Philosophical issues related to risks and values

: This paper begins with the assumption that the concept of risk implies an entanglement between facts and values. This is not an arbitrary assumption since it can directly be deduced from the standard notion of risk. The value-ladenness of risk raises at least two further issues: the first one concerns the scales adopted to evaluate the severity of risks; the second concerns the commensurability/comparability of risks to human health and the environment. Some additional light is shed on those issues whether the models used in risk analysis were understood as fictions limited by the values that they can include. From this point of view, controversies on the limited scope of standard risk assessments are not only descriptive but also evaluative.


Philosophical issues related to risks and values
Renato Rodrigues Kinouchi 1 Are risks value-free?
Risk is a polysemic word usua ly associated with the general idea of danger, harm or loss resulting from a ions whose consequences cannot be completely foreseen.Its etymology is uncertain because similar words can be traced back to Greek, Latin and Arabic, but there is enough evidence on its Mediter anean origin and its ecific relatedness with maritime voyage hazards such as strong storms, hi den reefs, pirate attacks, etc. (Knutsen et al., 2012).This ancient nautical slang was incorporated into economic and legal jargon with the rise of merchant capitalism and the increasing demand for maritime commerce insurance 2 .Today the word risk is employed in multifarious theoretical and pra ical contexts -for instance, statistics, decision theory, finance, medicine, engineering, ecology, climatology, sociology, etc. -which contributes to the plurality of its meanings.
In order to a dress the initial question, it is necessary to begin with one of the availa le definitions of risk, and it seems reasona le to choose the standard technical definition usua ly adopted by risk analysts: risk is the expectation alue of an unwanted event defined as the product of the probability of the event -expressed as a number within the interval [0, 1] -by a quantitative estimate of its severity -that is, the magnitude of the harm (cf.Hansson, 2013, p. 9-10).It is important to note that, according to such definition, risk varies as a function of both the probability of the event and the amount of damage it may cause.For instance, sup osing the probability of an earthquake in Alaska and in California were the same, the latter would pose a higher risk due to the significant difference in both the material losses and the number of people potentia ly affected.Thus, even for this technical definition, risk cannot be value-free since one of the equation terms must imply some measure of a value (human lives, material goods, etc.) under risk.The philosopher Sven Hansson (2004Hansson ( , 2005Hansson ( , 2009Hansson ( , 2012Hansson ( , 2013) ) has often emphasized that "risk always refers to the possibility that something undesira le wi l hap en.Due to this component of undesirability, the notion of risk is value-laden" (Hansson, 2013, p. 10).
Such value-ladenness does not mean, however, that risk is then a fact-free concept grounded exclusively upon human value judgements.Risk assessments may be seen as socially constructed in the same trivial sense that any human inquiry depends on social cooperation, lin uistic conventions, etc., but, although it is true that people actua ly have very different risk perceptions about, say, earthquakes, it is beyond any reasona le doubt that strong seismic events are rea ly much more frequent along the boundaries of tectonic plates, which in turn justifies investments in building design, evacuation alarm systems and other socia ly constructed safety mechanisms in cities located near geological faults.For Hansson,

In this way, risk is both fact-laden and value-laden […]
A notion of risk that connects in a reasonable way to the conditions of human life will have to accommodate both its fact-ladenness and its value-ladenness.The real challenge is to disentangle the facts and the values sufficiently from each other to make well-informed and well-ordered decision processes possible (Hansson, 2013, p. 11).
The fact-value distinction played an important role in early philosophical works on risk analysis since it made possile to detect hi den value assumptions usua ly overlooked by scientists, engineers and risk analysts. 3More recently, Mö ler (2012) reframed this issue by considering the concept of risk as a thic ethical conce t.Thick ethical concepts are concepts that have both descriptive and evaluative contents such as, say, "cruel" , "brave" and "selfish" (cf.Wi liams, 2006).These concepts are har ly reduci le to thin ethical conce ts like "right" or "bad" because they exhibit descriptive features absent in the latter.As an i lustration, to say that a certain person is selfish does not only mean he or she behaves ba ly but denotes a ecific way he or she gives priority to himself or herself over others.Thick concepts, in short, belong to a grey zone where descriptive and evaluative a ects, related to a given state of affairs, merge into each other.According to Mö ler, "that something is safe is a positive feature of the entity, and that something car ies a risk is a negative feature of it.But it is not simply positive or negative, it is positive or negative in a certain ay; it has certain descriptive shape" (Mö ler, 2012, p. 75, original italics).
