TokenAI

The TokenAI Concept

Integrating AI into high-stakes assessment requires a balance of security, ethics, and critical thinking. Our framework treats the LLM as a finite, monitored resource.

mindmap root("Constrained/restricted AI based on a metric (token) in a pedagogical situation (exam/lab)") Inequalities Knowledge of tools Training Critical thinking Analysis of biases Access to the resource Tensions Among students: knowledge of the tools Between teaching staff: uses for/against IT department for management Dependence Limited resources Instrumented tools Social impact Sovereignty LLM specific to the institution

Digital Sovereignty

Institutions deploy centralized, secure AI platforms internally. This avoids dependency on external tech giants and ensures absolute data privacy for student sessions.

Token-Based Assistance

Usage is limited via a token system. Professors set a specific budget, forcing students to strategically decide when a hint or verification is truly necessary.

Environmental Impact

Real-time carbon tracking educates students about the physical cost of AI. Minimal usage is rewarded with grade bonuses, encouraging mindful interactions.

Security Architecture

No File Uploads: Students must formulate their own prompts, requiring active thinking.
Aggregated Models: Teachers select models per specific subject needs.
Locked Kiosk: Operates through a secure browser (SEB) to block external internet access.

Implementation

Our model ensures that students lacking fundamental understanding still struggle. By restricting tokens (like limiting a cheat sheet), we prevent students from delegating all reasoning to the algorithm.

Teachers gain flexibility: a large budget might test prompt engineering and critical analysis, while a small budget tests basic troubleshooting or knowledge retrieval.

Post-Exam Analytics

After the session, all prompts are accessible to teachers. This data allows for a deep analysis of student reasoning and their interactions with AI.

Analyze request formulation patterns for each question

Adjust future exams based on LLM output trends

Enhance integration of AI in lectures

Teacher Dashboard