Joshua Holstein

Background

Current Generative AI systems are often built to maximize productivity, measured by the speed of supported work processes. However, this focus frequently leads to cognitive offloading of the human decision-maker—where they defer the thinking to the AI system and unreflectively rely on its outputs.

To counter this, we must pivot the design of generative AI systems toward Tools for Thought (TfT). TfTs prioritize the process of reasoning, memory, and critical thinking over the mere completion of a task. The goal is therefore not to provide final advice, but to act as a reflection partner. The "sweet spot" for cognitive augmentation occurs when design and usage strategies force the user to bridge gaps in their own understanding through active reasoning and reflection. We are seeking two Master’s students to explore this from two interrelated angles: Strategy (the "how") and Outcome (the "what").

Thesis A: Strategies for Thoughtful Cognition

Research Goal

This thesis focuses on the theoretical identification and systematization of design and usage strategies that prevent cognitive offloading. The goal is to move beyond "one-off" designs toward a transferable framework of GenAI strategies that spark active reasoning and critical thinking.

Research Objectives:

• Systematic Literature Review: Identify existing design patterns and theories from HCI, Learning Sciences, and Cognitive Psychology (e.g., Desirable Difficulties, Scaffolding, Productive Failure) that can be applied to GenAI.
• Strategy Taxonomy: Systematize identified examples into tangible Design Strategies (how the system is built) and Usage Strategies (how the user interacts) to support forward-reasoning.
• Conceptual Framework: Develop a framework that maps specific AI roles (e.g., Socratic Tutor, Facilitator, Provocateur) to the cognitive functions they are intended to augment.
• Prototypical Validation: Implement a focused "proof-of-concept" based on the developed framework to demonstrate how a specific strategy shifts the user from passive consumer to active thinker.

Your Profile

• Strong interest in HCI Theory and Cognitive Science.
• Ability to synthesize complex literature into structured taxonomies or frameworks.
• Solid experience in Python/Web-prototyping to build demonstrative "Tools for Thought".

Thesis B: Evaluating Cognitive Augmentation

Research Goal

This thesis focuses on defining and validating metrics for the "invisible" outcomes of using a TfT. The focus is on how to measure the evolution of a user’s Domain Mental Model and the quality of their internal reasoning process.

Research Objectives:

• Metric Synthesis: Conduct a theoretical analysis of how cognitive gain and reflection are measured in related fields (e.g., educational assessment, sensemaking) and adapt them for GenAI contexts.
• Domain Mental Model Elicitation: Develop and test methods to capture the "before and after" state of a user’s mental model when using a TfT (e.g., through concept mapping or structured reflection prompts).
• Process-Outcome Mapping: Analyze the relationship between Intermediary Artifacts (the notes and "failed" attempts made during a task) and the eventual depth of the user's domain understanding.

Your Profile

• Interest in Empirical Research Methods and Metacognition.
• Experience in (or willingness to learn) qualitative and quantitative data analysis (e.g., coding Think-Aloud protocols or analyzing interaction logs).
• You are driven by the question "How do we prove this tool actually changed the way someone thinks?"

Please send your current transcript of records, a short CV, and a brief motivation (3–4 sentences) to: Joshua.Holstein@kit.edu