For Mö ler, risk should be considered a thick concept since to affirm that the situation S is risky involves both a descriptive dimension about the potentia ly harmful event, which amounts to the likelihood of S coming about (e.g., the probability of an earthquake and the estimated distance of its epicenter from highly populated areas), and an evaluative dimension related to a precise forecast of its severity (e.g., the number of residents proba ly affected, the existence of chemical industries or nuclear plants in the area).This thickness of risk involves a functional distinction less committed to the ontological assumptions present in the traditional dichotomy judgements of facts versus values judgements.Importantly, Mö ler reframed the discussion with an ap roach resistant to reductionist views: "there is an essential interdependence between the natural-descriptive a ects and the normative a ects [...] The output of recent moral philosophy is skepticism of the reductive claim for thick concepts such as risk and safety" (Mö ler, 2012, p. 76).

Risks and value comparisons
What is the role played by values in risk analysis?If we begin with the preliminary question "what are values?", the usual answer includes a huge variety of desira le things worthy of pursuance by human beings, from personal needs, wants and pleasures to more objective goods such as health, wealth and, above a l, the preservation of our own lives.The field of axiology, understood in a broad sense, includes a conste lation of further questions such as "what is the nature of values?", "are values subjective or objective?","are there intrinsic values?","can values be compared?",and so on.The last question, about value comparisons, is decisive to clarify how risk analysts measure and compare the severities of potential harms.
In standard risk assessments, the quantity of deaths (casualties) is a widespread evaluative measure of severity.In order to calculate the expectation value of a risk, each life is valued the same and therefore can be arithmetica ly a ded to other lives, making it possi le to construct an interval scale where positive integers (1, 2, 3, …, n) define degrees of severity.So, the value of life is disposed into an interval scale because "for the expected value of harm to be we l defined [...] we must be a le to decide not only that harm A is more severe than B, and that B is more severe than C, but also the relative severity between them" (Mö ler, 2012, p. 63).In other words, if each life is valued the same, then it is possi le to say that "five deaths are five times worse than one death" .More importantly, this assumption takes for granted value commensurability: casualties become a cardinal unit to measure severity.
One may say that risk analysis cannot rely only on such quantitative measure.Sup ose a risk that does not cause death but instead leads to limb amputation.For this case, is losing one leg equal to losing one arm?If so, is losing two legs twice worse than losing one arm?In short, is it reasona le to measure the severity of limb amputation by an interval scale?One may say that the (dis)value of limb amputation depends on personal preferences: for example, for Admiral Nelson the lack of an arm was not significant for the victory in the Battle of Trafalgar, but proba ly a naval officer without a leg would be very much limited in his ability to command (due to ship instabilities, decks with lot of stairs, etc.); on the other hand, for the aviator Captain Douglas Ba ler, the lack of both legs did not jeopardise his performance in the Battle of Dunkirk, but he would har ly be a fighter ace without an arm.Although it does not sound reasona le to propose an interval scale to measure the severity of limb amputations, severity may be compara le by ordinal scales of preference: for Admiral Nelson, a leg was more valua le than an arm; for Captain Ba ler, an arm was more valua le than a leg.
The value-ladenness of risk involves the issue of commensurability/comparability of the unwanted outcomes.According to Ruh Chang, "two items are incommensura le just in case they cannot be put on the same scale of units of value, that is, there is no cardinal unit of measure that can represent the value of both items" (Chang, 2015, p. 205).As to incomparability, "two items are incompara le just in case they fail to stand in an evaluative comparative relation, such as being better than or worse than or equa ly as good as the other" (Chang, 2015, p. 205).From a logical point of view, incomparability entails incommensurability but the reverse does not hold.In what ecifica ly regards risks, Nicolas Espinoza describes those relations in the fo lowing way: I shall say that two risks are evaluatively incommensurable if and only if there is no cardinal scale with respect to which the severity of both risks can be compared.In addition, two risks are evaluatively incomparable if and only if it is not the case that they can be ordinally ranked, which is to say that is not the case that one risk is better, worse or equally as good as the other.Note that incommensurability thus defined does not necessarily imply incomparability; the failure to compare two risks cardinally, for instance the failure to say that risk A is, say, three times more severe than B, does not automatically imply that we cannot say that risk A is more severe than risk B. It may be helpful to view the distinction between incommensurability and incomparability, namely that between ordinal and cardinal measurement, as analogous to the distinction between quantitative and qualitative comparison (Espinoza, 2009, p. 129, emphasis mine).
