Evaluating AI decision tools in Ecuador’s courts: efficiency, consistency, and uncertainty in legal judgments

This study explores the impact of AI-based decision support tools on judicial performance in Ecuador, a context characterized by institutional uncertainty and procedural inefficiencies. It assesses whether such tools improve efficiency, consistency, and the normative quality of legal reasoning in judicial decisions. A mixed-methods approach was applied to analyze fifty court cases before and after AI implementation. Quantitative analysis used <i>t</i>-tests, Levene's test, and Mann-Whitney U test to evaluate procedural duration and inter-rater agreement, while natural language processing techniques, including topic modeling (LDA) and sentiment analysis (VADER), assessed changes in semantic structure and argumentation. In parallel, a content analysis of twelve policy and regulatory documents was conducted to examine changes in algorithmic governance discourse. The results show a statistically significant reduction in case resolution time (-23.5 days), an increase in inter-evaluator consistency (Cohen's kappa from 0.65 to 0.80), a shift toward more neutral-technical language, and greater density of legal citations. Mentions of governance principles such as transparency and accountability also increased. These findings indicate that AI-based tools, when used as assistive systems, can enhance judicial decision-making in uncertain environments without displacing human deliberation. While the study provides robust initial evidence, its exploratory sample and reliance on interpretable NLP techniques reflect the constraints of a low-resource judicial context and highlight avenues for future research. This research contributes to the literature on advanced analytical methods for institutional decision-making under legal and epistemic uncertainty.