With the increasing use of AI in high-stakes domains such as medicine, law, and defense, organizations spend a lot of time and money to make ML models trustworthy. Many books on the subject offer deep dives into theories and concepts. This guide provides a practical starting point to help development teams produce models that are secure, more robust, less biased, and more explainable. Authors Yada Pruksachatkun, Matthew McAteer, and Subhabrata Majumdar translate best practices in the academic literature for curating (...) datasets and building models into a blueprint for building industry-grade trusted ML systems. With this book, engineers and data scientists will gain a much-needed foundation for releasing trustworthy ML applications into a noisy, messy, and often hostile world. (shrink)

Business and Society Review, Volume 127, Issue S1, Page 223-251, Spring 2022.

Derivational robustness may increase the degree to which various pieces of evidence indirectly confirm a robust result. There are two ways in which this increase may come about. First, if one can show that a result is robust, and that the various individual models used to derive it also have other confirmed results, these other results may indirectly confirm the robust result. Confirmation derives from the fact that data not known to bear on a result are shown to be relevant (...) when it is shown to be robust. Second, robustness may increase the degree to which the robust result is indirectly confirmed if it increases the weight with which existing evidence indirectly confirms it. This may happen when it strengthens the connection between the core and the robust result by showing that auxiliaries are not responsible for the result. (shrink)

The most common argument against the use of rational choice models outside economics is that they make unrealistic assumptions about individual behavior. We argue that whether the falsity of assumptions matters in a given model depends on which factors are explanatorily relevant. Since the explanatory factors may vary from application to application, effective criticism of economic model building should be based on model-specific arguments showing how the result really depends on the false assumptions. However, some modeling results in imperialistic applications (...) are relatively robust with respect to unrealistic assumptions. Key Words: unrealistic assumptions • economics imperialism • rational choice • as if • robustness. (shrink)

This paper provides a conceptual framework that allows for distinguishing between different kinds of generalisation and applicability. It is argued that generalising models may bring epistemic benefits. They do so if they show that restrictive and unrealistic assumptions do not threaten the credibility of results derived from models. There are two different notions of applicability, generic and specific, which give rise to three different kinds of generalizations. Only generalising a result brings epistemic benefits concerning the truth of model components or (...) results. Abstracting the model and applying the model into new systems are not intrinsically epistemically beneficial in this way. The Dixit-Stiglitz model of monopolistic competition is used as an illustration. (shrink)

Robustness may increase the degree to which the robust result is indirectly confirmed if it is shown to depend on confirmed rather than disconfirmed assumptions. Although increasing the weight with which existing evidence indirectly confirms it in such a case, robustness may also be irrelevant for confirmation, or may even disconfirm. Whether or not it confirms depends on the available data and on what other results have already been established.

Like other mathematically intensive sciences, economics is becoming increasingly computerized. Despite the extent of the computation, however, there is very little true simulation. Simple computation is a form of theory articulation, whereas true simulation is analogous to an experimental procedure. Successful computation is faithful to an underlying mathematical model, whereas successful simulation directly mimics a process or a system. The computer is seen as a legitimate tool in economics only when traditional analytical solutions cannot be derived, i.e., only as a (...) purely computational aid. We argue that true simulation is seldom practiced because it does not fit the conception of understanding inherent in mainstream economics. According to this conception, understanding is constituted by analytical derivation from a set of fundamental economic axioms. We articulate this conception using the concept of economists' perfect model. Since the deductive links between the assumptions and the consequences are not transparent in ‘bottom‐up’ generative microsimulations, microsimulations cannot correspond to the perfect model and economists do not therefore consider them viable candidates for generating theories that enhance economic understanding. (shrink)

This paper studies the welfare consequences of strategic voting in two commonly used parliamentary agendas by comparing the average utilities obtained in simulated voting under two behavioural assumptions: expected utility maximising behaviour and sincere behaviour. The average utility obtained in simulations is higher with expected utility maximising behaviour than with sincere voting behaviour under a broad range of assumptions. Strategic voting increases welfare particularly if the distribution of preference intensities correlates with voter types.

