For a general philosophy of computer simulations

But in any future study, these classes of simulations must be incorporated under the unificationist account of explanation as well (or at least analyzed with the possibility of an incorporation). As indicated in Section 3.2.2, this move would require a better analysis of these classes of computer simulations, one that may diverge from what I argued in Chapter 2 and Chapter 3.

The second part of this selection has to do with the simulations that I excluded in Section 4.2.1. There I argued that the metaphysics of scientific explanation de-manded us to be very careful in guaranteeing that the results of a simulation ac-tually represented the phenomenon to be explained. This demand was based on epistemological grounds: without the proper representation we cannot claim that an explanation yields understanding of that piece of the world just simulated. Com-puter simulations, however, are rich heuristic systems, understood as mechanisms for exploring the properties of models.⁷ Krohs poses another interesting example:

the Oregonator, a simulation of the Belousov-Zhabotinsky chemical reaction.⁸ The problem with this simulation was that the system of equations werestiff and lead to qualitatively erroneous results.⁹ It would be interesting to develop a theory of expla-nation that includes these simulations as genuine producers of explaexpla-nation despite the fact that they do not represent. This last point, I suspect, has direct connections with considering computer simulations asa priori experiments, where the reality of the world is subsumed to the rationality of representing it. The intricacy of the problem speaks for itself.

sentence “today is a hot day.” In this vein, some philosophical accounts of evidence prefer to describe what good evidence achieves, whereas others focus on methods by which evidence is generated. The former are necessarily post hoc in that one must already possess and have evaluated the evidence in order to do business in this tradition;¹² the latter is prior to the considerations of how evidence relates to hypotheses or theories.

The first account is typically found in two versions: the probability definition and theexplanation definition.¹³ Briefly, the probability definition e is evidence for some hypothesis h, given background b, if and only if p(h|e.b) > p(h|b); in other words, the probability of the hypothesis given e and b must be greater than the probability of the hypothesis without e, if and only if e is to count as evidence for the hypothesis. The more compelling or confirming e is, the greater the inequality between p(h|e.b) and p(h|b).¹⁴ Explanation definition, on the other hand, takes e to be potential evidence that h, if and only if e is true; in turn, h would correctly explain e if h were true¹⁵ These accounts (or any varieties of them), provide post hoc characterizations of good evidence for a hypothesis, rather than a guidance for the production of it. For the latter task we need to turn to methodological features for obtaining good evidence, a different chapter in the philosophical literature on evidence.

Two common features about evidence come from these assumptions: i) that evidence points beyond itself; and ii) that evidence is a relationship between a sentence and (a piece of) the world.¹⁶ The first assumption must be held if we want to avoid any circularity (i.e., evidence cannot be evidence for itself but must refer to something else). Therefore, the measurement of 45º C on a mercury thermometer can be evidence that the air temperature is hot, or that the mercury expanded a certain amount in space; but it cannot be evidence for itself (i.e., that it measured 45º C). The second assumption is also very strong in either account of evidence:

whether the probability definition or the explanation definition, both take evidence as causally generated (this is clearer for cases of production of evidence). In any case, causal interaction with the world grounds the relationship between a sentence and that piece of the world.

The challenge for computer simulations is to analyze these assumptions and to evaluate in what sense the data produced is evidence of the simulated phenomenon.

For instance, the second assumption entails that evidence is causally generated.

Thus, the measurement of the thermometer is actually due to the causal relation existing between the air temperature and the mercury. In what sense, then, can a non-causal device such as a computer simulation be evidence of a phenomenon in the

world? A possible solution to this problem lies in restricting the notion of evidence to a counterfactual situation. For instance, one can simulate what the measurement of the thermometer would have been, had the air temperature been such and such.

In these kinds of systems, there is no need for causal evidence for hypothetic systems.

Another possible solution is to modify the notion of evidence entirely, leaving some room for the data that computer simulations produce as evidence of the models implemented. This last solution has some kinship with the idea that computer simulations might have some use asthought experiments.¹⁷ I do not have the answers to these issues.

The general accomplishment of this study is that our understanding of the world is crossed by different methods and practices, all of which help in the general un-derstanding of it. The challenge for computer simulations is to show how they fit into the larger image of scientific practice and what sort of understanding they yield.

Finally, I would like to borrow Philip Kitcher’s words since, I believe, they express quite eloquently the feelings that most philosophers have at the end of their own work: “I doubt that I have done more than scratch the surface of unfamiliar terrain, but I am confident that epistemological rewards await those who are prepared to dig more deeply” (Kitcher, 1993, 389).

Notes

1There is another misleading interpretation presented by Wendy Parker that claims for the physical machine bringing about the empirical phenomenon (Parker, 2009). See (Durán, 2013b) for details.

2This is basically Hartmann’s claim (Hartmann, 1996). Parker (Parker, 2009) and Guala (Guala, 2002) follow this view.

