Justificatory explanations in machine learning: for increased transparency through documenting how key concepts drive and underpin design and engineering decisions

Given the pervasiveness of AI systems and their potential negative effects on people's lives (especially among already marginalised groups), it becomes imperative to comprehend what goes on when an AI system generates a result, and based on what reasons, it is achieved. There are consistent technical efforts for making systems more "explainable" by reducing their opaqueness and increasing their interpretability and explainability. In this paper, we explore an alternative non-technical approach towards explainability that complement existing ones. Leaving aside technical, statistical, or data-related issues, we focus on the very conceptual underpinnings of the design decisions made by developers and other stakeholders during the lifecycle of a machine learning project. For instance, the design and development of an app to track snoring to detect possible health risks presuppose some picture or another of "health", which is a key notion that conceptually underpins the project. We take it as a premise that these key concepts are necessarily present during design and development, albeit perhaps tacitly. We argue that by providing "justificatory explanations" about how the team understands the relevant key concepts behind its design decisions, interested parties could gain valuable insights and make better sense of the workings and outcomes of systems. Using the concept of "health", we illustrate how a particular understanding of it might influence decisions during the design and development stages of a machine learning project, and how making this explicit by incorporating it into ex-post explanations might increase the explanatory and justificatory power of these explanations. We posit that a greater conceptual awareness of the key concepts that underpin design and development decisions may be beneficial to any attempt to develop explainability methods. We recommend that "justificatory explanations" are provided as technical documentation. These are declarative statements that contain at its simplest: (1) a high-level account of the understanding of the relevant key concepts a team possess related to a project's main domain, (2) how these understandings drive decision-making during the life-cycle stages, and (3) it gives reasons (which could be implicit in the account) that the person or persons doing the explanation consider to have plausible justificatory power for the decisions that were made during the project.

Individual prevention and containment measures in schools in Catalonia, Spain, and community transmission of SARS-CoV-2 after school re-opening

BACKGROUND

Despite their clear lesser vulnerability to COVID-19, the extent by which children are susceptible to getting infected by SARS-CoV-2 and their capacity to transmit the infection to other people remains inadequately characterized. We aimed to evaluate the role of school reopening and the preventive strategies in place at schools in terms of overall risk for children and community transmission, by comparing transmission rates in children as detected by a COVID-19 surveillance platform in place in Catalonian Schools to the incidence at the community level.

METHODS AND FINDINGS

Infections detected in Catalan schools during the entire first trimester of classes (September-December 2020) were analysed and compared with the ongoing community transmission and with the modelled predicted number of infections. There were 30.486 infections (2.12%) documented among the circa 1.5M pupils, with cases detected in 54.0% and 97.5% of the primary and secondary centres, respectively. During the entire first term, the proportion of "bubble groups" (stable groups of children doing activities together) that were forced to undergo confinement ranged between 1 and 5%, with scarce evidence of substantial intraschool transmission in the form of chains of infections, and with ~75% of all detected infections not leading to secondary cases. Mathematical models were also used to evaluate the effect of different parameters related to the defined preventive strategies (size of the bubble group, number of days of confinement required by contacts of an index case). The effective reproduction number inside the bubble groups in schools (R*), defined as the average number of schoolmates infected by each primary case within the bubble, was calculated, yielding a value of 0.35 for primary schools and 0.55 for secondary schools, and compared with the outcomes of the mathematical model, implying decreased transmissibility for children in the context of the applied measures. Relative homogenized monthly cumulative incidence ([Formula: see text]) was assessed to compare the epidemiological dynamics among different age groups and this analysis suggested the limited impact of infections in school-aged children in the context of the overall community incidence.

CONCLUSIONS

During the fall of 2020, SARS-CoV-2 infections and COVID-19 cases detected in Catalan schools closely mirrored the underlying community transmission from the neighbourhoods where they were set and maintaining schools open appeared to be safe irrespective of underlying community transmission. Preventive measures in place in those schools appeared to be working for the early detection and rapid containment of transmission and should be maintained for the adequate and safe functioning of normal academic and face-to-face school activities.

Urban seismic monitoring in Brasília, Brazil

Urban seismology has gained scientific interest with the development of seismic ambient noise monitoring techniques and also for being a useful tool to connect society with the Earth sciences. The interpretation of the sources of seismic records generated by sporting events, traffic, or huge agglomerations arouses the population’s curiosity and opens up a range of possibilities for new applications of seismology, especially in the area of urban monitoring. In this contribution, we present the analysis of seismic records from a station in the city of Brasilia during unusual episodes of silencing and noisy periods. Usually, cultural noise is observed in high-fequency bands. We showed in our analysis that cultural noise can also be observed in the low-frequency band, when high-frequency signal is attenuated. As examples of noisy periods, we have that of the Soccer World Cup in Brazil in 2014, where changes in noise are related to celebrations of goals and the party held by FIFA in the city, and the political manifestations in the period of the Impeachment trial in 2016, which reached the concentration of about 300,000 protesters. The two most characteristic periods of seismic silence have been the quarantine due to the COVID-19 pandemic in 2020, and the trucker strike that occurred across the country in 2018, both drastically reducing the movement of people in the city.

