Browsing by Author "Nefdt, Ryan"

Now showing 1 - 5 of 5
  • No Thumbnail Available
    Item
    Open Access
    Epistemic opacity: a feature not a bug: an exploration into the relationship between brains and ANNs
    (2025) Schoeman, Keldt; Nefdt, Ryan
    AI in the 21st century has come to be dominated by one school in particular, connectionism. And its successes are all around us – in the media we consume, in the music we listen to, in the cold calls we receive, etc. While this school was founded by psychologists, logicians, and philosophers with the goal of replicating human-level intelligence, the field has undergone a drastic transformation in recent years, entering a paradigm which is now dominated by engineering goals. Within this new paradigm, connectionism is no longer characterized as a field modelling the brain, but rather a mere engineering tool with incredible powers of pattern recognition. However, while the move to employ connectionist AI as a tool has led to remarkable successes in a variety of fields, it has also come with issues such as the black box problem, or epistemic opacity. Within a strictly engineering paradigm, attempts to explain the internal reasoning of these networks remain unsatisfying. Therefore, I propose recoupling connectionist networks with their roots in brain modelling, which would in turn open rich, new explanations for problems like epistemic opacity. Simply put, when we place the problem of opacity within the context of brain modelling, it appears that it may not be a problem at all, but an emergent feature of a complex system. In other words, we are beginning to have difficulty understanding modern connectionist networks in much the same manner we struggle to understand brains. Hence, it might well be feature, not a bug, that these systems should disappear into the mists of complexity.
  • Thumbnail Image
    Item
    Open Access
    Examining Evidence for Passive in Yoruba
    (2021) Balogun, Bunmi; Nefdt, Ryan; Mesthrie, Rajend
    The construction of English passives is formed by placing the noun which denotes the subject of the action in the sentence to the object position and then making the object of the sentence dislocate to the subject position. Other transformations include the change of the auxiliary verb and the inflection of the main verb. The focus of this study however is on Yoruba passives, a topic that is very passive in the Yoruba grammar literature. The study's primary aim then is to investigate the passives in Yoruba. No study to the best of my knowledge has lay claim of the existence of passives in Yoruba, a Kwa language under the Niger-Congo language family which is spoken mostly in the Western part of Nigeria. This study provides evidence to argue for the existence of (forms of) passives in the language. In the realization of Yoruba passives, the subject in the active verb is physically dislocated to become an agent phrase of the passive sentence or deleted. However, the object's position in particular differs strikingly from the English passive as the object of the active verb does not move to the subject position. The pronoun a occupies the subject position, and it is in [Spec, TP] as the object makes no movement. In Yoruba, the auxiliary of the passive is not a consistent form in that auxiliaries change based on the sentence they are used in. Similarly, the Yoruba verbs are not morphologically inflected. Hence, the verb of the active sentence retains its form in the passive. I speculate that object DP cannot fulfil the requirement of T's EPP, hence its inability to move. The analysis of the study will be done in comparison to English passive constructions and selected languages that have peculiar cases of passives and analyses will be represented in tree diagrams. The paper is cast within the Minimalist Program of syntax.
  • No Thumbnail Available
    Item
    Open Access
    Fully moral artificial agents: future or fiction
    (2025) Moore, Jaimee; Nefdt, Ryan
    The rapid technological progress has resulted advancements in technology that includes the development of Artificially Intelligent agents (AI) for various purposes. The extensive progress in AI has reached a point where the introduction of AI agents as android robots or humanoids into society is plausible. This highlights the evolving relationship between humans and technology. This thesis explores the multidimensional aspects of artificial intelligence (AI) by examining the intersection of machine ethics, ethical theories, and deep learning. The central focus is on assessing the compatibility of ethical theories with the rapid advancements in technology, particularly the potential development of fully moral artificial agents. The aim of this thesis is to address the ethical concerns associated with the evolution of AI, particularly the introduction of artificially intelligent agents into various societal roles, such as care robots in healthcare. The need for an impartial approach to address these concerns is identified, leading to the proposal of machine ethics as a framework. Machine ethics, defined as ensuring moral behaviours in AI, provides a basis for evaluating the ethical capabilities of artificially intelligent agents. As technology continues to progress, the demand for a comprehensive ethical framework for AI decision-making becomes increasingly apparent. Machine ethics offers insights into the ethical processes involved in AI decision-making, allowing a closer examination of computational abilities and ethical capacities. The primary inquiry revolves around whether AI agents can achieve full morality and if existing ethical theories can govern their behaviour effectively. To explore these questions, the thesis draws upon three ethical theories—Utilitarianism, Deontology, and African Ubuntu Ethics—and their applicability to the development of fully moral artificial agents. It employs deep learning as a critical component in understanding moral agency within the context of AI. The analysis unfolds in four parts: first, providing a background on AI concepts; second, examining recent AI progress and the potential for ethical AI agents; third, exploring the future of AI in relation to morality through ethical theories; and fourth, establishing criteria for considering AI agents as moral agents. In conclusion, the thesis argues that, with the advancements in technology and the insights provided by machine ethics and ethical theories, the development of fully moral artificial intelligent agents is possible.
