In this paper I argue that Artificial Intelligence and the many data science methods associated with it, such as machine learning and large language models, are first and foremost epistemic technologies. In order to establish this claim, I first argue that epistemic technologies can be conceptually and practically distinguished from other technologies in virtue of what they are designed for, what they do and how they do it. I then proceed to show that unlike other kinds of technology (_including_ other (...) epistemic technologies) AI can be uniquely positioned as an epistemic technology in that it is primarily designed, developed and deployed to be used in _epistemic contexts_ such as inquiry, it is specifically designed, developed and deployed to manipulate _epistemic content_ such as data, _and_ it is designed, developed and deployed to do so particularly through _epistemic operations_ such as prediction and analysis. As has been shown in recent work in the philosophy and ethics of AI (Alvarado, AI and Ethics, 2022a), understanding AI as an epistemic technology will also have significant implications for important debates regarding our relationship to AI technologies. This paper includes a brief overview of such implications, particularly those pertaining to explainability, opacity, trust and even epistemic harms related to AI technologies. (shrink)

We address some of the epistemological challenges highlighted by the Critical Data Studies literature by reference to some of the key debates in the philosophy of science concerning computational modeling and simulation. We provide a brief overview of these debates focusing particularly on what Paul Humphreys calls epistemic opacity. We argue that debates in Critical Data Studies and philosophy of science have neglected the problem of error management and error detection. This is an especially important feature of the epistemology of (...) Big Data. In “Error” section we explain the main characteristics of error detection and correction along with the relationship between error and path complexity in software. In this section we provide an overview of conventional statistical methods for error detection and review their limitations when faced with the high degree of conditionality inherent to modern software systems. (shrink)

Some computational phenomena rely essentially on pragmatic considerations, and seem to undermine the independence of the specification from the implementation. These include software development, deviant uses, esoteric languages and recent data-driven applications. To account for them, the interaction between pragmatics, epistemology and ontology in computational artefacts seems essential, indicating the need to recover the role of the language metaphor. We propose a User Levels (ULs) structure as a pragmatic complement to the Levels of Abstraction (LoAs)-based structure defining the ontology and (...) epistemology of computational artefacts. ULs identify a flexible hierarchy in which users bear their own semantic and normative requirements, possibly competing with the logical specification. We formulate a notion of computational act intended in its pragmatic sense, alongside pragmatic versions of implementation and correctness. (shrink)

How can we be certain that software is reliable? Is there any method that can verify the correctness of software for all cases of interest? Computer scientists and software engineers have informally assumed that there is no fully general solution to the verification problem. In this paper, we survey approaches to the problem of software verification and offer a new proof for why there can be no general solution.

How can we be certain that software is reliable? Is there any method that can verify the correctness of software for all cases of interest? Computer scientists and software engineers have informally assumed that there is no fully general solution to the verification problem. In this paper, we survey approaches to the problem of software verification and offer a new proof for why there can be no general solution.

Digital artifacts often fail to perform as expected. It has recently been argued that this should not be analyzed as software malfunctioning. Rather, every case that is not the result of hardware failures would be due to design errors. This claim, which hinges on the notion of ‘implementation’, highlights a potential fundamental difference between software and other technical artifacts. It also implies that software engineers have more extensive responsibilities than creators of other artifacts. After reconstructing the argument, we show that (...) it rests on two presuppositions regarding the nature of software and the activities of its creators, which we call ‘complete control’ and ‘self-containedness’. Software is taken to be a self-contained unit through which the software engineer exerts complete control over a machine’s behavior. These two presuppositions as well as the usage of implementation and strong claims regarding responsibility can be found throughout the literature on software engineering. Still, we argue that the presuppositions are untenable in light of current practices. For one thing, software engineers cannot exert complete control because software has become too complex for any technique to reliably locate every fault in it. For another, the creation and running of software are nowadays heavily mediated processes involving many agential activities. As a result, the performance of digital artifacts may vary, often in undesirable ways, without either hardware failures or design errors. This undermines the conclusion that software cannot malfunction, while also revealing the complications in developing a metaphysical account of software that fits contemporary practices. (shrink)