3This is the view developed by Humphreys (Humphreys, 1991, 2004), Morrison (Morrison, 2009), and Rohrlich (Rohrlich, 1990). Lenhard (Lenhard, 2006) and Winsberg (Winsberg, 2009a, 2010) have a similar view on the matter.

4The interpretation of the notion of ‘basic mechanism of a phenomenon’ depends on the theory of scientific explanation: for the ontic account, for instance, it is identified with the set of causal relations that bring the phenomenon about (Salmon, 1984); for the unificationist, it consists in identifying the set of patterns of behavior that unify the phenomena. Note that the unificationist account takes unification as the systematization of the complete set of truths about the empirical world; that is, of a language that picks out genuine natural kinds, that includes all the true statements in that language (Kitcher, 1989, 495).

5Cf. (Kitcher, 1989, 478).

6See, for details (Kitcher, 1989, 447ff).

7See, for instance (Velasco, 2002; García and Velasco, 2013).

8See, for instance (Field and Noyes, 1974).

9See (Krohs, 2008, 281).

10It is worth noting that the unificationist account has also been used for explicatingconfirmation (Thagard, 1992).

11An illuminating example of the use and need of evidence in computer forensics is (Brown, 2006).

12See, for instance (Stegenga, unpublished).

13See also (Staley, 2004; Morrison, 1992; Maull, 1976; Janssen, 2002).

14Cf. (Achinstein, 1983, 328).

15Cf. (Achinstein, 1983, 335).

16See (Achinstein, 2001).

17For instance (Ristićand Radulović, 2001).

Summary (English)

Computer simulations have proven to be a fundamental tool for the advance-ment and developadvance-ment of scientific practice. This is especially true in areas related to scientific experimentation, where computer simulations stand out with a strong presence. Such pervasiveness in the sciences draws the attention of scientists and philosophers both interested in understanding their epistemic power. Naturally, one can find advocates for as well as detractors against computer simulations, holding strong arguments against each other. This work is my attempt to build a system-atic defense of the epistemic power of computer simulations, not only against the attacks perpetrated by its critics, but also for providing a substantial assessment of the claim about their epistemic power.

With these motivations firmly in mind, the general aim of this work is to show that computer simulations yield understanding of some aspects of the world by ex-plaining simulated phenomena. Since understanding is an epistemic concept par excellence, and since there is general agreement that scientific explanation is a coherence-making process, it follows that the epistemic power of computer simu-lations can be defended by showing in virtue of what computer simusimu-lations explain and what kind of understanding they yield. In this context, I have chosen the unifi-cationist theory of scientific explanation as a conceptual framework for explanation in computer simulations. Such is the general line of argumentation. In this thesis, I address the following questions:

1. In which respects do computer simulations differ from laboratory experimen-tation and why do they represent a novelty for the philosophy of science?

2. What are the advantages that computer software offer, which are absent in laboratory experimentation but which are beneficial for conceptualizing the epistemic power of computer simulations?

3. What is the nature of a computer simulation? Is it abstract, due to the mathematical/logical nature of algorithms? Or is it causal, due to the nature

4. What are the conditions that must be imposed onto a computer simulation in order for it to confer reliability as a process that genuinely represents an empirical target system?

5. How is the explanatory process carried out and in what sense does such an explanation yield understanding of the simulated phenomenon as well as the empirical phenomenon?

These questions parallel the structure of this work. The first question is addressed in Chapter 1, where I analyze current philosophical literature on models and scientific experimentation, with a strong emphasis on methodological and epistemological dif-ferences with computer simulations. In this chapter, I center on differencesvis à vis laboratory experiments, urging for a change in the way philosophers approach the epistemology of computer simulations. Concretely, I urge for analyzing computer simulations at face value rather than by means of comparing them with labora-tory experiments. The final section is dedicated to outlining some answers to the skepticism regarding the novelty of computer simulations in the philosophical arena.

Motivated by these results, I propose to systematically construct a working con-ceptualization of computer simulations; this undertaking begins in Chapter 2 and continues in Chapter 3. For this, I begin by studying the nature of computer soft-ware as expounded by philosophers of computer science. Three units of analysis are of major interest; namely, the specification, the algorithm, and the computer process. These units of analysis are meant to answer the second question above by showing the benefits of computer software (e.g., syntax manipulability and syn-tax transference) absent in laboratory experimentation that speak in favor of the epistemic power of the former. In addition, understanding the nature of computer software facilitates the conceptualization of computer simulations. In this sense, an evaluation of each unit of analysis and the relations among them is essential for the robustness of this work.

The third question is more complex since it is at the heart of the nature of computer simulations and, in a way, determines their fate as epistemic devices.

For these reasons, Chapter 3 is intended to work on several levels: first, it reviews the literature on simulations, examining the differences between analog and digital simulations. Second, it narrows down the universe of all computer simulations that can be found in scientific practice, dividing it into equation-based simulations on the one hand, and cellular automata, agent-based, and complex systems on the other. This distinction is meant to restrict the domain of computer simulations of interest to equation-based simulations. Third, it discusses the methodology of computer simulations and analyzes one concrete paradigmatic example of the class of