In order to point out a pro lem concerning risk comparisons, let's denominate as ca dinaliza le those values which can be ar anged into interval scales (e.g.cardinal utility, money, number of casualties, etc.), and let's denominate as o dinaliza le those values which can be ar anged into ordinal scales (e.g.ordinal utility, preferences, rankings).This distinction is usual in statistics, economics and related disciplines: for instance, the notion of cardinal utility has a long history going back to Bentham and it was further developed by neoclassical economists and by Von Neumann and Morgenstern; regarding the notion of ordinal utility, it is widely used by contemporary economists and Bayesian statisticians.
Nevertheless, there is a background question on whether every value can be adequately disposed into interval or ordinal scales.There is no impediment to sup ose that certain values could not be ar anged into any type of scale, resulting in their incomparability.For example, there are discussions about value incomparability in ethical debates concerning incompensa le har s4 (e.g.Thomson, 1986;Shrader-Frechette, 1991) and in empirical studies on the so-ca led protected alues5 (e.g.Baron and Spranca, 1997;Tetlock et al., 2000).As a matter of fact, "some people think that some of their values are protected from trade-offs with other values" and they hold those values "as possessing infinite or transcendental significance that precludes comparisons, trade-offs, or indeed any other mingling with bounded or secular values" (Espinoza, 2009, p. 130-131).And in a dition, "laypersons often feel that it is unethical to assign monetary prices to risk imposed upon humans or the environment" (Espinoza, 2009, p. 130).This issue embar asses the method of cost-benefit analysis (CBA), which tries to compare risks using assignments of a monetary price as a common measure for their severities: In a typical CBA, two or more options in a public decision are compared to each other by careful calculation of their respective consequences.These consequences can be different in nature, e.g. economic costs, risk of disease and death, environmental damage etc.In the final analysis, all such consequences are assigned a monetary value, and the option with the highest value of benefits minus costs is recommended or chosen [...] Cost-benefit analysis is controversial and has repeatedly been subject to severe criticism not least from philosophers.Most of this criticism has focused on two practices.One of these is the assignment of a monetary price to (the loss of) a human life.The other is contingent valuation, in which the prices of non-market goods such as environmental assets are determined by asking people what they are willing to pay for them (Hansson, 2007, p. 163-164;cf. Espinoza, 2009, p. 130).
Against such criticisms, experts and decision-makers tend to ar ue that cost-benefit analysis permits to estimate how much should be ent in a ions of risk prevention and risk management, with the intention of a locating finite resources to minimize harms of different sorts and magnitudes.From this point of view, sacred and protected values may be of interest to psychologists and philosophers, but "if incomparability is widespread, then what we do in most choice situations fa l outside the scope of pra ical reason" (Chang, 2015, p. 206).Cost-benefit analysis takes for granted that values under risk can be measured (by means of interval scales) or compared (by means of ordinal scales) because without "accurate comparisons in terms of severity, we wi l not be a le to perform accurate and cost-effective trade-offs between risks and their associated benefits" (Espinoza, 2009, p. 131).As stated above, the procedure of comparing values disposed into interval or ordinal scales is also taken for granted in several other fields of scientific inquiry.Risk analysts just fo low the same e a lished path, despite the wor y on the part of some laypersons and philosophers.

Limitations of standard risk assessments
Technological risks are particularly pro lematic for standard risk assessments.The main pro lem concerns the lack of statistical data to estimate the objective probability of potential harms.Standard risk assessments have been successfu ly ap lied to natural catastrophes, occupational accidents, shipwrecks and traffic accidents, etc., for which there are immense statistical recordings.The potential harms of technological innovations cannot be estimated in the same way for an obvious reason: they never occur ed before.Risk assessments on technological innovations require models based on conditional probabilities subjectively estimated by engineers and risk analysts.Indeed, due to the lack of previous statistical data, technological risks are not risks properly eaking, but uncertainties.
There is voluminous literature on the distinction between risk and uncertainty according to the type of probabilistic knowledge availa le.The distinction involves several complications which have been extensively discussed in decision theory, in economics, in statistics and in philosophy (see Hájek and Hitchcock, 2016).But besides the type of probabi-listic knowledge availa le, the values involved also influence our choices.Roughly eaking, the epistemic uncertainty related to our limited probabilistic knowledge comes accompanied by an evaluative uncertainty related to our value judgements.Such evaluative limitation should not be disregarded in matters such as, for example, biotechnological innovations whose large-scale side effects can be ir eversi le (Garcia and Martins, 2009;Lacey, 2005Lacey, , 2009;;Mariconda, 2014;Martins, 2012).Consider the following passage about risk assessment of transgenic agriculture: Standard risk assessment has kept some potentially harmful varieties of transgenics from being marketed.Nevertheless, it is unable to address, among other things, (a) potential social risks -e.g.monopolization of the world's food supply, undermining the conditions for other forms of farming, impoverishment and dislocation of small-scale farmers -and (b) potential risks to the environment occasioned by transgenics in virtue of the fact that usually they are commodities and integral to current projects of large corporations, and (c) ecological and long-term environmental risks that arise because of social mechanisms, e.g., the failure (or inability) of farmers to adhere to regulations that are assumed to be in place when judgments of risk in practice are based on standard risk assessment (Lacey, 2009, p. 853-854).
