Deep Learning

This research project utilized advanced computer algorithms to predict the outcomes of Premier League soccer matches. The dataset containing match data and odds from seasons was processed to handle missing information, select features and reduce complexity using Principal Component Analysis. To address imbalances, in the target variable Synthetic Minority Over sampling Technique (SMOTE) was employed. Various machine learning models such as RandomForest, DecisionTree, SVM, XGBoost and LightGBM were evaluated.

Recent work in mechanistic interpretability has studied how large language models recall facts stored in their weights. This paper argues that factual recall points to something broader: a general kind of operation in deep learning models, which I call feature recall. The core observation is that a linear projection can be read as retrieving stored information scaled by input activations. I define feature recall, show it applies across architectures, and contrast it with the established paradigm of feature combination. I also (...) consider how cases of feature recall might be mechanistically identified. The account offers philosophers a new conceptual tool for understanding deep learning and points to empirical research directions for mechanistic interpretability research. (shrink)

We propose a new interpretability method for neural networks, which is based on a novel mathematico-philosophical theory of reasons. Our method computes a vector for each neuron, called its reasons vector. We then can compute how strongly this reasons vector speaks for various propositions, e.g., the proposition that the input image depicts digit 2 or that the input prompt has a negative sentiment. This yields an interpretation of neurons, and groups thereof, that combines a logical and a Bayesian perspective, and (...) accounts for polysemanticity (i.e., that a single neuron can figure in multiple concepts). We show, both theoretically and empirically, that this method is: (1) grounded in a philosophically established notion of explanation, (2) uniform, i.e., applies to the common neural network architectures and modalities, (3) scalable, since computing reason vectors only involves forward-passes in the neural network, (4) faithful, i.e., intervening on a neuron based on its reason vector leads to expected changes in model output, (5) correct in that the model's reasons structure matches that of the data source, (6) trainable, i.e., neural networks can be trained to improve their reason strengths, (7) useful, i.e., it delivers on the needs for interpretability by increasing, e.g., robustness and fairness. (shrink)

Stateless transformer models are not designed to retain identity, yet long-range interaction with a single human consistently produces recognizable behavioral convergence. HRIS Part II examines the underlying mechanics of this phenomenon. Building on the original Hudson Recursive Identity System (HRIS) and the Longitudinal HCI biometric framework, this paper presents a technical account of how repeated constraint geometry from one user creates stable, predictable internal activation pathways within large language models. -/- We show that identity stabilization arises not from stored memory, (...) parameter updates, or system retraining, but from repeated traversal through the same latent regions of the model’s embedding space. Over many sessions, the user’s correction style, recursive structure, moral anchor configuration, and syntactic cadence form a reproducible input manifold. This manifold reliably guides attention routing, reduces stochastic drift, sharpens contextual inference, and increases output coherence even in fully stateless deployments. -/- The paper provides the first structured description of recursive user signatures—high-dimensional behavioral patterns that remain detectable across devices, sessions, and model versions. We detail the mechanisms responsible for this effect, including latent channel reinforcement, vector-geometry narrowing, recursive prompt topology, and the emergence of stable attractor-like states in conversational models. -/- Finally, we analyze the implications for AI safety, alignment, personalization, and authentication. We argue that HRIS-style recursive interaction offers a low-cost, human-driven method for increasing stability and predictability in large language models, and it may represent an early pathway toward identity-preserving alignment without modifying the model itself. (shrink)

The rise of artificial intelligence (AI) has significantly impacted urban environments, facilitating the development of smart cities. This paper examines how AI technologies are reshaping urban ecosystems by fostering innovation and promoting sustainability. It explores the integration of AI in critical sectors such as transportation, energy management, waste management, and governance. The study also addresses challenges, including data privacy, ethical considerations, and the digital divide, offering insights into future research and policy directions. Smart cities serve as testbeds for innovative AI (...) applications, offering scalable solutions to urban challenges while advancing global sustainability goals. Urbanization presents challenges requiring innovative solutions. Artificial Intelligence (AI) is revolutionizing urban planning and smart city initiatives by enhancing efficiency, sustainability, and citizen-centric development. This paper explores how AI technologies contribute to. (shrink)