This book provides the first comprehensive and critical examination of Mäki's realist philosophy of economics.

It is argued in this paper that amalgamating confirmation from various sources is relevantly different from social-choice contexts, and that proving an impossibility theorem for aggregating confirmation measures directs attention to irrelevant issues.

This paper provides the first systematic epistemological account of simulated data in empirical science. We focus on the epistemic issues modelers face when they generate simulated data to solve problems with empirical datasets, research tools, or experiments. We argue that for simulated data to count as epistemically reliable, a simulation model does not have to mimic its target. Instead, some models take empirical data as a target, and simulated data may successfully mimic such a target even if the model does (...) not. We show how to distinguish between simulated and empirical data, and we also offer a definition of simulation that can accommodate Monte Carlo models. We shed light on the epistemology of simulated data by providing a taxonomy of four different mimicking relations that differ concerning the nature of the relation or relata. We illustrate mimicking relations with examples from different sciences. Our main claim is that the epistemic evaluation of simulated data should start with recognizing the diversity of mimicking relations rather than presuming that only one relation existed. (shrink)

Kirjasymposio Eerik Lagerspetzin Social Choice and Democratic Values – kirjasta. Kenneth Arrowia seuraten Lagerspetz pitää sosiaalisen valinnan teorian suurimpana vahvuutena sitä, että sen tuloksia voidaan käyttää monissa erilaisissa yhteyksissä. Minä taas pidän teorian suurimpana heikkoutena sen vaikeutta: tutkijat eivät ole päässeet yhteisymmärrykseen erityisesti ns. epärelevanttien vaihtoehtojen riippumattomuusehdon tulkinnasta ja muotoilusta. Lagerspetz hyväksyy kirjassaan nähdäkseni seuraavat väitteet: kaikki demokraattiset äänestyssäännöt rikkovat riippumattomuusehtoa, mutta että tuo ehto on silti normatiivisesti perusteltu. Arrow’n ehdot ovat intuitiivisesti ottaen hyväksyttävissä. Koska riippumattomuusehdon rikkoutumisesta seuraa strategista äänestämistä (...) ja strateginen äänestäminen on normatiivisesti katsottuna ei-toivottavaa tai jopa moraalisesti kyseenalaista, strategisen äänestämisen poissulkeminen olisi paras mahdollinen argumentti riippumattomuusehdon puolesta. Lisäksi säännöt jotka toteuttavat riippumattomuusehdon ovat immuuneja strategiselle käyttäytymiselle. Hyvin toimiva äänestyssääntö luo puhtaasti mekaanisen yhteyden yksilöiden mielipiteiden ja kollektiivisen valinnan välille. Tämän kirjoituksen ensisijainen tavoite on kumota jokainen edellisessä kappaleessa esitetty väite. (shrink)

This paper provides a philosophical critique of social choice theory insofar as it deals with the normative evaluation of voting and voting rules. I will argue that the very method of evaluating voting rules in terms of whether they satisfy various conditions is deeply problematic because introducing strategic behaviour leads to a violation of any condition that makes a difference between voting rules. I also argue that it is legitimate to make interpersonal comparisons of utilities in voting theory. Combining a (...) realistic account of voters’ behaviour with a utilitarian evaluation of the outcomes then leads to the judgment that strategic voting is beneficial. If it is, then Arrow's theorem does not have far-reaching consequences for democracy because one of its conditions is not normatively acceptable. (shrink)

Arrow's Independence of Irrelevant Alternatives (IIA) has been under criticism for decades for not taking account of preference intensities. Computer-simulation results by Aki Lehtinen concerning strategic voting under various voting rules show that this intensity argument does not need to rest on mere intuition. Voters may express intensities by voting strategically, and that this has beneficial aggregate-level consequences: utilitarian efficiency is higher if voters engage in strategic behaviour than if they always vote sincerely. Strategic voting is thus unambiguously beneficial under (...) a utilitarian evaluation of outcomes. What has been considered the main argument for IIA turns out to be one against it. This paper assesses the implications of these results for interpretations of Arrow's theorem and the Gibbard-Satterthwaite theorem in a discussion on the methodological and philosophical arguments concerning preference intensities and IIA. (shrink)

