The word "delve" appeared in roughly 200 academic papers in 2020. By 2023, it showed up in over 800,000. This linguistic fingerprint of large language models reveals how deeply AI has already penetrated academic writing, regardless of institutional policy.
At the University of Zurich, where I studied Artificial Intelligence, ChatGPT arrived mid-curriculum. The institutional response was largely prohibitive. The problem with bans is straightforward: they are unenforceable and inequitable. Students who use ChatGPT despite the rules (for comprehension checks, code reviews, draft editing) gain a measurable advantage over those who comply. The ban does not eliminate AI use; it ensures that the benefit accrues unevenly and without transparency.
If a language model can reproduce the factual content of a blockchain course in seconds, the course needs to justify what it teaches beyond facts. The pedagogical challenge is real: curricula built around knowledge transmission become redundant when that knowledge is freely available through an API. What remains valuable is the capacity for critical evaluation, novel problem formulation, and the ability to identify when an AI-generated answer is subtly wrong. These are skills that require a different kind of teaching.
The situation in primary education is structurally different. A university student using ChatGPT to draft a literature review is automating a task they already know how to do. A primary school student using it to write an essay on Kafka's "The Metamorphosis" may never develop the underlying analytical skills. At this level, the product is not the point. The cognitive process of struggling with interpretation is. AI shortcuts that process entirely.
Teaching is already an intensely demanding profession: pedagogy, mentorship, administration, and emotional support compressed into insufficient hours. AI can relieve some of that pressure, particularly in grading and feedback. Automated first-pass assessment of written work could free hours per week for the interpersonal work that no model can replicate. The shift is from knowledge transmitter to learning facilitator (a role that requires more skill, not less).
Traditional exam-based assessment assumes that the student produced the work independently. That assumption is now unfalsifiable for take-home assignments. A teacher who has observed a student over months or years has a far richer basis for evaluation than any single test. Qualitative, continuous assessment (regular conversations, portfolio reviews, observed problem-solving) is more robust against AI-assisted cheating and more informative about actual capability. AI itself can contribute here by providing real-time feedback on exercises, allowing students to iterate before a human evaluator ever sees the work.
The most promising near-term application is AI as an always-available tutor. Current systems can explain a concept repeatedly from different angles, adapt to a student's pace, and provide immediate feedback rather than the weeks-long delay of traditional grading cycles. The constraint is that this works best when the student already has enough foundational knowledge to evaluate the AI's responses critically. This brings the argument back to the importance of building those foundations without shortcuts.
Students use tools like ChatGPT for comprehension questions, code reviews, and text improvement. Institutions that respond with blanket bans create a digital divide: students who use the tools despite prohibitions gain an advantage over those who comply. The surge of certain AI-favored phrases in academic papers such as 'delve,' which jumped from 200 to 800,000 occurrences between 2020 and 2023 shows that AI is already deeply embedded in academic writing.
In primary education, the learning process matters more than the final product. When students use AI to generate text interpretations, they risk never developing fundamental skills in critical thinking and independent analysis. The challenge is integrating AI without undermining the cognitive development that comes from struggling with a problem.
AI can take over time-intensive tasks like grading, freeing teachers to focus on pedagogical and interpersonal aspects of their work. The role shifts from knowledge transmitter toward learning facilitator and mentor, emphasizing personal development over information delivery.
Traditional grading systems and standardized tests are increasingly supplemented or replaced by qualitative evaluations and personal conversations. The trend is toward continuous, individualized assessment by teachers who observe students over extended periods. AI can support this by providing immediate feedback and tracking individual learning progress.
AI can function as an interactive learning tool that provides immediate feedback and adapts to individual learning pace. It can revisit concepts from different angles until they are understood, enabling personalized learning and continuous support outside the classroom.
Avoiding digital inequality requires fair and equal access to AI tools, clear policies for AI use, adequate technical infrastructure in schools, and training for both teachers and students. Blanket bans should be avoided, as they typically lead to hidden and unequal usage patterns.
.
Copyright 2026 - Joel P. Barmettler