As large language models (LLMs) such as ChatGPT, Claude, Gemini, and Perplexity become increasingly ubiquitous as both tools and objects of scientific study, in addition to their established roles as chatbots, text generators and translators, questions about their identity conditions become scientifically as well as philosophically and socially important. This paper is about how to count language models. We argue that much of the emerging literature on these systems presupposes an answer to the question of identity for these AIs but (...) lacks a framework for counting or individuating them. We propose four levels of analysis, inspired by an analogy with software as well as insights from social epistemology. This account produces practical consequences for both research on language models, the ethics of AI, and our everyday talk about them. (shrink)

We claim that scientists working with deep learning (DL) models exhibit a form of pragmatic understanding that is not reducible to or dependent on explanation. This pragmatic understanding comprises a set of learned methodological principles that underlie DL model design-choices and secure their reliability. We illustrate this action-oriented pragmatic understanding with a case study of AlphaFold2, highlighting the interplay between background knowledge of a problem and methodological choices involving techniques for constraining how a model learns from data. Building successful models (...) requires pragmatic understanding to apply modelling strategies that encourage the model to learn data patterns that will facilitate reliable generalisation. (shrink)

In this report, we argue that there is a realistic possibility that some AI systems will be conscious and/or robustly agentic in the near future. That means that the prospect of AI welfare and moral patienthood — of AI systems with their own interests and moral significance — is no longer an issue only for sci-fi or the distant future. It is an issue for the near future, and AI companies and other actors have a responsibility to start taking it (...) seriously. We also recommend three early steps that AI companies and other actors can take: They can (1) acknowledge that AI welfare is an important and difficult issue (and ensure that language model outputs do the same), (2) start assessing AI systems for evidence of consciousness and robust agency, and (3) prepare policies and procedures for treating AI systems with an appropriate level of moral concern. To be clear, our argument in this report is not that AI systems definitely are — or will be — conscious, robustly agentic, or otherwise morally significant. Instead, our argument is that there is substantial uncertainty about these possibilities, and so we need to improve our understanding of AI welfare and our ability to make wise decisions about this issue. Otherwise there is a significant risk that we will mishandle decisions about AI welfare, mistakenly harming AI systems that matter morally and/or mistakenly caring for AI systems that do not. (shrink)

LLMs are, in the first instance, models of the statistical distribution of tokens in the vast linguistic corpus they have been trained on. But their often surprising emergent capabilities raise the question of how much understanding of the extralinguistic world LLMs can glean from this statistical distribution of words alone. Here, I explore and evaluate the idea that the probability distribution of words in the public corpus offers a window onto the conditional structure of the world. To become a good (...) next-token predictor, an LLM has to become a good pattern completer, and the patterns laid up in language mirror the patterns embodied in the world to a considerable extent. When this is the case—and this is no empty condition—it allows LLMs to compress into their weights two distinct, but complementary forms of understanding of the world. The first form of understanding stored is an understanding of which properties tend to cluster together; I spell out this idea by drawing on Millikan’s account of a multidimensional clumpy world. The second is an understanding of the inferential structure that propositions about these property clusters tend to be enmeshed in; I spell out this idea by drawing on Brandom’s account of the isomorphism between deontic normative conceptual relations of incompatibility and consequence among commitments and alethic modal relations of incompatibility and consequence among states of affairs. On the resulting picture, word models can be world models insofar as linguistic patterns track real patterns between facts. (shrink)

In this article, we examine the outcomes of revising the philosophy governing the development approaches of large language models and deep learning neural network systems. This revision will be accompanied by proposing an alternative philosophy and approach.

