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

TL;DR

Endless Tuning addresses the problem of AI deployment that risks human replacement and responsibility gaps in high-stakes decisions. The authors propose a relational, double-loop protocol with soft interpreters and hermeneutic explanations, implemented in three prototypical decision tasks and evaluated via domain-expert interviews. Key contributions include a modular protocol, an interface that fosters reflection and traceability, and a governance-friendly log for accountability. The findings suggest users retain control and that accountability can be bridged, though data quality and interface design pose ongoing challenges.

Abstract

The Endless Tuning is a design method for a reliable deployment of artificial intelligence based on a double mirroring process, which pursues both the goals of avoiding human replacement and filling the so-called responsibility gap (Matthias 2004). Originally depicted in (Fabris et al. 2024) and ensuing the relational approach urged therein, it was then actualized in a protocol, implemented in three prototypical applications regarding decision-making processes (respectively: loan granting, pneumonia diagnosis, and art style recognition) and tested with such as many domain experts. Step by step illustrating the protocol, giving insights concretely showing a different voice (Gilligan 1993) in the ethics of artificial intelligence, a philosophical account of technical choices (e.g., a reversed and hermeneutic deployment of XAI algorithms) will be provided in the present study together with the results of the experiments, focusing on user experience rather than statistical accuracy. Even thoroughly employing deep learning models, full control was perceived by the interviewees in the decision-making setting, while it appeared that a bridge can be built between accountability and liability in case of damage.

Figures (7)

• Figure 1: A graphical representation of the double loop of Endless Tuning (image taken from fabris2024towards; originally made by the same author of this paper; received permission by the editors).
• Figure 2: Diagram of the proposed Endless Tuning protocol in a decision-making setting. In order: the user is asked their first impression; is given suggestions through explanations before knowing the outcome; is given suggestions by comparing the present case with similar ones in the training set; looks at the outcome with confidences; after a sufficient number of sessions, the model is finetuned.
• Figure 3: (a) A screenshot of the suggestion extracted from the decision tree w.r.t the case n. 144, which happened to be chosen by the bank director. The translation of the actual previous note, given w.r.t the first impression, is: « Worth considering. Perhaps the years of employment [in a hesitant tone]; they are not too few, but not too much either. We [the bank] would also observe the solvency of the employer, mildly». (b) Similarity table w.r.t case n. 144 (excluding plot).
• Figure 4: (a) The saliency map produced with the https://github.com/jacobgil/pytorch-grad-camjacobgilpytorchcamselvaraju2017grad explainer, regarding the picture chosen in the art style recognition test. (b) The corresponding https://github.com/eclique/RISEpetsiuk2018rise explanation (both maps presented before knowing the outcome, exploring a hermeneutic strength). (c) The picture chosen by the professor of aesthetics. Notably, it is a painting of 1659 by Francisco de Zurbarán representing Saint Francis.
• Figure 5: An example of the recorded log (art style recognition test).