Search

The pursuit of trust in and fairness of AI systems in order to enable human-centric goals has been gathering pace of late, often supported by the use of explanations for the outputs of these systems. Several properties of explanations have been highlighted as critical for achieving trustworthy and fair AI systems, but one that has thus far been overlooked is that of descriptive accuracy (DA), i.e., that the explanation contents are in correspondence with the internal working of the explained system. Indeed, the violation of this core property would lead to the paradoxical situation of systems producing explanations which are not suitably related to how the system actually works: clearly this may hinder user trust. Further, if explanations violate DA then they can be deceitful, resulting in an unfair behavior toward the users. Crucial as the DA property appears to be, it has been somehow overlooked in the XAI literature to date. To address this problem, we consider the questions of formalizing DA and of analyzing its satisfaction by explanation methods. We provide formal definitions of naive, structural and dialectical DA, using the family of probabilistic classifiers as the context for our analysis. We evaluate the satisfaction of our given notions of DA by several explanation methods, amounting to two popular feature-attribution methods from the literature, variants thereof and a novel form of explanation that we propose. We conduct experiments with a varied selection of concrete probabilistic classifiers and highlight the importance, with a user study, of our most demanding notion of dialectical DA, which our novel method satisfies by design and others may violate. We thus demonstrate how DA could be a critical component in achieving trustworthy and fair systems, in line with the principles of human-centric AI.

We introduce a conceptualisation for generating argumentation frameworks (AFs) from causal models for the purpose of forging explanations for mod-els’ outputs. The conceptualisation is based on reinterpreting properties of semantics of AFs as explanation moulds, which are means for characterising argumentative relations. We demonstrate our methodology by reinterpreting the property of bi-variate reinforcement in bipolar AFs, showing how the ex-tracted bipolar AFs may be used as relation-based explanations for the outputs of causal models. We then evaluate our method empirically when the causal models represent (Bayesian and neural network) machine learning models for classification. The results show advantages over a popular approach from the literature, both in highlighting specific relationships between feature and classification variables and in generating counterfactual explanations with respect to a commonly used metric.

A user receiving an explanation for outcomes produced by an artificially intelligent system expects that it satisfies the key property of descriptive accuracy (DA), i.e. that the explanation contents are in correspondence with the internal working of the system. Crucial as this property appears to be, it has been somehow overlooked in the XAI literature to date. To address this problem, we consider the questions of formalising DA and of analysing its satisfaction by explanation methods. We provide formal definitions of naive, structural and dialectical DA, using the family of probabilistic classifiers as the context for our analysis. We evaluate the satisfaction of our given notions of DA by several explanation methods, amounting to two popular feature-attribution methods from the literature and a novel form of explanation that we propose and complement our analysis with experiments carried out on a varied selection of concrete probabilistic classifiers.

BackgroundHealthcare workers (HCW) are at increased risk of being infected with SARS-CoV-2, yet limited information is available on risk factors of infection.MethodsWe pooled data on occupational surveillance of 10,654 HCW who were tested for SARS-CoV-2 infection in six Italian centers. Information was available on demographics, job title, department of employment, source of exposure, use of personal protective equipment (PPE), and COVID-19-related symptoms. We fitted multivariable logistic regression models to calculate odds ratios (OR) and 95% confidence intervals (CI).FindingsThe prevalence of infection varied across centers and ranged from 3.0% to 22.0%, being strongly correlated with that of the respective areas. Women were at lower risk of infection compared to men. Fever, cough, dyspnea and malaise were the symptoms most strongly associated with infection, together with anosmia and ageusia. No differences in the risk of infection were detected between job titles, or working in a COVID-19 designated department. Reported contact with a patient inside or outside the workplace was a risk factor. Use of a mask was strongly protective against risk of infection as was use of gloves. The use of a mask by the source of exposure (patient or colleague) had an independent effect in reducing infection risk.