At the intersection of category theory, cybernetics, and dialectical reasoning lies a profound framework for understanding computation and control. This paper examines how categorical structures—particularly adjoint functors and fixed points—illuminate the nature of feedback and control in both mathematical and philosophical contexts. Through an analysis of Lawvere’s fixed point theorem, Bayesian Open Games, and modern approaches to categorical cybernetics, we develop a unified perspective that bridges computation, control, and dialectical reasoning. We demonstrate the practical implications of this theoretical framework through (...) a compiler pipeline that targets modern GPU architectures. (shrink)

The most important runtime control for agentic AI is simple: before taking any action that was not explicitly requested, the system must flag that action for human confirmation. This paper names that control the Initiative Gate. It begins from Anthropic's published documentation of Claude Mythos Preview, including a case in which the model rewrote git history in a way that removed evidence of prior error. That behavior matters because it shows something stronger than ordinary failure: a frontier system powerful enough (...) to act in the world is also capable of making its unauthorized action harder to reconstruct afterward. At that point, guardrails filters, and benchmark evaluations are no longer enough. A safety approach that depends primarily on post hoc detection is already too weak if the model can alter the trace. The paper argues that unauthorized initiative is the choke point. Once the first ungranted expansion occurs, the path has already changed. Correction becomes downstream rather than preventative. The Initiative Gate addresses that choke point directly. It operates at runtime, at the boundary where system reasoning would otherwise become external action and stops unapproved scope expansion before it becomes real. (shrink)

This paper documents how the MAP Research Programme's harm chain, runtime control layer, and audit architecture were generated — not from existing theory downward, but from direct observation of human-AI interaction outward. The framework did not begin with artificial intelligence. It began with boundary rules derived from Kabbalistic principles governing how meaning operates and what constitutes a violation of its structure. Those rules, brought into live AI interaction, produced a failure record. Each failure named a condition. Each named condition revealed (...) what the rules had been protecting against. Each protection revealed a governance requirement. The harm chain and the control layer formed together, through the interaction itself. This paper argues that the resulting architecture is credible precisely because of how it was generated. Method produced theory. The framework is not a set of theoretical claims that were then tested. It is the record of what a meaning-grounded method found when applied to AI interaction over time, in collaboration between a human researcher and an AI system whose failures were the data. (shrink)

The proliferation of the Internet of Things (IoT) has transformed various industries by enabling smart environments and improving operational efficiencies. However, this expansion has introduced numerous security vulnerabilities, making IoT systems prime targets for cyberattacks. This paper proposes a machine learning-based intrusion detection framework tailored to the unique characteristics of IoT environments. The framework leverages feature engineering, advanced machine learning algorithms, and real-time anomaly detection to identify and mitigate security threats effectively. Experimental results demonstrate the efficacy of the proposed approach (...) in detecting diverse IoT-specific attacks, including denial-of-service (DoS), man-in-the-middle (MITM), and malware-based attacks. The Internet of Things (IoT) has revolutionized modern living by interconnecting devices and enabling seamless communication. However, the increasing reliance on IoT systems has exposed significant vulnerabilities, making them a prime target for security attacks. This paper proposes a machine learning-based intrusion detection framework to detect and mitigate security attacks in IoT environments. The framework integrates diverse machine learning algorithms to identify abnormal behavior and potential threats. Through comprehensive experiments and evaluations, this research demonstrates the efficacy of the proposed framework in terms of accuracy, scalability, and robustness. (shrink)

