As machine learning has gradually entered into ever more sectors of public and private life, there has been a growing demand for algorithmic explainability. How can we make the predictions of complex statistical models more intelligible to end users? A subdiscipline of computer science known as interpretable machine learning (IML) has emerged to address this urgent question. Numerous influential methods have been proposed, from local linear approximations to rule lists and counterfactuals. In this article, I highlight three conceptual challenges that (...) are largely overlooked by authors in this area. I argue that the vast majority of IML algorithms are plagued by (1) ambiguity with respect to their true target; (2) a disregard for error rates and severe testing; and (3) an emphasis on product over process. Each point is developed at length, drawing on relevant debates in epistemology and philosophy of science. Examples and counterexamples from IML are considered, demonstrating how failure to acknowledge these problems can result in counterintuitive and potentially misleading explanations. Without greater care for the conceptual foundations of IML, future work in this area is doomed to repeat the same mistakes. (shrink)

One major fault line in foundational theories of cognition is between the so-called “representational” and “non-representational” theories. Is it possible to formulate an intermediate approach for a foundational theory of cognition by defining a conception of representation that may bridge the fault line? Such an account of representation, as well as an account of correspondence semantics, is offered here. The account extends previously developed agent-based pragmatic theories of semantic information, where meaning of an information state is defined by its interface (...) role, to a theory that accommodates a notion of representation and correspondence semantics. It is argued that the account can be used to develop an intermediate approach to cognition, by showing that the major sources of tension between “representational” and “non-representational” theories may be eased. (shrink)

This article contributes to the debate around the abilities of large language models such as GPT-3, dealing with: firstly, evaluating how well GPT does in the Turing Test, secondly the limits of such models, especially their tendency to generate falsehoods, and thirdly the social consequences of the problems these models have with truth-telling. We start by formalising the recently proposed notion of reversible questions, which Floridi & Chiriatti propose allow one to ‘identify the nature of the source of their answers’, (...) as a probabilistic measure based on Item Response Theory from psychometrics. Following a critical assessment of the methodology which led previous scholars to dismiss GPT’s abilities, we argue against claims that GPT-3 completely lacks semantic ability. Using ideas of compression, priming, distributional semantics and semantic webs we offer our own theory of the limits of large language models like GPT-3, and argue that GPT can competently engage in various semantic tasks. The real reason GPT’s answers seem senseless being that truth-telling is not amongst them. We claim that these kinds of models cannot be forced into producing only true continuation, but rather to maximise their objective function they strategize to be plausible instead of truthful. This, we moreover claim, can hijack our intuitive capacity to evaluate the accuracy of its outputs. Finally, we show how this analysis predicts that a widespread adoption of language generators as tools for writing could result in permanent pollution of our informational ecosystem with massive amounts of very plausible but often untrue texts. (shrink)

This paper addresses the problem of upgrading functional information to knowledge. Functional information is defined as syntactically well-formed, meaningful and collectively opaque data. Its use in the formal epistemology of information theories is crucial to solve the debate on the veridical nature of information, and it represents the companion notion to standard strongly semantic information, defined as well-formed, meaningful and true data. The formal framework, on which the definitions are based, uses a contextual version of the verificationist principle of truth (...) in order to connect functional to semantic information, avoiding Gettierization and decoupling from true informational contents. The upgrade operation from functional information uses the machinery of epistemic modalities in order to add data localization and accessibility as its main properties. We show in this way the conceptual worthiness of this notion for issues in contemporary epistemology debates, such as the explanation of knowledge process acquisition from information retrieval systems, and open data repositories. (shrink)

In current philosophy of information, different authors have been supporting the veridicality thesis (VT). According to this thesis, an epistemically-oriented concept of information must have truth as one of its necessary conditions. Two challenges can be raised against VT. First, some philosophers object that veridicalists erroneously ignore the informativeness of false messages. Secondly, it is not clear whether VT can adequately explain the information considered in hypothetical reasoning. In this sense, logical diagrams offer an interesting case of analysis: by manipulating (...) a logical diagram we can verify that a certain conclusion follows from a set of premises, but it cannot help us to determine the actual truth-value of a given set of propositions. Focusing on the latter challenge, in this paper I claim that logical diagrams set out potential counterexamples to VT and, consequently, pose a real challenge to this thesis. First, a veridicalist analysis of logical diagrams requires the assumption of metatheoretical properties which are not satisfied by some logical systems (and, consequently, are not satisfied by some systems of logical diagrams). So, VT does not fit well as a general framework for a theory of logical diagrammatic information. Secondly, based on semantic inferentialism, one can propose a normative interpretation of the inferential content of logical diagrams not exposed to the problems faced by VT. Moreover, there are several reasons to believe that veridicalism cannot accommodate such a normative interpretation. In other words, normativism represents a real (though still underexplored) alternative to veridicality. Due to these reasons, I conclude that, until further research, we should adopt a more parsimonious standpoint and say that logical diagrams provide inferential information simply. (shrink)

