R&D-02
System
Explainable AI Research
Type
Computational Systems Research
Domain
Artificial Intelligence · Explainability · Human-Computer Interaction
Year
2020
Location
Europe
As machine learning systems become increasingly complex, their decision processes grow opaque, limiting human understanding and institutional trust in AI systems. // As machine learning systems scale, their decision processes become increasingly opaque, limiting interpretability, trust, and effective human interaction.
Structural Gap
Black-box models prevent meaningful interpretability, making it difficult for researchers, institutions, and users to understand how algorithmic decisions are produced. // Explainability methods remain fragmented and lack a user-centered framework, making it difficult to determine which explanation is appropriate for a given use case and audience.
Interpretability is often treated as a universal property, rather than a context-dependent and user-specific phenomenon.
Intervention
Research investigating structural approaches to explainability within machine learning systems.
The work explored methods for translating complex computational processes into interpretable representations accessible to human reasoning.
The research sits at the intersection of explainable AI, human-computer interaction, and systemic modeling. // Development of a user-centric framework for explainable AI based on the premise that interpretability is inherently dependent on the recipient.
The work introduced a recommendation mechanism that maps user profiles, use cases, and explanation features to identify the most suitable explanation method.
This included:
– structuring the conceptual foundations of interpretability
– clustering users based on expertise and preferences
– identifying patterns between user profiles and preferred explanation characteristics
Outcome
Development of research exploring interpretability, transparency, and cognitive accessibility in machine learning systems, contributing to broader conversations on responsible and human-centered AI. // A proof of concept demonstrating that interpretability is a polylithic and user-dependent construct, and that personalized explanation selection improves understanding, satisfaction, and trust in AI systems.
The work establishes a foundation for adaptive, user-specific explainability frameworks in machine learning.