In recent years, machine learning (ML) models have been successfully applied in a variety of real-world applications. However, they are often complex and incomprehensible to human users. This can decrease trust in their outputs and render their usage in critical settings ethically problematic. As a result, several methods for explaining such ML models have been proposed recently, in particular for black-box models such as deep neural networks (NNs). Nevertheless, these methods predominantly explain outputs in terms of inputs, disregarding the inner workings of the ML model computing those outputs. We present Argflow, a toolkit enabling the generation of a variety of ‘deep’ argumentative explanations (DAXs) for outputs of NNs on classification tasks.

Despite the rapid growth in attention on eXplainable AI (XAI) of late, explanations in the literature provide little insight into the actual functioning of Neural Networks (NNs), significantly limiting their transparency. We propose a methodology for explaining NNs, providing transparency about their inner workings, by utilising computational argumentation (a form of symbolic AI offering reasoning abstractions for a variety of settings where opinions matter) as the scaffolding underpinning Deep Argumentative eXplanations (DAXs). We define three DAX instantiations (for various neural architectures and tasks) and evaluate them empirically in terms of stability, computational cost, and importance of depth. We also conduct human experiments with DAXs for text classification models, indicating that they are comprehensible to humans and align with their judgement, while also being competitive, in terms of user acceptance, with existing approaches to XAI that also have an argumentative spirit.

Background: The present study was aimed at evaluating the effectiveness of BNT162b2 among HCWs of a university hospital while a recrudescence of pandemics was hitting the province, with a high rate of the B.1.1.7 variant. Methods: The study was performed in the context of health surveillance at the workplaces. We monitored the SARS-CoV-2 infection and COVID-19 symptoms among HCWs classified by having received the entire vaccine schedule or not; the latter further classified in not vaccinated workers and workers who had received the first shot more than 14 days earlier. The SARS-CoV-2 infection was diagnosed by conventional RT-PCR on rhino-pharyngeal swabs, followed by gene sequencing in positive vaccinated HCWs. The cumulative incidence of infections in the period was normalised to 100,000 people. Results: At the end of the observation period, HCWs that had completed the full schedule were at lower infection risk than both unvaccinated HCWs and the workforce who had not yet gained the complete theoretical protection from the vaccine (by 2.4-folds). Overall, ninety-two SARS-CoV-2 infections were observed among HCWs, mostly among not protected workers (52 cases) but none of them showed symptoms requiring hospitalisation. Conclusions: The vaccination campaign effectively reduced the appearance of symptoms and the incidence of infections among vaccinated HCWs. Among vaccinated HCWs, gene sequencing was possible in five cases only, 4 B.1.1.7 and 1 B1.525 variants. The high rate of unsuccessful gene sequencing observed among infected vaccinated workers could be explained by a low viral burden. Vaccination for COVID-19 should be mandatory in occupational settings with a high infective risk.

BACKGROUND: Prostate cancer (PCa) is the second cause of mortality in men worldwide. The prostate-specific antigen (PSA) test is routinely adopted in diagnosis; nevertheless more reliable biomarkers are continuously under investigation by monitoring the release of molecules into the bloodstream. The serum protein profiles appear to provide cancer-specific fingerprints that help to discriminate patients (especially with low PSA level) from controls, improving the performance of existing clinical tests. METHODS: Samples from healthy controls and PCa patients with low (4ng/mL) levels were analyzed by MALDI profiling, and by a multi fractionation approach coupled to ESI-MS for peaks identification. RESULTS: MALDI profiling achieved to detect 10 and 14 changed peaks (p-value

