Learning Outcomes of AI in Education

🎯 Featured Research: The DRIVE Framework

Assessing Students' DRIVE: A Framework to Evaluate Learning Through Interactions with Generative AI

Oliveira, M., Zednik, C., Bombaerts, G., Sadowski, B., & Conijn, R. (2025)

Groundbreaking research from TU Eindhoven introduces a practical framework for assessing student learning through their interactions with GenAI. This moves beyond concerns about "cheating" to recognize and evaluate productive AI collaboration in authentic classroom settings.

70
Essays Analyzed
1,450
AI Interactions Coded
r=0.54
Correlation with Traditional Scores
92%
Students Using GenAI (2025)

Key Finding: Process-focused assessment of GenAI interactions correlates strongly with traditional essay quality (r=0.54, p<.001), validating this approach for capturing student learning.

→ View full research and materials

What is the DRIVE Framework?

DRIVE stands for Directive Reasoning Interaction and Visible Expertise – two complementary dimensions for evaluating student learning through their GenAI interactions.

DRI

Directive Reasoning Interaction

Question: How actively and purposefully does the student steer the AI?

High DRI indicates:

  • Taking a leading role in the dialogue
  • Critically questioning AI outputs
  • Using personal reasoning to guide interaction
  • Maintaining "human-in-command" approach

Aligns with the ICAP framework (Chi & Wylie, 2014) and principles of self-determined learning (heutagogy)

VE

Visible Expertise

Question: Is the student's knowledge visible in their prompts?

High VE demonstrates:

  • Introducing specific course concepts
  • Applying unique insights and ideas
  • Critically evaluating AI outputs
  • Building on AI with domain knowledge

Resonates with "Making Thinking Visible" (Ritchhart, 2011) – for assessment, thinking must be observable

Key Research Findings

Analysis of 70 essays and 1,450 student-GenAI interactions revealed three distinct profiles:

High Essay Scores

"Targeted Improvement Partnership"

Students focused on systematic text refinement (13.4% of interactions), critical evaluation of AI content, and strategic conceptual integration.

High Interaction Scores

"Collaborative Intellectual Partnership"

Students centered on idea co-development, conceptual clarity (1.3%), and critical engagement with AI outputs.

Below-Average Scores

"Basic Retrieval / Passive Delegation"

Students primarily used AI for simple information gathering (5.6%) or passive task specification without strategic engagement.

Critical Insight

  • Assessment focus shapes student behavior: Traditional output-focused assessment reinforced text optimization strategies, while process-focused evaluation rewarded exploratory intellectual partnership with AI.
  • Strong validation: The correlation between traditional essay scores and GenAI interaction quality (r=0.54) demonstrates that analyzing the interaction process provides meaningful insights into student learning.
  • Three interaction categories emerged: Writing (41.3%), Content (28.7%), and Argument (22.3%) – providing a practical taxonomy for educators.

Implementing Process-Focused Assessment

Evidence-based recommendations for educators:

1. Dual Assessment Approach

Combine traditional essay assessment (captures text quality and conceptual integration) with interaction log evaluation (reveals critical thinking and collaborative strategies). Together, they provide a comprehensive view of student competencies.

2. Grading Rubric for GenAI Interactions

Consider evaluating three dimensions (adapted from TU/e courses):

3. Design Principles

4. Student Preparation

As implemented at TU/e, include instruction on:

GenAI Assessment: Global Context

Rapid Adoption Rates

The HEPI Student Generative AI Survey 2025 found that GenAI use among undergraduate students jumped from 66% in 2024 to 92% in 2025, with only 12% reporting they have not used GenAI for assessments (down from 47% the previous year). This unprecedented rate of adoption demonstrates that GenAI is now deeply embedded in higher education.

Assessment Transformation

Recent research from the Association of Pacific Rim Universities (2025) emphasizes that assessment reform is a critical priority, with institutions worldwide exploring authentic assessments that leverage GenAI as a learning tool rather than viewing it solely as a threat.

Authentic Assessment Response

As Kofinas et al. (2025) note, managing GenAI's impact requires a paradigm shift in assessment philosophy – intelligent assessment design can convert GenAI into a tool for deeper learning by:

Resources and Further Reading

TU/e Research

Assessment & GenAI in Higher Education

Student-AI Interaction Patterns

Competency Frameworks

Beyond AI: The New Essential Skills Reshaping Student Success (Research by Manuel Barbosa de Oliveira)

In today's rapidly evolving educational landscape, students find themselves at a fascinating crossroads. The traditional skills that once formed the bedrock of education - memorizing facts, producing standard reports, completing routine tasks - are giving way to something far more profound. Instead, students must become fluent in a new language: the language of AI literacy. This means not just understanding how to use AI tools, but truly grasping their potential and limitations, much like a craftsperson understanding their tools while recognizing when to rely on human touch instead.

