The Endless Tuning. An Artificial Intelligence Design To Avoid Human Replacement and Trace Back Responsibilities

AI systems can be designed to keep humans in control rather than replace them. The Endless Tuning method was tested across three real-world applications (loan decisions, medical diagnosis, and art recognition) and domain experts reported feeling they maintained "full control" over decision-making, even when using deep learning models.

A "responsibility gap" can be addressed through thoughtful system design. The paper proposes that by using a "double mirroring process," developers can create a bridge between accountability and liability, helping clarify who is responsible when AI systems cause harm—a persistent challenge in current AI deployment.

Explainability tools work better when reversed and adapted for context. Rather than simply explaining what an AI decided, the protocol uses a "reversed and hermeneutic deployment" of explainability algorithms, suggesting that how we present AI reasoning to users matters as much as the reasoning itself.

User experience and trust matter more than raw accuracy metrics. The research prioritized testing how domain experts felt about the system and whether they understood their role, rather than focusing solely on statistical performance—indicating that responsible AI requires measuring human confidence and comprehension alongside technical metrics.

Different ethical approaches to AI are needed. The paper explicitly adopts what it calls "a different voice" in AI ethics (referencing Gilligan's work), suggesting that current frameworks may be incomplete and that alternative philosophical perspectives should inform how we design AI systems.