In this paper I argue that, according to a particular physicalist conception of information, information is both alethically neutral or non-alethic, and is intrinsically semantic. The conception of information presented is physicalist and reductionist, and is contrary to most current pluralist and non-reductionist philosophical opinion about the nature of information. The ontology assumed for this conception of information is based upon physicalist non-eliminative ontic structural realism. However, the argument of primary interest is that information so construed is intrinsically semantic on (...) a reductionist and non-alethic basis where semantic content is constituted by indication along causal pathways. Similar arguments have been presented by philosophers with respect to representation. I suggest the conception of information that I present is correct by the lights of the best applied mathematical and scientific theories of information. If so, there is no need for any separate theory of semantic information. Thus I present a theory of intrinsically semantic information which also constitutes an informational theory of truth where truth reduces to information. In the last section I discuss weakly and strongly semantic information, and reject them in favour of alethically neutral intrinsically semantic information. (shrink)

I will analyse Floridi’s rejection of digital ontologies and his positive proposal for an informational structural realism. I intend to show that ISR is still fundamentally a digital ontology, albeit with some different metaphysical commitments to those that Floridi rejects. I will argue that even though Floridi deploys the method of levels of abstraction adapted from computer science, and has established a Kantian transcendentalist conception of both information and structure, ISR still reduces to a discretised binary, and therefore digital, ontology. (...) The digital ontologies that Floridi rejects are John Wheeler’s “It from Bit” conception and computational metaphysics. They’re rejected predominantly on the basis that they rely upon a false dichotomy between digital discrete and continuous metaphysics. ISR involves a Kantian transcendentalist conception of de re relations that is intended to avoid this false dichotomy. However, I’ll argue that the binary, discrete, digital component of digital ontology is retained in ISR, and therefore ISR is still a digital ontology since its conception of information reduces to binary discrete de re relations. As such, ISR comes down on one side of the rejected ontic dichotomy of digital metaphysics, and so an informational metaphysics that is not a digital ontology is still a promissory note in the philosophy of information. (shrink)

There are many ways of understanding the nature of philosophical questions. One may consider their morphology, semantics, relevance, or scope. This article introduces a different approach, based on the kind of informational resources required to answer them. The result is a definition of philosophical questions as questions whose answers are in principle open to informed, rational, and honest disagreement, ultimate but not absolute, closed under further questioning, possibly constrained by empirical and logico-mathematical resources, but requiring noetic resources to be answered. (...) The article concludes with a discussion of some of the consequences of this definition for a conception of philosophy as the study (or “science”) of open questions, which uses conceptual design to analyse and answer them. (shrink)

Three classic distinctions specify that truths can be necessary versus contingent,analytic versus synthetic, and a priori versus a posteriori. The philosopher reading this article knows very well both how useful and ordinary such distinctions are in our conceptual work and that they have been subject to many and detailed debates, especially the last two. In the following pages, I do not wish to discuss how far they may be tenable. I shall assume that, if they are reasonable and non problematic (...) in some ordinary cases, then they can be used in order to understand what kind of knowledge the maker’s knowledge is. By this I mean the sort of knowledge that Alice enjoys when she holds the information that Bob’s coffee is sweetened because she just put two spoons of sugar in it herself. The maker’s knowledge tradition is quite important but it is not mainstream in modern and analytic epistemology and lacks grounding in terms of exactly what sort of knowledge one is talking about. My suggestion is that this grounding can be provided by a minimalist approach, based on an information-theoretical analysis. In the article, I argue that we need to decouple a fourth distinction, namely informative versus uninformative, from the previous three and, in particular, from its implicit association with analytic versus synthetic and a priori versus a posteriori; such a decoupling facilitates, and is facilitated by, moving from a monoagent to a multiagent approach: the distinctions qualify a proposition, a message, or some information not just in themselves but relationally, with respect to an informational agent; the decoupling and the multiagent approach enable a re-mapping of currently available positions in epistemology on these four dichotomies; within such a re-mapping, two positions, capturing the nature of a witness’ knowledge and of a maker’s knowledge, can best be described as contingent, synthetic, a posteriori, and uninformative and as contingent, synthetic, weakly a priori, and uninformative respectively. In the conclusion, I indicate why the analysis of the maker’s knowledge has important consequences in all those cases in which the poietic intervention on a system determines the truth of the model of that system. (shrink)

: This article provides replies to, and comments on, the contributions to the special issue on the philosophy of information. It seeks to highlight con‐vergences and points of potential agreement, while offering clarifications and further details. It also answers some criticisms and replies to some objections articulated in the special issue.

