Aiming for a next generation Responsible AI that can enhance the effectiveness of inspectors.
This can only be achieved by developing algorithms and approaches that ensure optimal support for inspectors. We, as ICAI Lab Al4Oversight, are committed to making this possible.
In our lab different PhD researchers tackle a piece of this puzzle, each in their way contributing to a next generation of accurate, responsible and explainable algorithms that work effectively together with the inspector.
Find out below what we are currently working on!
Research Themes
Excellent & Learning Algorithms
Maximizing the performance of algorithms while adapting to the changing dynamics in the target population
Explainable and Fair Algorithms
Developing methods and applications for responsible AI-use, including fairness, explainability, sustainability, and understanding bias and uncertainties.
Effective Human-AI Interaction
Improving the computer-inspector interaction, where the inspector is supported by algorithms, and the experiences of the inspector improve the algorithm.
Research Projects
We aim to create AI models that are both fair and accurate for inspections. By integrating ethical, legal, and societal values directly into the AI training process, we ensure fair treatment for everyone from the start.
Trustworthy AI
We develop techniques like Active Learning to help inspectors efficiently label large amounts of data. By selecting the most useful cases for inspection, we aim to save time and resources while ensuring accurate and effective data collection.
Continuous Learning
Hybrid AI
We create adaptive systems that help inspectors by providing real-time, context-aware guidance.
Social Simulations
We develop an agent-based simulation platform to evaluate the effectiveness of inspection policies on a large number of inspectees before their deployment in practice.
We explore the variability in norm definitions, the answering of norms, and explaining the adherence to open norms in written reports.
Handling Ambiguity
Leveraging Dependencies
We develop methods to leverage interdependencies between rules and violations to become more effective.