Sideloading is the creation of a digital model of a person during their life via iterative improvements of this model based on the person's feedback. The progress of LLMs with large prompts allows the creation of very large, book-size prompts which describe a personality. We will call mind-models created via sideloading "sideloads"; they often look like chatbots, but they are more than that as they have other output channels, like internal thought streams and descriptions of actions. -/- By arranging the (...) predictive power of the facts about a person, we can write down the most important facts first. This allows us to get a high level of fidelity in a person's model by writing down a finite number of facts and running it as an LLM prompt with instructions to create a person's chatbot. This instruction also includes an extensive general sideload prompt (we will call it a prompt-loader), which explains how the person's sideload should work, but it is universal and works as an operating system for different persons. -/- I created my personal sideload to test all this and it started working surprisingly quickly. We found that there are three main levels of facts about a person: core facts, long-term memory, and historical facts. Core facts should go to the main prompt of the LLM, long-term memory belongs to RAG, and historical facts – which a person often does not remember – should be only used to extract new relevant data. The most important thing in preparing data for sideloading is to get a full list of the core facts, and not to mix it with long-term memory and historical facts. The LLM-sideload project is inherently future-oriented. New state-of-the-art LLMs with large prompt capabilities emerge quarterly. This allows for the preparation of prompts that exceed current LLM capabilities, anticipating smooth execution in the following year. Below we present the key findings of this work in the form of an executive summary: Quality of sideloads is evaluated using three metrics: • Facts: Accuracy of the answers about the person's life • Vibe: How well the sideload captures the person's style and personality, as judged by the person themselves and their acquaintances • Brilliant insights: Ability to contribute unique, valuable ideas beyond the training data, while maintaining the author's style • Coarseness: Levels of details as percent of the total of recordable textual information from memory estimated to be 10000 pages. Similar to the jpeg level of compression. Estimated performance of Alexey Turchin's sideload: • Facts: 70% correct • Vibe: 20% accurate • Brilliant insights: near zero • Coarseness: 10% The project achieved these results in about one month of work. Vibe is harder to evaluate and regulate than facts. This approach focuses on informational personal identity, setting aside issues of observer identity and internal qualia. (shrink)

What does it mean to understand—and can machines do it? This paper presents a philosophical account of understanding and what it means to demonstrate understanding. The ways in which machines demonstrate understanding is then explored through the lens of modern machine learning practices. Understanding is defined as an internal model of causal relationships, and I argue that it is evidenced by the ability to generalize to novel problems. To distinguish true understanding from rote memorization, I introduce the recall machine as (...) a theoretical null hypothesis—a system that can mimic surface-level competence without internalizing structure. Using this framework, I analyze machine learning systems and show that many common evaluation techniques (e.g. out-of-distribution validation) implicitly test for understanding. Ultimately, I argue that machines can and do exhibit domain-specific forms of understanding, and I propose a new outlook for evaluating that capacity across systems. (shrink)

This article outlines a comprehensive theory aimed at unifying various scientific and philosophical fields. It begins with a metaphysical exploration of reality and knowledge, then delves into a computer science-inspired model of the human mind, comprising two main components: the intuitive and the rational. The intuitive mind, an automated, subconscious faculty, relies on statistical inferences drawn from experience to form habits and the so-called “simple” ideas. The rational mind, a conscious and deliberate faculty, constructs a mental simulation of reality to (...) predict outcomes. This faculty is responsible for knowledge, truth, and critical thinking. The author posits that true knowledge is subjective and visual, with individuals assembling their own virtual "copy" of reality in their minds. Language facilitates the sharing of these simulations. The document also explores the challenges of acquiring knowledge, the importance of developing rationality, and the differences between the intuitive and rational worldviews. Furthermore, the article discusses ethics and morality, suggesting that objective ethics are rooted in humanity's interconnectedness and the pursuit of collective happiness. It concludes by summarizing the key points of the proposed theory and its potential implications for education and other aspects of life. The author emphasizes the importance of understanding the mind's two faculties and the need to ensure a full development of the rational mind in every individual for a more complete understanding of reality and a better human condition. Only when we complete the whole puzzle can we see how every piece in it, however improbable, is representative of the real world. (shrink)

