Source-linked AI summary
Grounds for trust: Essential Epistemic Opacity and Computational Reliabilism
Juan M. Durán, Nico Formanek
TL;DR
Computer simulations raise the problem of epistemic opacity: researchers may be unable to survey every process involved. The paper develops computational reliabilism to explain how simulation results can nevertheless be trusted when the processes are reliable.
Problem
Epistemic opacity leaves open how researchers can trust computer-simulation results when they cannot survey every relevant process.
Source: §1 · p. 2, §2.1 · p. 4, §2.3 · p. 8Method
The paper reconstructs epistemic opacity and develops computational reliabilism, identifying verification and validation, robustness analysis, implementation history, and expert knowledge as reliability sources.
Source: §1 · p. 3, §4 · p. 14Results
The paper argues that computer simulations can provide knowledge despite epistemic opacity when their results are produced by reliable processes.
Source: §1 · p. 3Takeaways & Limitations
Reliability sources differ in strength: verification and validation are more epistemically secure than expert knowledge alone.
Source: §4 · p. 14Trust requires a chain of reliable processes, since the process used to attribute reliability could itself be epistemically opaque or unreliable.
Source: §3.2 · p. 13
Abstract
from arXiv · showhide
Several philosophical issues in connection with computer simulations rely on the assumption that results of simulations are trustworthy. Examples of these include the debate on the experimental role of computer simulations \cite{Parker2009, Morrison2009}, the nature of computer data \cite{Barberousse2013, Humphreys2013}, and the explanatory power of computer simulations \cite{Krohs2008, Duran2017}. The aim of this article is to show that these authors are right in assuming that the results of computer simulations are to be trusted when computer simulations are reliable processes. After a short reconstruction of the problem of \textit{epistemic opacity}, the article elaborates extensively on \textit{computational reliabilism}, a specified form of process reliabilism with computer simulations located at the center. The article ends with a discussion of four sources for computational reliabilism, namely, verification and validation, robustness analysis for computer simulations, a history of (un)successful implementations, and the role of expert knowledge in simulations.
1 Introduction
The article focuses on epistemic opacity, one of four philosophical issues Humphreys identifies for computer simulations. It argues that opacity can persist even with software-engineering practices, and develops a framework for trusting simulation results without dismissing it.
1 Introduction
- The article focuses on epistemic opacity, one of four philosophical issues Humphreys identifies for computer simulations.The other issues are simulation semantics, computational temporal dynamics, and the distinction between in-principle and in-practice limits.Source: §1 · pp. 1–2
- Newman argues that sound software engineering can avoid opacity, but the authors maintain that information hiding can leave agents unable to relate programs to their physical instantiation.Information hiding preserves details that matter at lower-level processing but not in software design and programming contexts.Source: §1 · pp. 2–3
- The authors seek a general solution that acknowledges epistemic opacity without requiring researchers to give up knowledge from computer simulations.They frame this against the question of whether opacity undermines knowledge and whether scientists overstate simulation success.Source: §1 · pp. 2–3
- The article reconstructs opacity through accessibility and surveyability, develops computational reliabilism, and identifies four sources of reliability for simulations.The proposed sources are verification and validation, robustness analysis, histories of successful and unsuccessful implementations, and expert knowledge.Source: §1 · p. 3
2 Epistemic Opacity (EO)
The article reconstructs epistemic opacity as an agent's inability to access and survey the steps of a justification, then distinguishes human-essential opacity from the conditional possibility of justified belief through reliable computational processes.
2.1 What is epistemic opacity?
- The skeptical challenge is how mathematicians justify general claims that exceed what they can inspect directly, a problem the authors connect to computational methods.The analogy is that neither every mathematical instance nor every step in computational processes can be surveyed by a human.Source: §2.1 · p. 4
2.2 Reconstruction
- The reconstruction treats knowing as a matter of accessibility and surveyability rather than committing to a JTB analysis of knowledge.It allows informal and non-deductive justifications and does not require computational processes to be formalized in a particular logic.Source: §2.2 · p. 5
- A process is epistemically opaque to an agent when the agent lacks access to, or cannot survey, all steps of its justification.A justification is surveyable when it is finite in length and checkable in finite time, with the relevant limits depending on the agent.Source: §2.2 · pp. 5–6
2.3 Strengthening EO with Essential Epistemic Opacity
- The authors argue that essential opacity requires restricting possibility to metaphysical or physical possibilities, whose scope depends on contested or theory-specific choices.For physical possibility, the relevant theory must be specified because Newtonian and relativistic possibilities differ.Source: §2.3 · pp. 7–8
- Under the authors’ common-sense reading of possibility, human limits of space, time, and mental capacity make most computational-science processes essentially opaque to humans.This yields the skeptical challenge that, if most such processes are essentially opaque, their results may not be justified beliefs.Source: §2.3 · p. 8
- Computational reliabilism permits justified belief in computational results under reliability conditions by relaxing accessibility and surveyability demands.The authors present it as a response to the skeptical challenge posed by essential epistemic opacity.Source: §2.3 · p. 8
3 Dissolving EEO with computational reliabilism
The section argues that researchers can retain knowledge claims about simulations despite epistemic opacity by grounding trust in computational reliabilism. This account specifies what counts as a reliable simulation process and requires a retrospective chain of reliable sources to support that reliability.
