funded by the Hans-Böckler Foundation
The TranKI project develops standards for the user interfaces of AI systems that allow users of the systems to better understand and contextualize the system’s results.
It’s neither necessary nor possible for users to understand, trace, or verify decision-making or recommendation systems powered by AI (for example, systems for personnel and shift scheduling, recommendation systems for further training, or systems supporting hiring processes) in detail. These systems utilize large datasets that continuously change. The underlying models and algorithms are also complex and time-consuming to comprehend, even for experts.
However, for many applications, it is necessary to be able to understand the outputs of the system, i.e., the factors influencing the system results, to accept or possibly correct them. This goal can be partially achieved at the level of the user interface:
- by displaying variables that influence the result,
- by graphically representing the uncertainty of results,
- by offering options to influence outcomes by altering variables,
- by displaying metrics of fairness (e.g., „37% of displayed profiles are female“),
- through counterfactual explanations (e.g., „a person with 2 years less professional experience would have been recommended this position“).
The developed standards standardize the auditing of AI systems and can be utilized in the development, selection, and deployment of HR systems.
Project Duration
October 1st, 2023 – September 30th, 2026
Project Lead
Project Staff
Student Assistant
B.Sc. Thao Do
Funding Agency