  • Thumbnail Image
    Item
    Open Access
    The metaphysics of time investigations in tense-logic and a B-series semantics
    (2010) Nefdt, Ryan
    The view that time flows from the future to the present and then recedes into the past is both natural and deeply problematic. So called 'A-theories' about time claim that this is the fundamental nature of temporality. This is not the view which will be defended in this paper. Rather I shall argue for what is known as a 'B-theory' analysis of time and language, one in which the relations of 'earlier than' and 'later than' are necessary and sufficient for any analysis of time. The structure of this paper is tripartite. The first part will address the philosophical and metaphysical tenets of both the A and B-theories of time. In this section, McTaggart's 1908 argument for the unreality of time will be presented along with some objections to it. I will show that some seemingly convincing arguments against 'McTaggart's paradox' are unsuccessful and his paradox does indeed call the A-series into question. This section will lay the groundwork for further discussions relating to other disciplines which are concerned with this debate by describing the issues and points of tension.
  • No Thumbnail Available
    Item
    Open Access
    On naturalistic saltation genealogies of content: an exploration of continuity & discontinuity in nonreductive diachronic explanations of content by Adam H. Schroeder supervised by University of Cape Town
    (2024) Schroeder, Adam; Weiss, Bernhard; Nefdt, Ryan
    Explanations involving the relationship between naturalism, normativity and intentional content are often plagued with difficulty. Incorporating any two in an explanation usually leads to challenges with integrating the third. My thesis aims to explore this tension, by considering an ambitious version of naturalistic explanation for the emergence of normative intentional content. I wish to cast doubt on this explanation by showing how it fails to alleviate the aforementioned tension. In fact, I will argue that this explanation is inconsistent owing to the inimical relationship between naturalism, normativity and intentional content. The explanation I will be considering is one amongst a variety of naturalistic explanations for solving the placement problem, i.e., how to locate intentional content, or items involving intentional content, “in a world exhaustively characterized in terms of the … collective posits of the … sciences.” 1 Traditionally, the answer to this location problem has been pursued by trying to naturalise content. In general, this involves providing some reductive relation between content and accepted natural facts in the hope of demonstrating that content can be understood in completely naturalistic non-contentful terms. This strategy has faced several difficulties in relation to normativity such as the disjunction problem and gerrymandering objection inter alia. As a prophylactic measure to deal with these difficulties, a strategy has recently come into vogue which forgoes attempts to provide purely reductive explanations of content; rather, it aims to explain how it is possible for content to nonreductively emerge in the natural world. In other words, it aims to explain the natural origins of content, rather than naturalise content. I name accounts fitting this strategy Naturalistic Saltation Genealogies of Content. Despite the benefits this strategy affords in avoiding the perennial objections naturalisation projects face in relation to normativity, my aim is to show that it is susceptible to its own set of difficulties due to the tension between naturalism, normativity and intentional content. More specifically, my aim is to show that the central assumptions of this strategy are inconsistent, and as a result, entail that they are discontinuous explanations of content. Alternatively put, these genealogical explanations cannot succeed in answering the placement problem. A consequence of this aim will be that if one is committed to the continuity of a naturalistic saltation genealogy of content, then this can be shown to implicitly entail the reduction of the normative to the nonnormative or the use of some non-naturalistic resources in explanation. The aim and consequence of my argumentation can by captured by the following slogan: (S) Naturalistic saltation genealogies of content are either discontinuous explanations or implicitly entail the reduction of the normative to the nonnormative or the rejection of naturalism. This is the same as saying that: (S*) Naturalistic saltation genealogies of content are either discontinuous explanations or self-defeating. Whereby ‘self-defeating' I mean that proponents of these genealogies unwittingly revert to other strategies for solving the placement problem – strategies that naturalistic saltation genealogies of content were precisely aimed at avoiding. This slogan will be demonstrated to be a product of the structural objection named the Continuity-Discontinuity Regress Argument. I.e., for every continuous naturalistic saltation genealogy, there is a Discontinuity Argument against it; and for every Discontinuity Argument there must be a sub-continuous naturalistic saltation genealogy constructed in response. This sets off a regress which results in an infinite regress of location problems.