The process of completing, correcting and prioritising specifications is an essential but very complex task for the maintenance and improvement of software systems. The preservation of functionalities and the ability to accommodate changes are main objectives of the software development cycle to guarantee system reliability. Logical theories able to fully model such processes are still insufficient. In this paper we propose a full formalisation of such operations on software systems inspired by the Alchourrón–Gärdenfors–Makinson paradigm for belief revision of human epistemic (...) states. We represent specifications as finite sets of formulas equipped with a priority relation that models functional entrenchment of properties. We propose to handle specification incompleteness through ordered expansion, inconsistency through ordered safe contraction and prioritisation through revision with reordering, and model all three in an algorithmic fashion. We show how the system satisfies basic properties of the AGM paradigm, including Levi’s and Harper’s identities. We offer a concrete example and complexity results for the inference and model checking problems on revision. We conclude by describing resilience and evolvability of software systems based on such revision operators. (shrink)

Malware has been around since the 1980s and is a large and expensive security concern today, constantly growing over the past years. As our social, professional and financial lives become more digitalised, they present larger and more profitable targets for malware. The problem of classifying and preventing malware is therefore urgent, and it is complicated by the existence of several specific approaches. In this paper, we use an existing malware taxonomy to formulate a general, language independent functional description of malware (...) as transformers between states of the host system and described by a trust relation with its components. This description is then further generalised in terms of mechanisms, thereby contributing to a general understanding of malware. The aim is to use the latter in order to present an improved classification method for malware. (shrink)

This paper investigates how unsupervised machine learning methods might make hermeneutic interpretive text analysis more objective in the social sciences. Through a close examination of the uses of topic modeling—a popular unsupervised approach in the social sciences—it argues that the primary way in which unsupervised learning supports interpretation is by allowing interpreters to discover unanticipated information in larger and more diverse corpora and by improving the transparency of the interpretive process. This view highlights that unsupervised modeling does not eliminate the (...) researchers’ judgments from the process of producing evidence for social scientific theories. The paper shows this by distinguishing between two prevalent attitudes toward topic modeling, i.e., topic realism and topic instrumentalism. Under neither can modeling provide social scientific evidence without the researchers’ interpretive engagement with the original text materials. Thus the unsupervised text analysis cannot improve the objectivity of interpretation by alleviating the problem of underdetermination in interpretive debate. The paper argues that the sense in which unsupervised methods can improve objectivity is by providing researchers with the resources to justify to others that their interpretations are correct. This kind of objectivity seeks to reduce suspicions in collective debate that interpretations are the products of arbitrary processes influenced by the researchers’ idiosyncratic decisions or starting points. The paper discusses this view in relation to alternative approaches to formalizing interpretation and identifies several limitations on what unsupervised learning can be expected to achieve in terms of supporting interpretive work. (shrink)

Conventional accounts of epistemic opacity, particularly those that stem from the definitive work of Paul Humphreys, typically point to limitations on the part of epistemic agents to account for the distinct ways in which systems, such as computational methods and devices, are opaque. They point, for example, to the lack of technical skill on the part of an agent, the failure to meet standards of best practice, or even the nature of an agent as reasons why epistemically relevant elements of (...) a process may be inaccessible. In this paper I argue that there are certain instances of epistemic opacity— particularly in computational methods such a computer simulations and machine learning processes—that do not arise from, are not responsive to, and are therefore not explained by the epistemic limitations of an agent. I call these instances agent-neutral and agent-independent instances of epistemic opacity respectively. As a result, I also argue that conventional accounts of epistemic opacity offer a limited understanding of the full spectrum of kinds and sources of epistemic opacity, particularly of the kind found in computational methods. In particular, as I will show below, the limitations of these accounts are reflected in the way they fail to provide satisfactory explanations when faced with certain instances of opacity. (shrink)

