🌱 Tony's Memex Tony's Intertwingled Memex
Digital Garden | OER Curriculum
Home / Digital Garden / Authentic AI Policy and Redline Audits

Authentic AI Policy and Redline Audits

Problem Statement

Standard institutional policies that attempt to outright ban artificial intelligence create an adversarial classroom climate, incentivizing covert cheating and arms-race policing. When students face uncompromising deadlines or complex calculus derivations, prohibition does not foster discipline; it drives students to copy AI outputs blindly to harvest points. This practice prevents learners from engaging in the cognitive struggle necessary to build deep mathematical intuition. More critically, total prohibition deprives engineering students of the opportunity to develop quality assurance and error-detection skills—learning how to audit, verify, and correct automated outputs, which is a vital competency in modern engineering practice.

Measurable Goals

  • Reframe AI as a Conceptual Tutor: Shift student behavior from using LLMs as automated answer keys to leveraging them as interactive, low-stakes tutors during the initial discovery phase.

  • Establish Clear Performance Boundaries: Train students to delineate between support-assisted conceptual exploration ("anything goes to learn") and unassisted paper-and-pencil mastery ("closed-notes to finish").

  • Develop Engineering QA/QC Verification Skills: Cultivate rigorous error-detection habits by requiring students to audit, redline, and correct AI-generated solutions containing mathematical hallucinations or flawed derivations.

Course Solutions & Policies

"Anything Goes to Learn, Closed-Notes to Finish" Policy

The course establishes a realistic, transparent boundary for digital tool usage:

  • Discovery Phase: During initial homework problem sets on Knewton Alta, students are permitted to use any support tool available—including AI models, peer study groups, video tutorials, and office hours—to build geometric or algebraic intuition.
  • Mastery Phase: Before closing out an assignment or sitting for exams, students are required to re-attempt problem sets unassisted on paper, with closed notes, to ensure true procedural and conceptual competence.
  • Exam Integrity: Exams remain strictly closed-book, closed-notes, paper-and-pencil assessments where only approved scientific or graphing calculators are permitted.

Bi-Weekly Redline Audits

In professional engineering, calculations and automated simulations are subjected to strict Quality Assurance / Quality Control (QA/QC) peer review. Bi-weekly Redline Audits replicate this professional workflow:

  • Audit Workflow: Students evaluate calculation sets containing intentional errors or common LLM hallucinations. Students must manually evaluate, redline, and correct these solutions using standard engineering markup conventions.
  • Critical Verification: By comparing human derivations against AI outputs, students learn firsthand that generative tools frequently hallucinate mathematical steps, reinforcing the professional duty never to trust automated output blindly.

Grade Validation via Weight Structures & Safety Nets

To support process and effort over pure numerical correctness, course grading weights are structured to reduce panic-driven cheating:

  • Effort-Based Safety Nets: Homework counts for 20% of the course grade, protected by the 4-Hour Effort Rule and a 7-day rolling grace period to absorb schedule friction.
  • Process & Communication: Technical Communication and Quality Control (Redline Audits and Memos) account for 15% of the total grade, emphasizing documentation quality and error auditing.
  • Retroactive Grade Replacement: Initial exam stumbles do not permanently decay student grades; completing exam corrections and maintaining an 80% global homework attempt rate allows the final exam score to replace the lowest midterm grade.

Student Avatar Context: Maya

Maya is a high-achieving freshman who relied on memorization and formula-plugging in high school. When faced with multi-step engineering modeling in calculus, she feels overwhelmed and considers using AI solvers to generate homework answers. Our syllabus explicitly addresses this temptation: rather than banning AI with unworkable threats, it establishes an "anything goes to learn, closed-notes to finish" study strategy. Maya uses AI to explain geometric intuitions during homework, but practices solving final problems unassisted. During bi-weekly Redline Audits, she audits AI-generated solutions, catches mathematical hallucinated errors, and builds the quality control skills required of professional engineers.

Related Network Nodes