The article addresses the problem of how semantic information can be upgraded to knowledge. The introductory section explains the technical terminology and the relevant background. Section 2 argues that, for semantic information to be upgraded to knowledge, it is necessary and sufficient to be embedded in a network of questions and answers that correctly accounts for it. Section 3 shows that an information flow network of type A fulfils such a requirement, by warranting that the erotetic deficit, characterising the target (...) semantic information t by default, is correctly satisfied by the information flow of correct answers provided by an informational source s. Section 4 illustrates some of the major advantages of such a Network Theory of Account (NTA) and clears the ground of a few potential difficulties. Section 5 clarifies why NTA and an informational analysis of knowledge, according to which knowledge is accounted semantic information, is not subject to Gettier-type counterexamples. A concluding section briefly summarises the results obtained. (shrink)

Contemporary science seems to be caught in a strange predicament. On the one hand, it holds a firm and reasonable commitment to a healthy naturalistic methodology, according to which explanations of natural phenomena should never overstep the limits of the natural itself. On the other hand, contemporary science is also inextricably and now inevitably dependent on ever more complex technologies, especially Information and Communication Technologies, which it exploits as well as fosters. Yet such technologies are increasingly “artificialising” or “denaturalising” the (...) world, human experiences and interactions, as well as what qualifies as real. So the search for the ultimate explanation of the natural seems to rely upon, and promote, the development of the artificial, seen here as an instantiation of the non-natural. In this article, I would like to try and find a way out of this apparently strange predicament. I shall argue that the naturalisation of our knowledge of the world is either philosophically trivial, or mistaken. First, I shall distinguish between different kinds of naturalism. Second, I shall remind the reader that the kinds of naturalism that are justified today need to be protected and supported pragmatically, but they are no longer very interesting conceptually. We know how to win the argument. We just have to keep winning it. Whereas the kind of naturalism that is still interesting today is now in need of revision in order to remain acceptable. Such a kind of naturalism may be revised on the basis of a realistic philosophy of information, according to which knowing is a constructive activity, through which we do not merely represent the phenomena we investigate passively, but create more or less correct informational models of them, proactively and interactively. I shall conclude that the natural is in itself artefactual, and that the information revolution is disclosing a tension not between the natural and the non-natural, but a deeper one between a user’s and a producer’s interpretation of knowledge. The outcome is a philosophical view of knowledge and science in the information age that may be called constructionist and a revival of philosophy as a classic, foundationalist enterprise. (shrink)

Luciano Floridi’s informational structural realism takes a constructionist attitude towards the problems of epistemology and metaphysics, but the question of the nature of the semantical component of his view remains vexing. In this paper, I propose to dispense with the semantical component of ISR completely. I outline a Syntactical version of ISR. The unified entropy-based framework of information has been adopted as the groundwork of SISR. To establish its realist component, SISR should be able to dissolve the latching problem. We (...) have to be able to account for the informational structures–reality relationship in the absence of the standard semantical resources. The paper offers a pragmatic solution to the latching problem. I also take pains to account for the naturalistic plausibility of this solution by grounding it in the recent computational neuroscience of the predictive coding and the free energy principle. (shrink)

Research in information systems includes a range of approaches which make varied contributions in terms of knowledge, understanding, or practical developments. In these days of “fake news” and spurious internet content, scholarly research needs to be able to demonstrate its validity – are its finding true, or its recommendations correct? We argue that there are fundamental validation criteria that can be applied to all research approaches despite their apparent diversity and conflict. These stem from current views of the nature of (...) truth, and the related but wider concept correctness, within philosophy. There has been much debate about the nature of truth – is it correspondence, coherence, consensual or pragmatic? Current debates revolve around the idea of a pluralist view of truth – that there are different forms of truth depending on the context or domain. Related to truth is the wider concept of correctness – propositions may be true but correctness can also be applied to actions, performances or behavior for which truth is not appropriate. We develop a framework for research validity and apply it to a range of research forms including positivist, interpretive, design science, critical and action-oriented. The benefits are: i) a greater and more explicit focus on validity criteria will produce better research; ii) having a single framework can provide some commonality between what at times seem conflicting approaches to research; iii) having criteria made explicit should encourage debate and further development. The framework is applied to a variety of empirical papers employing varied research approaches. (shrink)

This chapter addresses two questions. First, if knowledge is accounted information, how are we supposed (to apply this analysis in order) to understand perceptual knowledge and knowledge by testimony? In the first part of the chapter, I articulate an answer in terms of a re-interpretation of perception and testimony as data providers rather than full-blown cases of knowledge. Second, if perception and testimony are correctly understood as data providers, how are we supposed (to apply this analysis in order) to understand (...) the semantic value of the data provided by such processes? In the second part of the chapter, I argue in favour of a constructionist hypothesis about how data may become meaningful for human cognitive agents through a process of repurposing of natural data/signals. The conclusion of the chapter is that human agents are natural-born data hackers. (shrink)