Real-world sequential decision-making tasks are generally complex, requiring trade-offs between multiple, often conflicting, objectives. Despite this, the majority of research in reinforcement learning and decision-theoretic planning either assumes only a single objective, or that multiple objectives can be adequately handled via a simple linear combination. Such approaches may oversimplify the underlying problem and hence produce suboptimal results. This paper serves as a guide to the application of multi-objective methods to difficult problems, and is aimed at researchers who are already familiar with single-objective reinforcement learning and planning methods who wish to adopt a multi-objective perspective on their research, as well as practitioners who encounter multi-objective decision problems in practice. It identifies the factors that may influence the nature of the desired solution, and illustrates by example how these influence the design of multi-objective decision-making systems for complex problems.

Background The second wave of the COVID-19 pandemic was more aggressive in Brazil compared to other countries around the globe. Considering the Brazilian peculiarities, we analyze the in-hospital mortality concerning socio-epidemiological characteristics of patients and the health system of all states during the first and second waves of COVID-19. We performed a cross-sectional study of hospitalized patients with positive RT-PCR for SARS-CoV-2 in Brazil. Data was obtained from the Influenza Epidemiological Surveillance Information System (SIVEP-Gripe) and comprised the period from February 25, 2020, to April 30, 2021, separated in two waves on November 5, 2020. We performed a descriptive study of patients analyzing socio-demographic characteristics, symptoms, comorbidities, and risk factors stratified by age. In addition, we analyzed in-hospital and intensive care unit (ICU) mortality in both waves and how it varies in each Brazilian state. Between February 25, 2020 and April 30, 2021, 678 235 patients were admitted with a positive RT-PCR for SARS-CoV-2, with 325 903 and 352 332 patients for the first and second wave, respectively. The mean age of patients was 59$·$65 (IQR 48$·$0 - 72$·$0). In total, 379 817 (56$·$00%) patients had a risk factor or comorbidity. In-hospital mortality increased from 34$·$81% in the first to 39$·$30% in the second wave. In the second wave, there were more ICU admissions, use of non-invasive and invasive ventilation, and increased mortality for younger age groups. The southern and southeastern regions of Brazil had the highest hospitalization rates per 100 000 inhabitants. However, the in-hospital mortality rate was higher in the northern and northeastern states of the country. Racial differences were observed in clinical outcomes, with White being the most prevalent hospitalized population, but with Blacks/Browns (Pardos) having higher mortality rates. Younger age groups had more considerable differences in mortality as compared to groups with and without comorbidities in both waves. We observed a more considerable burden on the Brazilian hospital system throughout the second wave. Furthermore, the north and northeast of Brazil, which present lower Human Development Indexes, concentrated the worst in-hospital mortality rates. The highest mortality rates are also shown among vulnerable social groups. Finally, we believe that the results can help to understand the behavior of the COVID-19 pandemic in Brazil, helping to define public policies, allocate resources, and improve strategies for vaccination of priority groups. Coordinating Agency for Advanced Training of Graduate Personnel (CAPES) (C.F. 001), and National Council for Scientific and Technological Development (CNPq) (No. 309537/2020-7).

The recent paper “Reward is Enough” by Silver, Singh, Precup and Sutton posits that the concept of reward maximisation is sufficient to underpin all intelligence, both natural and artificial, and provides a suitable basis for the creation of artificial general intelligence. We contest the underlying assumption of Silver et al. that such reward can be scalar-valued. In this paper we explain why scalar rewards are insufficient to account for some aspects of both biological and computational intelligence, and argue in favour of explicitly multi-objective models of reward maximisation. Furthermore, we contend that even if scalar reward functions can trigger intelligent behaviour in specific cases, this type of reward is insufficient for the development of human-aligned artificial general intelligence due to unacceptable risks of unsafe or unethical behaviour.