SRL is developing a digital research environment for AI-supported research. It stems from the question of how complex research processes can be continued across individual working sessions while keeping sources, work states, checks and decisions traceable.
The digital research environment serves as a research tool. Its development and application also raise questions about the division of labour between humans and AI and the technical and methodological conditions this requires.
Synthesis as a Development Principle
The research infrastructure grows out of specific requirements arising from research practice. To address these requirements, SRL examines established models, technical approaches and best practices and combines those components that are essential to the problem at hand.
This results in a solution hypothesis that is put into practice, tested and iteratively developed. Additional structures and controls are developed where they make a tangible contribution to the reliability, traceability or effectiveness of the research process.
Infrastructure development thus follows the same fundamental research logic as SRL’s substantive research: existing knowledge is related to a specific question, examined within the limits of what it can support, and condensed into a verifiable synthesis.
Digital Research Environment
A key component of the research environment is the persistence of research states beyond individual AI working sessions. This allows lines of inquiry to be resumed and developed further over longer periods.
The research environment is designed to preserve previous states and make changes traceable. A research output, its review, the decision about its use and its current standing within the research process are treated as distinct processes. New findings may alter existing results while their origin and revision remain traceable.
At the same time, the research environment enables a far-reaching division of labour between humans and AI. Research, comparison, analysis and the development of alternative explanations can be supported by generative AI. Technically verifiable properties can be checked automatically. Decisions about research questions, evaluation, research direction and published results remain distinct from these processes.
An Example: Quality Assurance
In developing quality assurance procedures, SRL examines established audit, risk and assurance models for their applicability to AI-supported research processes.
In one such investigation, SRL examined Failure Mode and Effects Analysis (FMEA), including its treatment in a guide published by NASA’s Goddard Space Flight Center. For the specific problem under investigation, the distinction between possible failure modes, their causes, effects and detectability proved particularly relevant. Further aspects were drawn from other audit, review and risk models.
The components of such models that are relevant to the specific problem are combined with further findings and condensed into a solution hypothesis that is then tested in research practice.
Development in Research Practice
Parts of the research infrastructure are already being used and further developed in the Vindobona project. There, the digital research environment supports the continuation of complex research processes through persistent work states, traceable revisions and controlled transitions between different working and decision states.
Its application also raises new research questions: Which work steps can meaningfully be transferred to generative AI? What information must be retained outside a model across individual working sessions? Which checks can be automated? Where is human decision-making required? And how can a far-reaching division of labour be structured so that provenance, uncertainty and responsibility remain traceable?
The answers to these questions, in turn, feed into the further development of research methods and infrastructure.