Bridging Human Expertise and Intelligent Agents
An interactive, community-driven handbook for selecting knowledge elicitation techniques, built on reciprocal human-agent collaboration and machine-readable semantic models.
The Elicitation Challenge
Traditional knowledge elicitation techniques (such as structured interviews) are indispensable for extracting expert logic, yet they often introduce cognitive biases, omissions, or memory misattributions. Selecting the correct elicitation framework is essential to preserve analytical accuracy.
Reciprocal Elicitation
We introduce human-agent cooperation to elevate elicitation. Reciprocal Knowledge Elicitation (RKE) describes the continuous, mutual exchange of knowledge extraction and feedback provision between humans and artificial agents in a collaborative team structure.
Community-Driven Origins
This platform was conceived through the RKEHAC Workshop Series (Reciprocal Knowledge Elicitation in Human-Agent Collaboration) to design a digitized taxonomy of elicitation practices. A primary objective is delivering semantic, machine-readable representations using RDF to enable seamless operational execution by AI agents.
The REXPEK Project
This handbook is partly motivated and funded by the REXPEK (Reproducing Expert Knowledge) project. In modern industrial operations, expert operators possess crucial, often mathematically unwritten, domain setpoints and controller-tuning skills that dictate system efficiency. REXPEK aims to capture and digitize this tacit expertise to optimize system design cycles.
By analyzing operational telemetry data during system loops without disturbing workflow, REXPEK extracts setpoints and optimization behaviors directly from normal daily activities.
Instead of asking operators to write complex mathematical formulas, the system presents intuitive prompts (e.g., rating, ranking, or rejecting suggestions) to confirm criteria implicitly.
The captured models empower junior operators to execute controller tuning loops and system diagnoses with efficiency equivalent to, or surpassing, that of a veteran operator.
A Joint Research Contribution
The methodologies, semantic taxonomies, and handbook items are a collaborative research result of four leading academic and technical institutions.