Este artigo objetiva discutir algumas questões relacionadas à importância da pronúncia para professores brasileiros de inglês, tais como: O que os professores pensam sobre este assunto? Os professores de inglês da rede pública deveriam ter conhecimento de Fonética e Fonologia para ensinar inglês? A falta deste conhecimento pode influenciar o aprendizado dos alunos? Os professores tiveram aulas de pronúncia ao longo de sua formação? Os professores têm dificuldade de pronúncia em inglês? Quais? Estes professores consideram sua própria pronúncia importante para seus alunos? Para discutir essas e outras perguntas, um questionário foi desenvolvido e aplicado para um grupo de professores de inglês formados em Letras que trabalham em escolas da rede pública da região metropolitana de Curitiba. Para fins comparativos, o mesmo questionário também foi aplicado a um grupo de professores de inglês de cursos ou escolas particulares e professores universitários. Os resultados apontam as dificuldades, os desafios e a preocupação destes professores de inglês em relação a sua própria pronúncia e ao ensino desta na sala de aula.

Indoleamine 2,3-dioxygenase 1 (IDO1) is a potent immunoregulatory enzyme that catalyses the degradation of the essential amino acid tryptophan (Trp) along the kynurenines pathway. Significant changes in systemic Trp catabolism have been reported in many diseases, including cancer and autoimmunity. In female nonobese (NOD) mice, IDO1 expression and hence immune tolerance to pancreatic b-cell autoantigens are defective in conventional dendritic cells stimulated with IFN-γ, the main IDO1 inducer. Although the evidences in NOD mice suggest that IDO1 function is impaired, the existence of the IDO1 defect in human T1D (type 1 diabetes) has not been proven yet. Here we monitored the IDO1 expression and activity in peripheral blood mononuclear cells (PBMCs) of children with T1D as compare to age-matched control subject, in response to IFN- γ. Results from kynurenines assay and Western blot analysis demonstrate that the majority of patients with T1D is characterized by defective Trp catabolism. Moreover, our data indicated that this defect is mainly imputable to a SOCS3-mediated, dysregulated IL-6 signaling that would favor IDO1 proteasomal degradation in inflammatory environments, i.e. dominated by IFN-γ. To confirm this, we measured IDO1 expression and activity in PBMCs co-incubated with IFN-γ and Tocilizumab (TCZ), a licensed IL-6 receptor blocker. Results showed that TCZ is able to restore normal levels of IDO1 catalytic activity in response to IFN-γ in approximately 30% of the examined T1D population. Besides further confirming the heterogeneity of the disease, our data indicate the existence of a subset of individuals with T1D who may gain clinical benefit in restoring immunoregulatory mechanisms by treatment with tocilizumab.

To compare data obtained in a pilot study (2002) which evaluated the risk from exposure to stress and burnout in health care workers, the same three subjective questionnaires were administered: the Job Content Questionnaire (JCQ), the State-Trait Anxiety Inventory (STAI), the Maslach Burnout Inventory (MBI). The evaluation considered 294 workers employed in eight units: two of the six previous units were reorganized in four subunits. Preliminary data showed that questionnaires' scores have a concordant trend in the different units, as observed in 2002, confirming the validity of the instruments adopted. Results indicated a decreased level of the perceived stress in all the units, especially in the Third Division. We can hypothesize that the new organization, workers' turnover during the 5 years and the effect due to individual variables could have contribute to the observed variation. No significant associations, compared to 2002, between questionnaires' scores and task were found. Further evaluations, including measurement of objective parameters, will be carried out to complete the follow-up study and to determine which variables could have a role in the variation of the levels of stress and burnout's subjective perception.

Our Institute has been following for 20 years a group of workers of a ferroalloy industry in order to evaluate neurobehavioral effects due to manganese exposure. Five years after the last study we have planned another one, to evaluate differences in neuromotor e cognitive functions between exposed and controls and to perform a longitudinal evaluation of the results. Environmental and biological sampling were collected, liver and kidney functionality, haemochrome, iron metabolism and sieric prolactine were evaluated. Several tests were administered: postural evaluation, tremor, four tests of the SPES battery, Pursuing Aiming, five tests of the Luria Nebraska Motor Battery, Raven Progressive Matrices, Trail Making Test, Mood Scale, Brief Symptoms Inventory, neuropsychological symptoms questionnaire. Personal habits and working, living and clinical histories were collected. We evaluated 43 exposed workers and 40 controls. Exposure indicators resulted all significantly higher in exposed workers. Neuropsychological examination showed differences in Raven Progressive Matrices and Pursuit Aiming, higher tremor values and differences in postural evaluation between exposed and controls.

