Explainability of Algorithms

Opacity comes in two distinct forms: Technical opacity results from the genuine complexity of algorithms like neural networks that process data through multiple hidden layers that are inherently difficult for humans to understand, while intentional opacity occurs when companies deliberately hide how their systems work for proprietary or commercial reasons—these require different ethical and regulatory approaches.

Black box algorithms affect all stakeholders equally: Even the designers and developers of complex AI systems cannot fully access or explain their inner workings, meaning no single group has complete understanding of how these systems make decisions that impact people's lives.

Complexity alone doesn't justify opacity: The paper distinguishes between unavoidable technical limitations and deliberate concealment, suggesting that commercial AI systems hiding their decision-making processes for business reasons raise distinct ethical concerns beyond what technical complexity alone would warrant.

Explainable AI (XAI) solutions exist but remain incomplete: While computer scientists have developed various methods to make AI systems more transparent and interpretable, the paper indicates these explanatory approaches still face "numerous challenges" and cannot fully resolve the opacity problem.

Explainability is essential for ethical AI development: The paper identifies algorithm opaqueness as "one of the main obstacles to the ethical development of artificial intelligence," suggesting that understanding how AI systems work is a prerequisite for responsible deployment in high-stakes decisions.