Background:Moral distress has various adverse effects on nurses working in critical care. Differences in personal values, and between values and self-perception of behaviour are factors that may cause moral distress.Research aims:The aims of this study were (1) to identify ethical values and self-perception of behaviour of critical care nurses in Japan and (2) to determine the items with a large difference between value and behaviour and the items with a large difference in value from others.Research design:A nationwide, cross-sectional study was (...) conducted.Participants and research context:We developed a self-administered questionnaire with 28 items, which was completed by 1014 critical care nurses in Japan. The difference between value and self-perception of behaviour was calculated from the score of each value item minus the score of each self-perception of behaviour item. The size of the difference in value from the others was judged by the standard deviation of each item.Ethical considerations:The study was approved by the Ethics Committee of the Tokyo Medical and Dental University (approval nos. M2018-214, M2019-045).Results:The items with a large difference between value and behaviour sources were related to the working environment and decision-making support. The items with a large difference in value from others were related to hospital management and disclosure of information to patients.Discussion:Improving the working environment for nurses is important for reducing moral distress. Nurses are faced with a variety of choices, including advocating for patients and protecting the fair distribution of medical resources, and each nurse’s priorities might diverge from those of other team members, which can lead to conflict within the team.Conclusion:This study revealed items with particularly high risks of moral distress for nurses. The results provide foundational information that can guide the development of strategies to mitigate moral distress. (shrink)

A review of Donald Katzner's book on economic modelling is provided. In addition to characterising the book, I give critical comments on the distinction between primary and secondary assumptions.

Following an established tradition, the current special issue collects five articles that originate from papers presented at the IX Conference of the International Network for Economic Method. The conference took place in Helsinki on 1–3 September 2011 and was hosted by TINT (Trends and Tensions in Intellectual Integration), University of Helsinki. The conference was successful both in terms of the number of participants and the quality of the presentations. Although the sample of papers that made it to this special issue (...) is relatively small, we think that it is representative of some of the major contemporary currents in the wider field of the philosophy and methodology of economics. The papers examine both long-debated and more recent issues. (shrink)

This paper contributes to the philosophical accounts of generalisation in formal modelling by introducing a conceptual framework that allows for recognising generalisations that are epistemically beneficial in the sense of contributing to the truth of a model result or component. The framework is useful for modellers themselves because it is shown how to recognise different kinds of generalisation on the basis of changes in model descriptions. Since epistemically beneficial generalisations usually de-idealise the model, the paper proposes a reformulation of the (...) well-known distinction between abstraction and idealisation. A reformulated notion of abstraction is needed because the extant accounts yield wrong judgments when model-modifications introduce implicit assumptions. (shrink)

We claim that the process of theoretical model refinement in economics is best characterised as robustness analysis: the systematic examination of the robustness of modelling results with respect to particular modelling assumptions. We argue that this practise has epistemic value by extending William Wimsatt's account of robustness analysis as triangulation via independent means of determination. For economists robustness analysis is a crucial methodological strategy because their models are often based on idealisations and abstractions, and it is usually difficult to tell (...) which idealisations are truly harmful. (shrink)

Articles on Jaina philosophy and Jaina literature.

Aggressive behavior is instinctively driven behavior that helps animals to survive and reproduce and is closely related to multiple behavioral and physiological processes. The dorsal raphe nucleus (DRN) is an evolutionarily conserved midbrain structure that regulates aggressive behavior by integrating diverse brain inputs. The DRN consists predominantly of serotonergic (5‐HT:5‐hydroxytryptamine) neurons and decreased 5‐HT activity was classically thought to increase aggression. However, recent studies challenge this 5‐HT deficiency model, revealing a more complex role for the DRN 5‐HT system in aggression. (...) Furthermore, emerging evidence has shown that non‐5‐HT populations in the DRN and specific neural circuits contribute to the escalation of aggressive behavior. This review argues that the DRN serves as a multifaceted modulator of aggression, acting not only via 5‐HT but also via other neurotransmitters and neural pathways, as well as different subsets of 5‐HT neurons. In addition, we discuss the contribution of DRN neurons in the behavioral and physiological aspects implicated in aggressive behavior, such as arousal, reward, and impulsivity, to further our understanding of DRN‐mediated aggression modulation. (shrink)