But this is just the beginning of their journey. The heart of modern education lies in developing what AI cannot easily replicate: higher-order thinking skills that define human intelligence. Students are now challenged to think critically, solve complex problems, and create innovative solutions in ways that machines cannot. It's like developing a mental toolkit where each tool - from analytical thinking to creative problem-solving - serves a unique purpose that complements, rather than competes with, artificial intelligence.

Perhaps most crucially, adaptability has become the cornerstone of student success. In this new educational frontier, students must learn to be agile thinkers, capable of navigating an ever-changing technological landscape. They need to develop what we might call "intellectual flexibility" - the ability to not just work alongside AI, but to think deeply about how they think, understanding their own learning processes and explaining their reasoning. This represents a fundamental shift from being passive receivers of knowledge to becoming active architects of their understanding, preparing them for a future where the ability to adapt and think creatively is more valuable than ever.

For the full report, click here

How Different Learning Objectives affect the Design of Educational Chatbots

As artificial intelligence continues to transform education, understanding the different types of chatbots and their specific applications becomes crucial for effective implementation in educational settings. Defining distinct chatbot types helps institutions and educators make informed decisions about which tools best suit their specific educational needs and objectives.

The classification of educational chatbots serves multiple purposes:

Evaluation and Impact

Different types of chatbots require distinct evaluation approaches because their goals and success metrics vary significantly. For instance, an information retrieval chatbot's success might be measured by response accuracy and speed, while a creativity-stimulating chatbot's value lies in its ability to inspire novel thinking and engagement.

Expected Benefits

For students, these specialized chatbots can offer:

For teachers, the benefits include:

Types of Educational Chatbots and Their Applications

Chatbot Type Objective Teaching Activities That Can Be Replaced/Substituted Accuracy Flexibility/Functionality User-Centric Evaluation Ongoing Pilots
Information Retrieval Provide quick access to course information, such as exam dates, policies, or schedules. Answering routine student inquiries about schedules, deadlines, policies, and course logistics. High accuracy is critical; errors in factual details can lead to confusion. Should support multiple courses and adapt to various course structures. Users value efficiency and reduced search time. Interface should be intuitive. Alexandria.cx, Tilburg.ai
Content Inquiry Clarify and explain course concepts. Answering repetitive questions on fundamental course concepts and providing general explanations. Responses must align closely with course content; hallucination minimized. Allows adaptation to different educational theories and inclusion of course-specific data. Users should trust the chatbot for reliable course explanations; satisfaction tied to learning gains. Tilburg.ai, Alexandria.cx
Discussion Enhancement Encourage critical thinking and argumentation skills. Initiating structured discussions, suggesting prompts, and moderating debates. Accuracy in prompts is important but creativity and engagement take priority. Should adapt to various collaborative contexts and integrate with discussion formats. Users value interactive, engaging discussions that enhance critical thinking and collaboration skills.
Creativity-Stimulating Foster creativity through brainstorming, scenario simulation, and idea expansion. Facilitating brainstorming sessions, generating idea prompts, and simulating creative scenarios. Accuracy in factual inputs is secondary; primary focus is on stimulating divergent and innovative thinking. Tools should allow personalized prompts, fine-tuning, and adaptability to diverse creative tasks. Users value perceived autonomy and the chatbot's ability to inspire new ideas and maintain engagement.
Assessment Support Generate and assess questions or quizzes. Creating quizzes, grading objective assessments, and providing automated feedback. High accuracy needed to align with learning objectives and ensure fairness in evaluation. Allows customization based on instructional goals and assessment frameworks. Users value fairness, accuracy, and reduced time investment in creating assessments.
Learning Analytics Track and analyze learning outcomes, such as participation, grades, or critical thinking progress. Collecting and analyzing student performance data, identifying patterns, and generating reports. Data accuracy is paramount for reliable analytics. Must integrate seamlessly with various learning management systems and course tools. Users value actionable insights into their learning progress; perceived trust and privacy in data use are essential.

Contact & Collaboration

For questions about implementing the DRIVE framework or collaborating on AI in education research: