The FATE flagship develops AI capabilities for a digital assistant that acquires and extends its expertise through continuous learning from multiple potentially confidential and biased (subject) data sources and from human experts who add to and reflect on the AI-outcomes. The system provides decision support for multiple user roles, such as a researcher, consultant and subject.
The aim of the FATE flagship is to develop an expert assistant, where both the system and the user learn from each other through iterative interaction. The resulting classifications, predictions and advices will comply with the applicable fairness principles and will be communicated in an understandable and trustworthy way to the direct stakeholders.
The developed FATE Flagship architecture aims at easy adoption of the various linked use cases, through the creation of reusable functional components, that also reflect the key research topics in FATE:
- Fair AI: being able to detect, mitigate and evaluate bias in data and AI system.
- Explainable AI: being able to explain the reasons behind an advice to various user roles.
- Co-learning: being able to learn from users at the system and the individual level. At the individual level the task is to learn from the user (and adapt) and observe that the user has learned from the given advice.
- Secure learning: being able to handle (distributed) sensitive data by identifying and assessing potential information leaks and proposing secure-by-design alternatives.
Humans can interact with the system in various roles such as e.g. an AI developer/data scientist, practitioner (subject-matter or domain expert) or subject (affected person). These roles are the viewpoints of our research and system development. The three user roles that we distinguish are:
- Researcher (a domain researcher, e.g. a medical researcher, a criminologist, a sociologist),
- Consultant (advisor of a subject, e.g. a medical doctor, a judge, a civil servant),
- Subject (affected person, e.g., a patient/consumer, a convicted person, someone receiving social assistance).
These roles define three different types of users: one who has a research interest in the domain of the use case, one who uses the system to advise or intervene a subject’s actions or behaviour, and one whose actions or behaviour the system attempts to influence.
FATE has been running for three years. In year one we adopted a healthcare use case (decision support for diabetes), in year two a juridical use case (AI4Justice) and this year Skills Matching is adopted as use case, in which we take inspiration from a job seeking and vacancy/CV skills matching scenario.
- Milena Kooij-Janic, Sr Project Leader, TNO, e-mail: firstname.lastname@example.org
- Joachim de Greeff, Sr Consultant, TNO, e-mail: email@example.com