How Not to Become a Cheating Society

1 | THE PRODUCTIVITY PARADOX

Use AI Aggressively Where Capability Already Exists - Protect Capability While It Is Being Built

The argument for protecting human thinking is not an anti-technology argument. AI is already improving productivity in knowledge work. Microsoft Research reported that in a large randomized field experiment, workers with access to generative AI spent materially less time on email and completed documents faster. [1]

THE POLICY DISTINCTION AI + developed judgment can create leverage. AI replacing the development of judgment creates dependency risk.

2 | TEN WAYS TO ADAPT - PART I

1-3: Measure Process, Preserve Authentic Performance, Require Explanation

1. 1

   ASSESS PROCESS, NOT ONLY PRODUCT

   Build evidence of learning into the path: planning notes, drafts, checkpoints, source trails, revisions, reflections, and explanation of how decisions changed. A polished final product should not be the only proof of competence.

2. 2

   PRESERVE MEANINGFUL UNAIDED DEMONSTRATIONS

   Where society truly needs assurance of independent competence, keep some supervised, oral, practical, handwritten, live, or otherwise authentic demonstrations in which AI assistance is intentionally limited or absent.

3. 3

   REQUIRE STUDENTS AND PROFESSIONALS TO DEFEND IMPORTANT WORK

   Short oral defenses, live Q&A, demonstrations, and follow-up problems can reveal ownership and understanding far better than simply asking whether AI was used.

DESIGN PRINCIPLE If a credential matters, build at least one credible pathway for the human being to demonstrate the capability directly.

3 | TEN WAYS TO ADAPT - PART II

4-6: Define AI Use, Teach AI Literacy, Redesign the Work

1. 4

   REPLACE VAGUE "DO NOT CHEAT" RULES WITH AI-SPECIFIC EXPECTATIONS

   Define prohibited, limited, disclosed, and encouraged uses. A Green/Yellow/Red framework can help students understand the difference between feedback, assistance, material shaping, and substitution.

2. 5

   TEACH AI LITERACY AS A CORE HUMAN SKILL

   Students should learn how models can be wrong, biased, incomplete, persuasive without evidence, and dependent on prompts and context. UNESCO's framework calls for critical judgment, human agency, ethics, and accountability. [2]

3. 6

   REDESIGN ASSIGNMENTS AROUND JUDGMENT, CONTEXT, AND ITERATION

   Ask for local evidence, personal reasoning, oral explanation, comparison of sources, reflection on failed approaches, critique of AI output, and revision. Design work that values what the learner notices and decides - not only what text appears at the end.

4 | TEN WAYS TO ADAPT - PART III

7-10: Do Not Outsource Integrity to Detection, Train Adults, Protect Credentials, Keep Humans Accountable

1. 7

   DO NOT MAKE AI DETECTION THE ENTIRE INTEGRITY SYSTEM

   Detection tools may offer signals, but institutions still need fair process and positive evidence of learning. Common Sense Media reported demographic differences in teens saying teachers falsely flagged work as AI-generated, illustrating why accusation systems require caution. [3]

2. 8

   TRAIN EDUCATORS AND LEADERS

   Teachers, administrators, counselors, tutors, and supervisors need practical understanding of AI capabilities, limitations, privacy, assessment design, and appropriate disclosure. A rule nobody understands will not become a culture.

3. 9

   PROTECT HIGH-TRUST CREDENTIALS

   Medicine, engineering, law, accounting, education, and other high-stakes fields should identify competencies that must remain demonstrably human. Accreditation and licensing bodies should update assessment models rather than assuming old take-home formats still prove what they once proved.

4. 10

   PRESERVE HUMAN ACCOUNTABILITY

   Organizations may use AI to advise, draft, predict, summarize, or recommend. A responsible human still needs authority, understanding, and accountability for consequential decisions. NIST's AI risk-management work provides a useful governance model for managing GenAI risks across organizations. [4]

THE GOVERNANCE QUESTION When AI contributes to a consequential outcome, who is responsible for checking it, challenging it, approving it, and owning the result? If the answer is nobody, the system is not ready.

5 | WHO SHOULD DO WHAT

Different Institutions Have Different Responsibilities

K-12 SCHOOLS Teach age-appropriate AI literacy; define classroom rules; preserve core skill formation; support educators; communicate with families; protect student privacy.

COLLEGES & UNIVERSITIES Publish course-level expectations; redesign assessment; preserve authentic demonstrations; teach professional AI use; create fair integrity processes; support faculty.

GOVERNMENT Fund independent research and educator training; establish procurement/privacy expectations; support public AI literacy; coordinate standards for high-stakes sectors; avoid one-size-fits-all classroom mandates.

EMPLOYERS Use AI for productivity while testing real capability in hiring, promotion, and high-stakes roles; train workers to verify AI; maintain clear accountability.

ACCREDITORS / LICENSING BODIES Define what knowledge and judgment must remain independently demonstrable; modernize exams and practical assessments; protect public trust in credentials.

AI PROVIDERS Design education modes that can support hints, questioning, citations, transparency, age-appropriate safeguards, privacy, and learner agency rather than default completion.

6 | WHAT BETTER ASSESSMENT CAN LOOK LIKE

Move From "Can You Produce This?" to "Can You Understand, Defend, and Use This?"

ASSESSMENT SHOULD EVOLVE WITH THE TOOL If AI can legitimately do part of the real-world job, education should teach students how to supervise that part. But it should also verify the human capabilities society still expects the graduate to possess.

7 | THE HUMAN-JUDGMENT PROBLEM

The More Powerful AI Becomes, the More Valuable Independent Judgment Becomes

A future filled with capable AI does not remove the need for critical thinking. It changes where critical thinking happens. Microsoft Research found that GenAI shifts knowledge-work critical thinking toward verification, integration, and task stewardship - while higher confidence in GenAI was associated with less critical-thinking effort. [5]

THE PURPOSE OF EDUCATION IN THE AI AGE Machines will increasingly produce answers. Education must produce human beings capable of deciding which answers deserve to be trusted, challenged, rejected, improved, or acted upon.

8 | A 90-DAY INSTITUTIONAL START

Do Not Wait for Perfect Rules Before Building Better Practice

DAYS 1-30: DEFINE Inventory where AI is already being used. Identify high-stakes competencies. Publish interim definitions of prohibited, disclosed, and encouraged use. Establish a fair process for suspected misuse.

DAYS 31-60: REDESIGN Choose a small number of courses, assessments, hiring tasks, or credential checkpoints. Add process evidence, oral defense, supervised demonstration, or verification tasks.

DAYS 61-90: BUILD CAPABILITY Train staff. Teach AI literacy. Collect feedback. Measure whether the new system produces better evidence of learning and competence. Revise the rules from evidence rather than anxiety.