1,720,967 research outputs found
Analysis of some experiments of human evaluation of XAI
No full textThe black-box problem of Machine Learning is often dealt from a mathematical and computational standpoint. With this lead, the risk is high of missing the primary point of eXplainable AI (XAI) research: are the numerical artifacts we produce through these new methods good explanations of an AI's output? How should we evaluate explanations in relation with their human use?Here is a report of several years of investigations regarding human evaluation of XAI, by a PhD student raised in mathematics and informatics. I analysed 26 articles that provide objective results in human-AI interfaces, focusing on human-AI performance, trust and understanding. The results of this literature are often statistically weak or seamingly contradictory: they do not give any positive clue to direct technical research inside post-hoc explanation. However, they put an end to some stereotypes, like explanations being an hypnotic means to guide people towards obeying AI (which is a legitimate concern).Caution! This publication is made using experiments pre-dating 2024. The author estimate the expiration date of this article to early 2027, regarding any scientific credit you could be willing to give it. If you live in a post-2028 world, please try updating this knowledge. This work is a v1 intended as a scientific report, without scientific redaction style. Style should be updated later.</div
Développement d’interaction homme-IA pour l’explication des décisions des algorithmes : application au domaine financier
No full textArtificial intelligence (AI) has established itself for over a decade in the form ofMachine Learning (ML) to carry out numerous prediction tasks. However, ML models are oftenblack boxes, whose behavior is difficult to interpret, even for their designers. The generation of automatic explanations is a potential solution to this problem, but presents two key challenges: how to produce such explanations, and how to ensure their relevance?This work focuses on post-hoc local explanation generation. Our applications at CréditMutuel Arkéa involved detecting illicit behaviors, through the development of an interfacefor fraud analysts and adapting Feature Importance techniques to a deep representationmodel for transaction flows. We then explored evaluation through human performance, viewing explanation as an aid to human-AI cooperation. We conducted experiments outside the banking domain, first focusing on SHAP, followed by a system using multiple forms of explanation. Although our explanations did not improve decision-making performance, they revealed key factors for refining both the explanatory tools and our evaluation methods. These results open promising avenues for better adapting explanations to users’ needs.L’intelligence artificielle (IA) s’est imposée depuis plus d’une dizaine d’années sous la forme du Machine Learning (ML) pour de nombreuses tâches de prédiction. Toutefois, les modèles de ML sont souvent des boîtes noires dont le comportement est difficile à interpréter, même pour leurs concepteurs. La génération automatique d’explications est une piste pour résoudre ce problème, qui amène deux défis : comment produire de telles explications, et comment s’assurer de leur pertinence? Ce travail se focalise sur les techniques d’explication locale post-hoc. Nos applications au sein du Crédit Mutuel Arkéa ont portés sur la détection de comportements illicites, avec le développement d’une interface pour les analystes bancaires, et l’adaptation de Feature Importances à un modèle profond de représentation des transactions. Ensuite, nous avons exploré la piste d’uneévaluation par les performances humaines, qui traite l’explication comme une aide à la coopération dans l’interaction homme-IA. En s’appuyant sur une tâche extra-bancaire, nousavons réalisé des expérimentations portant d’abord sur SHAP, puis sur un système à explications multiples. Nos résultats, bien que négatifs en termes de performance, révèlent des facteurs clés pour perfectionner les outils explicatifs et nos évaluations
Développement d’interaction homme-IA pour l’explication des décisions des algorithmes : application au domaine financier
No full textArtificial intelligence (AI) has established itself for over a decade in the form ofMachine Learning (ML) to carry out numerous prediction tasks. However, ML models are oftenblack boxes, whose behavior is difficult to interpret, even for their designers. The generation of automatic explanations is a potential solution to this problem, but presents two key challenges: how to produce such explanations, and how to ensure their relevance?This work focuses on post-hoc local explanation generation. Our applications at CréditMutuel Arkéa involved detecting illicit behaviors, through the development of an interfacefor fraud analysts and adapting Feature Importance techniques to a deep representationmodel for transaction flows. We then explored evaluation through human performance, viewing explanation as an aid to human-AI cooperation. We conducted experiments outside the banking domain, first focusing on SHAP, followed by a system using multiple forms of explanation. Although our explanations did not improve decision-making performance, they revealed key factors for refining both the explanatory tools and our evaluation methods. These results open promising avenues for better adapting explanations to users’ needs.L’intelligence artificielle (IA) s’est imposée depuis plus d’une dizaine d’années sous la forme du Machine Learning (ML) pour de nombreuses tâches de prédiction. Toutefois, les modèles de ML sont souvent des boîtes noires dont le comportement est difficile à interpréter, même pour leurs concepteurs. La génération automatique d’explications est une piste pour résoudre ce problème, qui