Academic AI Usage & Assessment Integrity Matrix

Disciplinary Vulnerability Landscape

X-Axis Variable

Y-Axis Variable

Filter Discipline

Disciplinary Impact Analysis

As highlighted in recent academic analysis by The Economist, Computer Science and Business departments experience disproportionately high AI adoption rates among students, driven by LLMs' proficiency in synthetic code generation and structured prose.

Humanities face distinct challenges with nuanced text generation, leading to an arms race between detection tools and student evasion tactics, while educators struggle to redefine "original work."

Separate authored assessment scores from misconduct evidence

Your browser does not support video.

Read the explanation

This integrity matrix contains five source-authored discipline records. Their AI usage values are eighty-eight, seventy-six, sixty-four, fifty-two and seventy, averaging seventy. Those numbers and claims about students are not independently verified survey findings. Four saved integrity-risk values are strictly greater than sixty; that is a catalog count, not a measured misconduct rate. A separate assessment formula halves the selected field risk, adds point four times take-home weight and subtracts point three times oral-defense weight, then rounds and clamps between five and ninety-nine. With the default computer-science record eighty-five, take-home fifty and oral twenty, the raw score is fifty-six point five, rounded to fifty-seven. The bars grow from the forty-two point five field contribution to the complete fifty-seven score on a shared eight-pixel scale. The original chart depends on D3, unavailable in this offline saved page, so initialization does not build the matrix or discipline-selection list. We explain the exact formula without claiming those missing views rendered. The native sliders have listeners installed before chart initialization, so their numerical simulator can still update even when the external visualization library is missing. Holding oral weight at twenty and moving take-home from zero to one hundred shifts the computer-science score from thirty-seven to seventy-seven. The forty-point change follows the point four coefficient; it is not an observed effect of take-home exams. Holding take-home at one hundred and raising oral from twenty to one hundred subtracts twenty-four score points, ending at fifty-three. Caption thresholds are strict: above seventy is critical vulnerability, below forty is resilient, and forty through seventy is moderate. Rounding happens before this comparison. With oral zero and computer-science risk eighty-five, take-home seventy produces seventy-one after rounding and enters the critical branch. Take-home sixty-eight produces seventy and stays moderate. These are rules for interface labels, not disciplinary judgments about a person. Another displayed percentage uses an independent formula: saved unauthorized usage times one plus take-home minus oral divided by one hundred. The selected computer-science record uses forty-two. Default fifty minus twenty gives a factor one point three and fifty-five percent after rounding. Equal take-home and oral weights always return forty-two. At one hundred take-home and zero oral the factor is two, giving eighty-four; at zero take-home and one hundred oral it is zero, giving zero. These outputs are arithmetic estimates from an authored baseline, not detected misconduct or a probability that any student cheated. Filtering the chart changes visibility and axes change plotting coordinates; neither changes the selected field automatically. The JSON export always includes the complete five-record dataset, current controls and selected computer-science record. Local tests use actual downloads and native sliders, and separately record the missing D3 matrix and field-list limitation. Source summaries, detection claims and adoption rates remain unverified.

Academic AI Integrity Matrix Mechanics & Evaluation

How does this simulator model academic AI integrity risk, and what mathematical assumptions govern the calculated vulnerability scores?

This interactive matrix visualizes illustrative risk profiles across five academic fields (Computer Science, Business & Finance, Humanities & Social Sciences, Natural Sciences, and Law & Policy). When exploring the disciplinary scatter plot or altering coursework structure in the Integrity Stress Simulator, the tool evaluates vulnerability using stylized heuristic linear equations rather than empirical psychometric data. Calculated vulnerability weights take-home assignments positively and oral defenses negatively against a baseline field vulnerability score.

All underlying metrics—including the 88% Computer Science adoption rate, detection evasion scores, and the unauthorized assistance formula—are static heuristic parameters embedded in the JavaScript data fixtures rather than verified institution-wide empirical research. In the stress simulator, calculated assessment vulnerability is derived from an arbitrary linear combination (0.5 × baseline + 0.4 × takeHome − 0.3 × oral), clamped between 5 and 99. These stylized values illustrate structural assessment trade-offs but do not predict real-world classroom cheating rates.

Try a worked example

In the Disciplinary Vulnerability Landscape panel, select 'Humanities' from the Filter Discipline dropdown (#filter-discipline). The matrix updates to isolate the Humanities & Social Sciences node, highlighting its preset 64% adoption rate and 68 baseline vulnerability index while filtering out the remaining four disciplines.

Integrity Stress Simulator Formulation

The client-side calculation evaluates Assessment Vulnerability as calculatedRisk = round((field.integrityRisk × 0.5) + (takeHome × 0.4) − (oral × 0.3)), clamped between minimum 5 and maximum 99. A score above 70 triggers a 'Critical Integrity Vulnerability' rating, while scores below 40 are classified as 'Resilient Assessment Structure'.

The estimated unauthorized assistance rate adjusts the discipline's baseline unauthorized percentage via the factor (1 + (takeHome − oral) / 100). This produces a linear penalty for take-home deliverables and a linear discount for proctored or interactive evaluations.

Disciplinary Landscape Mapping

The D3.js scatter plot maps each field across selectable attributes, including adoption rate, unauthorized percentage, code-to-text dependency ratio, and educator adaptation velocity. The axes dynamically rescale using a 20% lower buffer and 10% upper margin around the dataset extremes.