Lacey has detailed the controversies on transgenics in dozens of works, but here I can only briefly outline the general lines of his ar ument presented in "The interplay of scientific a ivity, worldviews and value outlooks" .For Lacey, biotechnologists confine their research within a decontextualized a proac (DA) in which values related to social we l-being and environmental safety are excluded and "empirical data are selected, sought out [...] and reported using descriptive categories that are genera ly quantitative, ap lica le in virtue of measurement, instrumental and experimental operations" (Lacey, 2009, p. 843).Consequently, standard risk assessments tend to focus only "on the quantitati e and probabilistic study of (anticipated) haza ds for health and the environment over the relatively short time scale of laboratory and contro led field studies, deploying categories accepta le within DA" (Lacey, 2009, p. 853, italics mine).The decontextualized ap roach goes hand in hand with a materialist worldview deeply informed by the values of technological progress.Nevertheless, for Lacey, materialism and the values of technological progress are insufficient "to justify adopting the decontextualized ap roach virtua ly to the exclusion of conducting research under competing strategies" (Lacey, 2009, p. 851).
In a dition, one may ar ue that socio-environmental values are extremely important and indeed strongly prefera le to the values of technological progress, up to the point that no further benefits from transgenics could balance the risks imposed on health and on the environment by their use.In the last years such an ar ument against transgenics, which evokes the precautionary principle, has be un to inform an increasing number of restrictive policies in the European Union. 6 I wish to conclude with the su gestion that some a ditional light is shed on this issue if we consider that risk analysis, like other scientific pra ices, requires the construction of models which must be assumed to be fictions.Several authors such as Cartwright (1983Cartwright ( , 1999Cartwright ( , 2010)), Contessa (2010), Fine (1993), Godfrey-Smith (2006, 2009), Fri g (2010a, 2010b, 2010c), Leng (2001), Morgan (2001), Rahman and Redmond (2015) have discussed such ap roach -usua ly ca led fictionalism in philosophy of science 7 -according to which "scientific textbooks and journal articles abound with passages that ap ear to be meaningful plain descriptions of physical systems […] but which do not describe actual systems and which would not be taken to do so by any competent pra itioner in the field" (Fri g, 2010b, p. 257, original italics).The analogy between models and fictions is controversial (see Giere, 2009) but, in the context of risk analysis, how could we anticipate potential harms without constructing risk models which are assumed to be fictions?Could we make more realistic experiments intending to know the real extent of risky situations?From a descriptive point of view, fictionalism does not require from risk analysis more than the latter can offer, namely, imaginary risk scenarios.
More important, however, is the claim that "we wi l have to distort the true picture of what hap ens if we want to fit it into the highly constrained structures of our mathematical theories" (Cartwright, 1983, p. 139).For Cartwright, "a model is a work of fiction" in the sense that "some properties ascribed to objects in the model wi l be genuine properties of the objects mode led, but others wi l be [...] introduced into this model as a convenience, to bring the objects mode led into the range of the mathematical theory" (Cartwright, 1983, p. 153).Specifica ly concerning standard risk assessments, it is important to realize how they are decisively constrained by the way in which risk analysts take values into account.Risk analysts take for granted some idealized assumptions on value measurement, commensurability, and comparability, which have been usual in contemporary social sciences and economics; nevertheless, the point is that those idealized assumptions limit the scope of standard risk assessments.So, significant potential harms from technological innovations may not be anticipated by models/fictions constructed to find out only the risks whose severity measures fit in the equations of standard risk assessments.
To regard risk models as fictions does not mean they are useless mathematical tools.The intention is to keep in mind how deceptive they may be in virtue of the limited range of values which they include.The value-ladenness of risk implies issues which indeed affect every field of scientific inquiry that demands some procedure of value measurement.The models used in standard risk assessments are fictions, like other scientific models, in the sense that they take for granted some idealized assumptions on values that define which potential harms we wi l anticipate and which ones we wi l not.This is particularly relevant in controversies about technological innovations, not only because determining the probability of those potential harms is a hard task, but mainly because we can fail to evaluate the severity of those harms.