Designing AI-Assisted Interaction Concepts for Collaborative Scientific Work
AI Assisted Electronic Laboratory Notebook
My Role
Student
Supervision / Team
Prof. Dr. Markus Rohde
Prof. Dr. Shadan Sadeghian
Dr. Ing. Sabine Theis (DLR)
Tools & Tech
Prototype
View Prototype
Abstract
Electronic Laboratory Notebooks (ELN) have become central infrastructures for scientific documentation, offering structured data storage, traceability, and regulatory compliance. However, their prevailing interaction paradigm remains predominantly form-based and keyboard-centered, which may conflict with laboratory realities such as sterility constraints, multitasking, and distributed collaboration. This thesis investigates how a conversational interaction concept for ELN can be designed and evaluated under simulated AI conditions, with a primary focus on interaction structure rather than algorithmic performance.
Grounded in Human–Computer Interaction (HCI) theory and informed by Computer-Supported Collaborative Learning (CSCL), the work conceptualizes the ELN as a socio-technical artifact that mediates cognition, documentation practices, and collaborative continuity. Design foundations were derived from literature-informed considerations, insights from a DLR user requirements survey, and contextual observations from Kadi4Mat and POLiS laboratory environments. These inputs were synthesized into structural design drivers, including multimodal interaction, in-situ documentation support, human-in-the-loop control, and transparency in AI-mediated workflows.
A Wizard-of-Oz (WoZ) study was conducted to evaluate the conversational ELN prototype under simulated AI conditions. Voice-based documentation and chatbot-style retrieval were manually simulated to isolate interaction effects from technical AI limitations. A baseline condition using traditional documentation was compared with the prototype. Quantitative measures included the System Usability Scale (SUS), NASA-TLX workload assessment, and a Trust in Automation scale, complemented by qualitative interviews. Due to the exploratory design and limited sample size, quantitative results were analyzed descriptively.
Findings suggest that conversational interaction structures appear promising for supporting perceived workflow fluidity and contextual retrieval when technical performance is assumed optimal. At the same time, issues of predictability, transparency, and epistemic responsibility emerged as critical design considerations. The thesis contributes to HCI research by empirically examining conversational interaction paradigms within high-stakes scientific documentation contexts using a methodologically controlled simulation approach.
Research Context
Theoretical Grounding
Human-Computer Interaction (HCI) & Computer-Supported Collaborative Learning (CSCL)
Study Method
Wizard-of-Oz (WoZ) evaluation under simulated AI conditions
Domain
Electronic Laboratory Notebooks for scientific documentation & collaboration
Interaction Concept: Two Modalities
The proposed concept reframes the ELN from a passive repository into an active partner in scientific sense-making, combining two complementary interaction channels:
Voice Interaction
Hands-free documentation during active lab work, enabling researchers to record observations without breaking workflow or sterility constraints.
- In-situ experiment documentation
- Hands-free operation for sterile environments
- Real-time voice-to-structured-data capture
- Reduced cognitive context-switching
Conversational Chatbot
A retrieval-oriented interface for exploring past experiments, shared knowledge, and maintaining continuity across team members and time.
- Contextual retrieval of past experiments
- Cross-team knowledge exploration
- Collaborative continuity support
- Progressive disclosure of AI suggestions
Design Principles
Transparency
AI suggestions remain understandable with clear rationale and provenance
Progressive Disclosure
Information revealed incrementally to reduce cognitive overload
Human-in-the-Loop
Researchers maintain control over documentation decisions
Epistemic Responsibility
Preserving scientific integrity in AI-mediated workflows
Wizard-of-Oz Evaluation
Simulated AI conditions for controlled interaction assessment
A Wizard-of-Oz study was conducted to evaluate the conversational ELN prototype. Voice-based documentation and chatbot-style retrieval were manually simulated to isolate interaction effects from technical AI limitations. A baseline condition using traditional documentation was compared with the prototype.
Evaluation Measures
SUS
System Usability Scale
NASA-TLX
Workload Assessment
TiA
Trust in Automation Scale
Key Findings
- Conversational interaction structures appear promising for supporting perceived workflow fluidity
- Contextual retrieval benefits emerged when technical performance is assumed optimal
- Predictability and transparency surfaced as critical design considerations
- Epistemic responsibility challenges require careful design attention in AI-mediated scientific work
Approach & Methodology
This thesis followed a user-centered design approach, grounded in Human–Computer Interaction (HCI) and Computer-Supported Collaborative Learning (CSCL) principles. Rather than proposing a fully automated system, the focus was on designing interaction concepts that carefully integrate AI as a supportive collaborator.
User needs and design requirements were derived from a combination of empirical data and literature, including surveys conducted within a large research institute, prior workshop findings, and existing research on collaborative work, distributed cognition, and cognitive load. These insights informed the design of an AI-assisted ELN interaction concept that combines two complementary modalities:
Hands-free voice interaction, allowing researchers to document experiments during lab work without breaking focus or workflow
A conversational chatbot interface, enabling researchers to retrieve past experiments, explore shared knowledge, and maintain continuity across team members and time
Rather than treating the ELN as a passive repository, the proposed concept reframes it as an active partner in scientific sense-making. The design emphasizes transparency, progressive disclosure, and user control, ensuring that AI suggestions remain understandable and trustworthy.
To evaluate the concept despite limited access to domain experts, a Wizard-of-Oz user study was conducted. This allowed realistic interaction scenarios to be tested while keeping the AI behavior controlled and interpretable. The evaluation combined standardized measures—such as perceived workload, usability, and trust—with qualitative feedback to capture both experiential and reflective aspects of use.
Read the Full Thesis
Access the complete research document with detailed methodology, findings, and design implications.
Key Outcomes & Impact
A structured set of user needs and design requirements for AI-assisted ELNs, grounded in empirical research
A conceptual prototype illustrating how voice interaction and conversational AI can complement existing laboratory workflows
Design insights into reducing cognitive load, supporting shared understanding, and maintaining continuity across experiments
Empirical findings showing how users perceive trust, usability, and workload when interacting with AI-supported documentation tools
Practical design guidelines for integrating AI into collaborative research systems from a human-centered perspective
Beyond the artifact itself, the thesis contributes a design-oriented perspective on AI in scientific work—highlighting that the success of AI systems depends less on intelligence alone and more on how thoughtfully they are embedded into human practices, collaboration, and learning.