3 Dissolving EEO with computational reliabilism
- Epistemic opacity challenges trust in simulation results because researchers cannot survey every computational step needed to justify them.The section argues that removing trust would also undermine grounds for treating simulations as experimental.Source: §3 · p. 9, §3.2 · p. 13
- Because inspection cannot establish reliability under epistemic opacity, the account appeals to procedures external to the simulation itself.Well-grounded scientific knowledge and trusted assumptions are necessary but not sufficient sources of reliability.Source: §3 · p. 9
3.1 Computational reliabilism
- A simulation counts as reliable when its process tends to produce trustworthy results, making belief in its next results more likely justified than belief in results produced by an unreliable process by luck.Reliability concerns a tendency across results, not success on one occasion.Source: §3.1 · pp. 10–12
- Computational reliabilism specifies process reliabilism for simulations: p concerns a truth-valued proposition about simulation results, and m is the simulation.The proposition may concern particular outcomes or general claims such as correctness for a target system, validity relative to knowledge, or accuracy for intended use.Source: §3.1 · p. 12, §3.2 · p. 14
3.2 Justifying computational reliabilism
- Computational reliabilism requires a retrospective chain of reliable sources because reliability attribution can itself be epistemically opaque.An unreliable pseudo-random generator supplying Schelling-model inputs would undermine trust in the simulation results.Source: §3.2 · pp. 13–14
- The paper leaves unanswered where the retrospective chain of reliable processes ends.The account also allows the possibility of errors in simulation results.Source: §3.2 · pp. 12–13
- Unlike standard process reliabilism, the account first permits reliance on a reliable computational method, then requires reasons for trusting methods used to attribute reliability.This addresses the JJ-principle and reflects scientific and engineering practices such as verification and validation.Source: §3.2 · pp. 13–14
4 Sources for computational reliabilism
The article identifies four sources of computational reliability, which contribute different degrees of warrant. It explains how verification and validation, robustness analysis, implementation history, and expert knowledge support trust in simulation results.
4 Sources for computational reliabilism
- The four sources contribute different degrees of reliability; verification and validation are epistemically stronger than expertise, but their relative degrees are not measured.The article offers an analysis of each source rather than a measurement of its reliability.Source: §4 · p. 14
4.1 Verification and validation
- Verification checks whether the implementation represents the mathematical model, while validation compares simulation results with the empirical world.Together, these methods support trust by checking implementation and by matching results to empirical data.Source: §4.1 · pp. 15, 17
4.2 Robustness analysis for computer simulations
- Robustness analysis varies model structures and looks for shared properties; Ajelli’s agent-based and GLEaM models agreed on epidemic timing, while epidemic size differed by 5% to 10%.The analysis treats the shared property as grounds for reliability, while the size difference is attributed to the metapopulation model’s homogeneity assumption.Source: §4.2 · pp. 18, 20
4.3 A history of (un)successful implementations
- Prior successes and failures in implementation supply independent warrant, although methods evolve and transferring a successful method across domains may introduce unreliability.The authors describe simulation methodology as dynamic and non-hierarchical, but note that models can become methodologically stabilized.Source: §4.3 · p. 21
4.4 Expert knowledge
- Experts contribute theory and implementation knowledge, shape assumptions and acceptable result ranges, and judge relevant similarities in robustness analysis.Expert knowledge can also be a source of errors.Source: §4.4 · pp. 22–23
5 Final remarks
The article argues that epistemic opacity can coexist with reasons for genuine knowledge from computer simulations. It presents computational reliabilism as a formal response while acknowledging that its account of reliability sources remains general.
5 Final remarks
- The proposed response allows epistemic opacity to coexist with reasons for genuine knowledge from computer simulations.The authors frame the article as a formal account addressing the skeptic’s challenge.Source: §5 · pp. 23–24
- The authors present computational reliabilism as novel for taking essential epistemic opacity seriously, addressing process reliabilism’s shortcomings, and systematically reviewing reliability sources.Source: §5 · p. 23
- The account remains general, and more specific work is needed on reliability sources and on how simulation uses affect their assessment.The authors note that climate simulations may rely more on expertise because data scarcity makes validation difficult, without endorsing weaker verification or validation standards.Source: §5 · pp. 23–24