This paper sheds light on the shift that is taking place from the practice of ‘coding’, namely developing programs as conventional in the software community, to the practice of ‘curing’, an activity that has emerged in the last few years in Deep Learning (DL) and that amounts to curing the data regime to which a DL model is exposed during training. Initially, the curing paradigm is illustrated by means of a study-case on autonomous vehicles. Subsequently, the shift from coding to (...) curing is analysed taking into consideration the epistemological notions, central in the philosophy of computer science, of function, implementation, and correctness. First, it is illustrated how, in the curing paradigm, the functions performed by the trained model depend much more on dataset curation rather than on the model algorithms which, in contrast with the coding paradigm, do not comply with requested specifications. Second, it is highlighted how DL models cannot be considered implementations according to any of the available definitions of implementation that follow an intentional theory of functions. Finally, it is argued that DL models cannot be evaluated in terms of their correctness but rather in their experimental computational validity. (shrink)

We show how more general knowledge can be built in information security, by the building of knowledge of mechanism clusters, some of which are multifield. By doing this, we address in a novel way the longstanding philosophical problem of how, if at all, we come to have knowledge that is in any way general, when we seem to be confined to particular experiences. We also address the issue of building knowledge of mechanisms by studying an area that is new to (...) the mechanisms literature: the methods of what we shall call mechanism discovery in information security. This domain offers a fascinating novel constellation of challenges for building more general knowledge. Specifically, the building of stable communicable mechanistic knowledge is impeded by the inherent changeability of software, which is deployed by malicious actors constantly changing how their software attacks, and also by an ineliminable secrecy concerning the details of attacks not just by attackers, but also by information security defenders as they protect their methods from both attackers and commercial competitors. We draw out ideas from the work of the mechanists Darden, Craver, and Glennan to yield an approach to how general knowledge of mechanisms can be painstakingly built. We then use three related examples of active research problems from information security to develop philosophical thinking about building general knowledge using mechanisms, and also apply this to develop insights for information security. We show that further study would be instructive both for practitioners and for philosophers. (shrink)

This paper addresses the methodological problem of analysing what it is to explain observed behaviours of engineered computing systems, focusing on the crucial role that abstraction and idealization play in explanations of both correct and incorrect BECS. First, it is argued that an understanding of explanatory requests about observed miscomputations crucially involves reference to the rich background afforded by hierarchies of functional specifications. Second, many explanations concerning incorrect BECS are found to abstract away from descriptions of physical components and processes (...) of computing systems that one finds below the logic circuit and gate layer of functional specification hierarchies. Third, model-based explanations of both correct and incorrect BECS that are provided in the framework of formal verification methods often involve idealizations. Moreover, a distinction between restrictive and permissive idealizations is introduced and their roles in BECS explanations are analysed. (shrink)

In this paper, I argue that machines can create works of art. My argument is based on an analysis of the so-called creative machines and focuses on technical functions and intentions. If my proposal is correct, then creative machines are technical artifacts with the proper function to bring about works of art. My account is based on sensible conceptual connections between makers, technical artifacts, intentions, and the creation of art. One upshot of the account presented here is that we do (...) not need a new conceptual framework or dubious assumptions about artistic agency on part of machines in order to arrive at the conclusion that creative machines make art. I will conclude the paper with some remarks regarding the artistic value of items produced by creative machines. (shrink)

The standard rule of single privative modification replaces privative modifiers by Boolean negation. This rule is valid, for sure, but also simplistic. If an individual a instantiates the privatively modified property (MF) then it is true that a instantiates the property of not being an F, but the rule fails to express the fact that the properties (MF) and F have something in common. We replace Boolean negation by property negation, enabling us to operate on contrary rather than contradictory properties. (...) To this end, we apply our theory of intensional essentialism, which operates on properties (intensions) rather than their extensions. We argue that each property F is necessarily associated with an essence, which is the set of the so-called requisites of F that jointly define F. Privation deprives F of some but not all of its requisites, replacing them by their contradictories. We show that properties formed from iterated privatives, such as being an imaginary fake banknote, give rise to a trifurcation of cases between returning to the original root property or to a property contrary to it or being semantically undecidable for want of further information. In order to determine which of the three forks the bearers of particular instances of multiply modified properties land upon we must examine the requisites, both of unmodified and modified properties. Requisites underpin our presuppositional theory of positive predication. Whereas privation is about being deprived of certain properties, the assignment of requisites to properties makes positive predication possible, which is the predication of properties the bearers must have because they have a certain property formed by means of privation. (shrink)

The birth, growth, stabilization and subsequent understanding of a new field of practical and theoretical enquiry is always a conceptual process including several typologies of events, phenomena an...