amène deux défis : comment produire de telles explications, et comment s’assurer de leur pertinence? Ce travail se focalise sur les techniques d’explication locale post-hoc. Nos applications au sein du Crédit Mutuel Arkéa ont portés sur la détection de comportements illicites, avec le développement d’une interface pour les analystes bancaires, et l’adaptation de Feature Importances à un modèle profond de représentation des transactions. Ensuite, nous avons exploré la piste d’uneévaluation par les performances humaines, qui traite l’explication comme une aide à la coopération dans l’interaction homme-IA. En s’appuyant sur une tâche extra-bancaire, nousavons réalisé des expérimentations portant d’abord sur SHAP, puis sur un système à explications multiples. Nos résultats, bien que négatifs en termes de performance, révèlent des facteurs clés pour perfectionner les outils explicatifs et nos évaluations
Going Beyond Counting First Authors in Author Co-citation Analysis
Full text linkThe present study examines one of the fundamental aspects of author co-citation analysis (ACA) - the way co-citation
counts are defined. Co-citation counting provides the data on which all subsequent statistical analyses and mappings
are based, and we compare ACA results based on two different types of co-citation counting - the traditional type that
only counts the first one among a cited work's authors on the one hand and a non-traditional type that takes into
account the first 5 authors of a cited work on the other hand. Results indicate that the picture produced through this non-traditional author co-citation counting contains more coherent author groups and is therefore considerably clearer. However, this picture represents fewer specialties in the research field being studied than that produced through the traditional first-author co-citation counting when the same number of top-ranked authors is selected and analyzed. Reasons for these effects are discussed
Learning how to use a Decision Support System: a simulation of Human-AI interaction frames
No full textMany studies address human-AI interaction by focusing on the direct behavioral effects of AI on humans, often assuming a static view of human decision-making. However, a defining feature of human intelligence is its capacity for continuous learning. Experiments designed to study this learning process are challenging to implement due to their duration and the presence of numerous uncontrollable variables. To overcome these limitations, it is essential to explore potential solutions through numerical simulations. This study proposes a simulation paradigm, and use it to explore new frames (trust quota and mentalisation) for human-automation teaming. We compare them with reference approaches (human alone and AI as an advisor), simulating the learning phenomenon using artificial neural networks. Our findings suggest that trust quotas can boost human-performances, but that only the effort to develop mental models of AI may foster a resilience to the loss of AI advice
Variations on the Author
Full text link“Variations on the Author” discusses two of Eduardo Coutinho’s recent films (Um Dia na Vida, from 2010, and Últimas Conversas, posthumously released in 2015) and their contribution to the general question of documentary authorship. The director’s filmography is characterized by a consistent yet self-effacing form of authorial self-inscription: Coutinho often features as an interviewer that rather than express opinions propels discourses; an interviewer that is good at listening. This mode of self-inscription characterizes him as an author who is not expressive but who is nonetheless markedly present on the screen. In Um Dia na Vida, however, Coutinho is completely absent form the image, while Últimas Conversas, on the contrary, includes a confessional prologue that moves the director from the margins to the center of his films. This article examines the ways in which these works stand out in the filmography of a director who offers new insights into the notion of cinematic authorship
Appropriate Similarity Measures for Author Cocitation Analysis
Full text linkWe provide a number of new insights into the methodological discussion about author cocitation analysis. We first argue that the use of the Pearson correlation for measuring the similarity between authors’ cocitation profiles is not very satisfactory. We then discuss what kind of similarity measures may be used as an alternative to the Pearson correlation. We consider three similarity measures in particular. One is the well-known cosine. The other two similarity measures have not been used before in the bibliometric literature. Finally, we show by means of an example that our findings have a high practical relevance.information science;Pearson correlation;cosine;similarity measure;author cocitation analysis
Dispelling the Myths Behind First-author Citation Counts
Full text linkWe conducted a full-scale evaluative citation analysis study of scholars in the XML research field to explore just how different from each other author rankings resulting from different citation counting methods actually are, and to demonstrate the capability of emerging data and tools on the Web in supporting more realistic citation counting methods. Our results contest some common arguments for the continued
use of first-author citation counts in the evaluation of scholars, such as high correlations between author rankings by first-author citation counts and other citation
counting methods, and high costs of using more realistic citation counting methods that are not well-supported by the ISI databases. It is argued that increasingly available digital full text research papers make it possible for citation analysis studies to go beyond what the ISI databases have directly supported and to employ more
sophisticated methods
- …
