The Paper

Research Paper

Reimagining Education
in the Age of Generative AI

Abstract

Research shows that heavy reliance on generative AI can significantly diminish independent reasoning, writing, and problem-solving skills. Koan addresses this challenge by integrating a Socratic AI teaching assistant that guides students through reflective problem-solving rather than providing answers.

September 14, 2025 · 15 min read · Koan Research Team

Download PDFRead Online
I

The Crisis

AI has fundamentally reshaped education.
Schools are not equipped to respond.

Generative AI tools like ChatGPT, Claude, and Microsoft Copilot are now embedded in how students research, write, and process information. Yet research shows that the manner in which students interact with these tools is proving to be problematic.

AI_ADOPTION

92%

of UK students use AI in their studies

up from 66% in 2024

// Freeman & HEPI, 2025

POLICY_GAP

31%

of US public schools have AI policies

leaving students without guidance

// NCES, 2025

SKILL_DEFICIT

82%

of students lack skills for responsible AI use

despite regular usage

// Dean, 2025

Research Finding

“Over a four-month period, LLM users consistently underperformed at neural, linguistic, and behavioral levels.”

- Maes et al., 2025

Problem Statement

“Modern education incentivizes polished outputs, not the messy, nonlinear process of learning. This has created a widening skills gap that erodes students’ critical thinking and leaves them unprepared for real-world problem solving.”

II

Theory

Understanding the Problem
Through Three Lenses

The issue is not that students are using AI tools, but rather how they are using them. Our response must address cognitive, pedagogical, and institutional gaps.

cognitivepedagogicinstitutionalthe problem

Cognitive

Understanding How We Learn

Cognitive Load Theory

Learners can only handle a limited amount of information at once. AI tools, when used appropriately, can alleviate ‘extraneous load’ by summarizing texts, organizing notes, or assisting with language translation.

(Sweller, 1988)

The Risk of Surface Learning

When AI is used to bypass ‘germane load’, the cognitive effort required to process, reflect on, and internalize new knowledge, students produce acceptable outputs without engaging in the cognitive processes necessary for long-term retention.

Metacognition

Students with low metacognitive awareness are more likely to rely on shortcuts. Without guided frameworks for reflective AI use, they risk reinforcing cognitive dependency and weakening critical thinking skills.

(Flavell, 1979; Zimmerman, 2002)

Pedagogic

How We Structure Learning

The Output Problem

Standardized curricula, high-stakes testing, and grade-driven assessment models promote a form of learning that prioritizes outputs over process, incentivizing students to use generative tools adversely.

(Au, 2007; Kohn, 2004)

Against Surveillance

Digital pedagogy scholars argue that emphasis on surveillance and punishment fails to recognize students as collaborators in their own learning. Trust-based, process-centered environments treat AI as a tool for dialogue.

(Morris & Stommel, 2018)

Community of Inquiry Framework

The CoI framework highlights three dimensions of meaningful learning: cognitive presence (making sense through questioning), social presence (authentic expression), and teaching presence (guiding the experience).

(Garrison, Anderson, & Archer, 2000)

Institutional

The Policy Gap

Inconsistent Guidance

Educational institutions are being forced to navigate students’ rapid adoption of AI technologies with inconsistent and slow-paced development of policies to manage their use in academic settings.

Failure of Current Approaches

Current approaches, often inconsistent and restrictive, leave learners without guidance on ethical use, digital literacy, or productive collaboration with AI.

The Stakes

Public education exists to prepare citizens for economic participation, democratic engagement, and social contribution. When graduates lack critical thinking or the ability to work effectively with emerging technologies, these systems fail in their fundamental mission.

Key Insight

“Our current education systems have a damaged incentive structure. The product of polished assignments has become more important than the practice of learning.”

III

The Solution

Socratic AI for Public
Cognitive Development

At the heart of our solution lies Aidan, a Socratic AI teaching assistant built into an AI-integrated Learning Management System. Unlike conventional generative AI, Aidan is designed not to provide direct answers, but to guide students through questions.

Example Interaction

If a student asks for an essay on inflation, Aidan does not supply a draft. It breaks the task into decisions and prompts:

“What have you noticed about prices in your community this year?”

“Which forces could explain those changes?”

“How might public policy shape what families can afford?”

“How would you organize an introduction, supporting paragraphs, and a conclusion so your claim is clear?”

Meet Aidan

Socratic Method

Aidan guides students through questions and reflective processes, never providing direct answers.

Dynamic Knowledge Tree

A pedagogical frame that adapts to each learner while maintaining consistent academic standards.

Process Checkpoints

Targeted hints and formative feedback loops keep students moving while preserving academic integrity.

Ethical Guardrails

Teachers control how directive or open the tutor should be through adjustable guidance levels.

The Koan Platform

Writing Workspace

Students draft, cite, revise, and submit in one place. Rich text, equations, figures, citations, and version history included.

Interactive Assignments

Flowchart builders, graph tasks, simulations, data tables, rationale fields, and reflection checkpoints, all within guided flows.

Teacher Dashboards

Real-time visibility into discourse quality, progress toward outcomes, and where students stall.

Integrity Tools

Similarity views, citation checks, and reflection compliance, transparency over detection.

Research Support

“Early research shows that general purpose AI used in isolation can harm learning. When placed in a safeguarded human-plus-AI workflow with clear teaching presence, it can increase engagement, comprehension, and critical thinking.”

- Kestin et al., 2025; Bastani et al., 2024; Shukla & Pandey, 2025
IV

The Evidence

Internal Survey Findings
Validating the Need

We conducted structured surveys targeting educators and students across multiple countries to investigate the challenges in an increasingly AI-driven learning environment.

92

Educators surveyed

18

Students surveyed

7

Countries represented

Student Findings

n = 18 (12 undergraduates, 5 secondary, 1 graduate)

use AI tools88%

ChatGPT, Claude, Gemini

use AI for assignments100%

at least sometimes

for writing/rephrasing83%

most common use case

concerned about skills89%

cognitive erosion worry

“Peers are using AI and completing assignments very quickly. If I don’t use AI, I feel like I am lagging behind.”

- Undergraduate student from Evanston, Illinois, USA

Educator Findings

n = 92 (primarily high school teachers and undergraduate lecturers)

use AI tools75%

planning, grading, communication

encounter AI problems95%

plagiarism, over-reliance

changed assessments69%

oral, handwritten, follow-ups

have policy authority78%

full or shared control

Key finding: 95% of educators report encountering at least one problem with student AI usage, with over-reliance replacing learning effort being the most frequently cited concern.

Interest in Koan

85%

Educators interested

95% CI: 77.5% - 92.5%

94%

Students interested

95% CI: 82.7% - 100%

Both surveys confirmed the global scale of the issue, with responses from Brazil, the United States, Canada, India, Ecuador, and Australia.

Conclusion

“Students are eager for guidance to use AI responsibly rather than simply banning it, signaling a clear need for a structured solution.”

It’s time to reimagine education for the AI era.

AI itself is not the problem. The true challenge is our outdated approach to learning, we overvalue polished outputs and undervalue the messy, reflective process that builds deep understanding.

Download PDFMeet Our Team

Why Koan?

Koan is built for students, by students. We are a dedicated team of university students from four continents with expertise in AI, research, and product development. We have completed ideation and preliminary market validation, demonstrating both the feasibility and demand for Koan.

For academic citations

Koan Research Team. (2025). Reimagining Education in the Age of Generative AI. Koan Learning Systems.