Aiming to register workplace violence in health care workers, we adopted two evaluation methods: direct and indirect. The direct method is represented by the Violent Incident Form (VIF), a 16 items checklist, which resumes the key aspects of violence events, identifying spatial and temporal circumstances, aggressor, type of violence and consequences. Indirect method is constituted by the observation of injury reports, through three sources: the Prevention and Protection Service (SPP), the Public Relation Office and the Police Position of the First Aid Unit. Although the instruments adopted, thus not comparable each other, pointed out a rather high number of violence events suffered by health care workers (direct method: 34 reports in the period 2002-2006; indirect method: 25 subjects, 8.5% of total); in most cases, the events were perpetrated by a patient, and directed against female of nurses. The most reported types of violence were: kicking, beating, hitting (indirect method) and verbal assault/threats (direct method). The study pointed out that workplace violence is an important risk factor, and therefore it is necessary to consider it for preventive intervention.

The existence of a risk of musculoskeletal disorders work-related of the upper extremity (UEWMSDs) in the textile industry, specifically in particular working phases as spinning, appears today controversial. The upper limb disorders, sometimes described, have not a location ever plausible with the level of the biomechanical overload, as when carpal tunnel syndromes are signalled during activities in which the shoulder is the only segment eventually interested. Moreover these findings are shown in workers appointed at tasks or actions not clearly identified at risk. For example the spinning activity, that requires the movement of shuttles of modest weight, sometimes on levels above the shoulder line, is organized with cyclical cadences sometimes as repetitive tasks concentrated in a short period, sometimes in longer during the entire shift. Distribution and number of the actions, would however allow a sufficient biomechanical recovery. The identification of the single technical action may result difficult, due to interindividual variability of actions, of their speed and complexity. Furthermore the other possible activities, alternate with specific spinning activities, can require strength or not correct posture even if they may have short duration. In this paper we present and discuss the results of the ergonomic survey for risk assessment of musculoskeletal disorders work-related of the upper limb. These analysis have been carried out in two textile plants, in which some cases of disorders of the upper limb in workers employed in spinning activities have been described by occupational health physicians. In addition to assessment of possible risk for UEWMSDs in spinning activities, we founded the capability of usual methods for ergonomic analysis to adequately examine work situations like spinning, in which the hazardous actions are diluted over the entire shift or concentrate in a short period and in which there is a specific biomechanically overloaded segment i.e. the shoulder. Finally attention was paid to evaluate psychosocial factors which appear to be important when biomechanical factors are low-lying.

A cross sectional field study was planned to assess neurotoxic effects caused by low-level occupational lead exposure. Two groups of 66 workers and 86 controls were examined with a battery including a questionnaire on neurotoxic symptoms, the measure of performance at neurobehavioral testing, the detection of visual contrast sensitivity, and the dosage of serum prolactin. Both current and cumulative exposure to lead were defined. The average PbB was 27.50 +/- 28 microg/dl (median 28, range 6-61) in the exposed and 8.11 +/- 4.47 microg/dl (median 7, range 2-21). The test results were controlled for possible confounders including age, schooling, alcohol and coffee intake. Significant differences were observed between exposed and controls regarding neurotoxic symptoms reporting, the exposed reporting more frequently mood changes and abnormal fatigue. The exposed subjects showed a decreased visual contrast sensitivity, and a marked increase of prolactin secretion. No changes emerged regarding neurobehavioral testing. The alterations observed resulted associated to the current lead exposure and not to the cumulative indices. A safe exposure level was calculated on the basis of dose-response relationship with prolactin alteration, yielding a PbB value of 10 microg/dl.