We present a review of Raymond Turner’s Book Computational Artifacts – Towards a Philosophy of Computer Science, focusing on three main topics: Design, Malfunction, and Validity.

This paper contributes to the computer ethics debate on software ownership protection by examining the ontological, methodological, and ethical problems related to property right infringement that should come prior to any legal discussion. The ontological problem consists in determining precisely what it is for a computer program to be a copy of another one, a largely neglected problem in computer ethics. The methodological problem is defined as the difficulty of deciding whether a given software system is a copy of another (...) system. And the ethical problem corresponds to establishing when a copy constitutes, or does not constitute, a property right infringement. The ontological problem is solved on the logical analysis of abstract machines, and the latter are argued to be the appropriate level of abstraction for software at which the methodological and the ethical problems can be successfully addressed. (shrink)

Software-intensive science challenges in many ways our current scientific methods. This affects significantly our notion of science and scientific interpretation of the world, driving at the same time the philosophical debate. We consider some issues prompted by SIS in the light of the philosophical categories of ontology and epistemology.

One might poetically muse that computers have the essence both of logic and machines. Through the case of the history of Separation Logic, we explore how this assertion is more than idle poetry. Separation Logic works because it merges the software engineer’s conceptual model of a program’s manipulation of computer memory with the logical model that interprets what sentences in the logic are true, and because it has a proof theory which aids in the crucial problem of scaling the reasoning (...) task. Scalability is a central problem, and some would even say the central problem, in appli- cations of logic in computer science. Separation Logic is an interesting case because of its widespread success in verification tools. For these two senses of model—the engineering/conceptual and the logical—to merge in a genuine sense, each must maintain their norms of use from their home disciplines. When this occurs, both the logic and engineering benefit greatly. Seeking this intersection of two different senses of model provides a strategy for how computer scientists and logicians may be successful. Furthermore, the history of Separation Logic for analysing programs provides a novel case for philosophers of science of how software engineers and computer scientists develop models and the components of such models. We provide three contributions: an exploration of the extent of models merging that is necessary for success in computer science; an introduction to the technical details of Separation Logic, which can be used for reasoning about other exhaustible resources; and an introduction to the problems, process, and results of computer scientists for those outside the field. (shrink)

Machine awareness is a disputed research topic, in some circles considered a crucial step in realising Artificial General Intelligence. Understanding what that is, under which conditions such feature could arise and how it can be controlled is still a matter of speculation. A more concrete object of theoretical analysis is algorithmic iteration for computational intelligence, intended as the theoretical and practical ability of algorithms to design other algorithms for actions aimed at solving well-specified tasks. We know this ability is already (...) shown by current AIs, and understanding its limits is an essential step in qualifying claims about machine awareness and Super-AI. We propose a formal translation of algorithmic iteration in a fragment of modal logic, formulate principles of transparency and faithfulness across human and machine intelligence, and consider the relevance to theoretical research on -AI as well as the practical import of our results. (shrink)

In this paper, I offer a contribution to the debate on testimony that rests on three elements: the definition of semantic information, the analysis of trust as a second?order property of first?order relations, and Floridi?s Network Theory of Account (NTA). I argue that testimony transmits semantic information and it is neither grounded on trust nor justified by it. Instead, I show that testimony is an occurrence of a first?order relation of communication affected by the second?order property of trust. I then (...) defend the view that an epistemic agent can acquire some knowledge, on the basis of the information communicated through testimony, if and only if the agent is able to connect the transmitted information to the conceptual network of interrelation to which it belongs. I refer to Floridi?s NTA to show how such a network allows the epistemic agent to achieve knowledge on the basis of semantic information. (shrink)

The reviewers Rapaport, Stephanou, Angius, Primiero, and Bringsjord of Turner cover a broad range of topics in the philosophy of computer science. They either challenge the positions outlined in Turner or offer a more refined analysis. This article is a response to their challenges.

The reviewers Rapaport, Stephanou, Angius, Primiero, and Bringsjord of Turner cover a broad range of topics in the philosophy of computer science. They either challenge the positions outlined in Turner or offer a more refined analysis. This article is a response to their challenges.