, highlighting the pedagogical importance of calibration and reflective judgment in AI-integrated learning. 2 Course Structure: an AI-Integrated Design Our design is guided by the premise that higher-order cognition is best developed through structured opportunities for evaluation and revision, particularly when learners engage with imperfect responses. 2.1 Core Topics The course is structured into a series of several interconnected modules that span one or two weeks. Topics include: introduction to Database Management System (DBMS) architecture, Entity–Relationship (ER) modeling, schema normalization, relational databases and SQL, NoSQL, key - value stores, and graph databases, culminating in a capstone integration project. 2.2 Module Components There are four core components that support scaffolded learning—lesson plans, short videos, AI - focused exercises, and curated readings. These components form a cohesive design that systematically advances students from foundational knowledge to higher - order analysis, evaluation, and creation, aligning with Bloom’s Taxonomy of learning. 2.3 Lesson Plans: What, Why, and How Each module begins with a detailed lesson plan that states explicit learning outcomes , clarifying what students should be able to do (e.g., “Design and implement a 3NF schema for a real - world scenario”). We frame these outcomes through three guiding questions: 1. What? the core concepts (e.g., “What is second - normal form?”), 2. Why? the pedagogical rationale (e.g., “Why does normalization reduce data redundancy and improve integrity?”), and 3. How? the procedural mechanics (e.g., “How do you decompose relations to eliminate partial dependencies?”). This What/Why/How structure emphasizes metacognitive reflection and primes students to connect theory with practice. 2.4 Short Video Components Each module includes a 4–10 minute introductory video that motivates the topic and sketches the big picture. These are supplemented by one or two supporting video clips; for example, a screencast demonstrating interactive normalization of a sample ER diagram or a short interview with industry practitioners on graph query use cases. Students view these videos before tackling hands - on exercises, ensuring a flipped - classroom dynamic. 2.5 AI Modules Central to our design is a weekly AI Module, structured to draw on Bloom’s taxonomy and metacognitive theory through an iterative critique-refinement cycle. • Prompting Strategies : Students engage at the level of application by learning techniques such as zero-shot vs. few-shot prompting, chain-of-thought prompting, and temperature tuning [ 6 ] . For example, when generating SQL snippets, students learn to include schema definitions and sample rows in their prompt to elicit accurate SELECT statements. • Failure-Mode Analysis : To promote analysis and evaluation , we present “what-can-go-wrong” cases such as an AI - generated ER diagram that violates referential integrity or produces nonsensical attribute

Read the complete text