Robert Sugden argues that robustness analysis cannot play an epistemic role in grounding model-world relationships because the procedure is only a matter of comparing models with each other. We posit that this argument is based on a view of models as being surrogate systems in too literal a sense. In contrast, the epistemic importance of robustness analysis is easy to explicate if modelling is viewed as extended cognition, as inference from assumptions to conclusions. Robustness analysis is about assessing the reliability (...) of our extended inferences, and when our confidence in these inferences changes, so does our confidence in the results. Furthermore, we argue that Sugden’s inductive account relies tacitly on robustness considerations. (shrink)

Odenbaugh and Alexandrova provide a challenging critique of the epistemic benefits of robustness analysis, singling out for particular criticism the account we articulated in Kuorikoski et al.. Odenbaugh and Alexandrova offer two arguments against the confirmatory value of robustness analysis: robust theorems cannot specify causal mechanisms and models are rarely independent in the way required by robustness analysis. We address Odenbaugh and Alexandrova’s criticisms in order to clarify some of our original arguments and to shed further light on the properties (...) of robustness analysis and its epistemic rationale. (shrink)

There has recently been some literature on the properties of a Health-Related Social Welfare Function (HRSWF). The aim of this article is to contribute to the analysis of the different properties of a HRSWF, paying particular attention to the monotonicity principle. For monotonicity to be fulfilled, any increase in individual health—other things equal—should result in an increase in social welfare. We elicit public preferences concerning trade-offs between the total level of health (concern for efficiency) and its distribution (concern for equality), (...) under different hypothetical scenarios through face-to-face interviews. Of key interests are: the distinction between non-monotonic preferences and Rawlsian preferences; symmetry of HRSWF; and the extent of inequality neutral preferences. The results indicate strong support for non-monotonic preferences, over Rawlsian preferences. Furthermore, the majority of those surveyed had preferences that were consistent with a symmetric and inequality averse HRSWF. (shrink)

A Scientific Integrity Consortium developed a set of recommended principles and best practices that can be used broadly across scientific disciplines as a mechanism for consensus on scientific integrity standards and to better equip scientists to operate in a rapidly changing research environment. The two principles that represent the umbrella under which scientific processes should operate are as follows: Foster a culture of integrity in the scientific process. Evidence-based policy interests may have legitimate roles to play in influencing aspects of (...) the research process, but those roles should not interfere with scientific integrity. The nine best practices for instilling scientific integrity in the implementation of these two overarching principles are Require universal training in robust scientific methods, in the use of appropriate experimental design and statistics, and in responsible research practices for scientists at all levels, with the training content regularly updated and presented by qualified scientists. Strengthen scientific integrity oversight and processes throughout the research continuum with a focus on training in ethics and conduct. Encourage reproducibility of research through transparency. Strive to establish open science as the standard operating procedure throughout the scientific enterprise. Develop and implement educational tools to teach communication skills that uphold scientific integrity. Strive to identify ways to further strengthen the peer review process. Encourage scientific journals to publish unanticipated findings that meet standards of quality and scientific integrity. Seek harmonization and implementation among journals of rapid, consistent, and transparent processes for correction and/or retraction of published papers. Design rigorous and comprehensive evaluation criteria that recognize and reward the highest standards of integrity in scientific research. (shrink)

ABSTRACTWe investigate the applicability of Rodrik’s accounts of model selection and horizontal progress to macroeconomic DSGE modelling in both academic and policy-oriented modelling contexts. We argue that the key step of identifying critical assumptions is complicated by the interconnectedness of the common structural core of DSGE models and by the ad hoc modifications introduced to model various rigidities and other market imperfections. We then outline alternative ways in which macroeconomic modelling could become more horizontally progressive.