Effect of wall type, delayed mortality and mosquito age on the residual efficacy of a clothianidin-based indoor residual spray formulation (SumiShield™ 50WG) in southern Mozambique

Indoor residual spraying (IRS) is one of the main malaria vector control strategies in Mozambique alongside the distribution of insecticide treated nets. As part of the national insecticide resistance management strategy, Mozambique introduced SumiShield™ 50WG, a third generation IRS product, in 2018. Its residual efficacy was assessed in southern Mozambique during the 2018-2019 malaria season. Using a susceptible Anopheles arabiensis strain, residual efficacy was assessed on two different wall surfaces, cement and mud-plastered walls, using standard WHO (World Health Organization) cone bioassay tests at three different heights. Female mosquitoes of two age groups (2-5 and 13-26 day old) were exposed for 30 minutes, after which mortality was observed 24h, 48h, 72h, and 96h and 120h post-exposure to assess (delayed) mortality. Lethal times (LT) 90, LT50 and LT10 were estimated using Bayesian models. Mortality 24h post exposure was consistently below 80%, the current WHO threshold value for effective IRS, in both young and old mosquitoes, regardless of wall surface type. Considering delayed mortality, residual efficacies (mosquito mortality equal or greater than 80%) ranged from 1.5 to ≥12.5 months, with the duration depending on mortality time post exposure, wall type and mosquito age. Looking at mortality 72h after exposure, residual efficacy was between 6.5 and 9.5 months, depending on wall type and mosquito age. The LT50 and LT10 (i.e. 90% of the mosquitoes survive exposure to the insecticides) values were consistently higher for older mosquitoes (except for LT10 values for 48h and 72h post-exposure mortality) and ranged from 0.9 to 5.8 months and 0.2 to 7.8 months for LT50 and LT10, respectively. The present study highlights the need for assessing mosquito mortality beyond the currently recommended 24h post exposure. Failure to do so may lead to underestimation of the residual efficacy of IRS products, as delayed mortality will lead to a further reduction in mosquito vector populations and potentially negatively impact disease transmission. Monitoring residual efficacy on relevant wall surfaces, including old mosquitoes that are ultimately responsible for malaria transmission, and assessing delayed mortalities are critical to provide accurate and actionable data to guide vector control programmes.

(Internet) Gaming Disorder in DSM-5 and ICD-11: A Case of the Glass Half Empty or Half Full: (Internet) Le trouble du jeu dans le DSM-5 et la CIM-11: Un cas de verre à moitié vide et à moitié plein

BACKGROUND

Diagnostic and Statistical Manual of Mental Disorders (DSM-5) included in 2013 Internet gaming disorder (IGD) as a condition for further study, and in 2018, the World Health Organization included gaming disorder (GD) as a mental disorder in the International Classification of Disease (ICD-11). We aim to compare disorders of gaming in both diagnostic systems using a sample of young adults in Mexico.

METHODS

Self-administered survey to estimate the prevalence of DSM-5 IGD and ICD-11 GD in 5 Mexican universities; 7,022 first-year students who participated in the University Project for Healthy Students, part of the World Health Organization World Mental Health International College Student Initiative. Cross-tabulation, logistic regression, and item response theory were used to inform on 12- month prevalence of DSM-5 IGD and ICD-11 GD, without and with impairment.

RESULTS

The 12-month prevalence of DSM-5 IGD was 5.2% (95% CI, 4.7 to 5.8), almost twice as high as the prevalence using the ICD-11 GD criteria (2.7%; 95% CI, 2.4 to 3.1), and while adding an impairment requirement diminishes both estimates, prevalence remains larger in DSM-5. We found that DSM-5 cases detected and undetected by ICD-11 criteria were similar in demographics, comorbid mental disorders, service use, and impairment variables with the exception that cases detected by ICD-11 had a larger number of symptoms and were more likely to have probable drug dependence than undetected DSM-5 cases.

CONCLUSION

DSM-5 cases detected by ICD-11 are mostly similar to cases undetected by ICD-11. By using ICD-11 instead of DSM-5, we may be leaving (similarly) affected people underserved. It is unlikely that purely epidemiological studies can solve this discrepancy and clinical validity studies maybe needed.