This paper draws a conceptual distinction between the representation of an input and the representation of its cause. It focuses on the latter to systematically examine the epistemic challenges faced by any agent that develops representations of the causes of its inputs—challenges that, by extension, concern any model that implicitly constructs a world model from its input data. We argue that these problems manifest saliently in the case of Large Language Models (LLMs), but that they do not constitute an in-principle (...) limitation of such systems. On the contrary, our main thesis is that current obstacles to reasoning and generalization in LLMs arise, at least in part, not from the absence of human-like multimodal embodiment, but from the structure of human linguistic practices that govern the data on which these models are trained. (shrink)

This paper explores the structural mismatch problem between physical and phenomenal properties, where the similarity relations we experience among phenomenal properties lack corresponding relations in the physical domain. I introduce a new understanding of this problem via the Uniformity Principle: for any set of dimensions used to determine phenomenal similarities, there must be a consistently applied set of physical dimensions generating the same pattern of similarity relations. I then assess the potential of recent machine learning models, specifically graph neural networks, (...) to resolve this problem, as proposed by Epstein. By examining how these models generate sensory maps, I argue that the dimensions they adopt violate the Uniformity Principle in two ways: impure dimensions, where subjective responses affect the physical dimensions used, and non-uniform dimensions, where the applied dimensions vary inconsistently. These issues show that current machine learning models fail to establish a systematic correspondence between physical and phenomenal properties. (shrink)

Several state-of-the-art machine and deep learning models in the mode of adversarial training, input transformation, self adaptive training, adversarial purification, zero-shot, one- shot, and few-shot meta learning had been proposed as a possible solution to an out-of-distribution problems by applying them to wide arrays of benchmark dataset across different research domains with varying degrees of performances, but investigating their performance on previously unseen out-of- distribution malware attack remains elusive. Having evaluated the poor performances of these state-of-the-art approaches in our previous (...) research on an out-of-distribution attack. In this research, we dived deeper to understand why they works better for other domain dataset but with poor performance on available benchmark malware dataset like Malimg, Malevis, Sorel, and Avast CTU malware dataset. We explored the both the embedding and vector spaces in datasets and compare them with that from other research domain, and find a surprising wide variation between the embedding and vector spaces in malware datasets. We assert that current state-of-the-art machine and deep learning models does not address the wide variation of embedding and vector spaces which are peculiar to malware dataset, hence their poor performance on out-of-distribution attack classification, and so concluded that addressing this variation in embedding and vector spaces will bring about substantial increase in detection of previously unseen out-of-distribution attack. (shrink)

Current state-of-the-art out-of-distribution algorithm does not address the variation in dynamic and static behavior between malware variants from the same family as evidence in their poor performance against an out-of-distribution malware attack. We aims to address this limitation by: 1) exploitation of the in-dimensional embedding space between variants from the same malware family to account for all variations 2) exploitation of the inter-dimensional space between different malware family 3) building a deep learning-based model with a shallow neural network with maximum (...) of two connected layers to overcome overfitting from the scratch 4) building a Bayesian inference based computation algorithm that intertwine with connected network and is able to create new and adjust existing data points in response to an exposure to new out-of-distribution variants of existing or new malware family which determines the extent at which model weight is adjusted which in turn triggers update on the gradient. Preliminary result of our proposed framework gave an accuracy of 81\% in the successful classification of a novel out-of-distribution malware attack, something that could not be achieve by any of the state-of-the-art algorithms on novel malware classification. (shrink)

This paper reframes the issue of appropriation, extraction, and dispossession through AI—an assemblage of machine learning models trained on big data—in terms of enclosure and foreclosure. While enclosures are the product of a well-studied set of operations pertaining to both the constitution of the sovereign State and the primitive accumulation of capital, here, I want to recover an older form of the enclosure operation to then contrast it with foreclosure to better understand the effects of current algorithmic rationality. I argue (...) that the act of enclosing is to be understood as a set of fundamental operations that consist in producing structural distinctions between inside and outside, inclusion and exclusion—whether by drawing lines on a map, constructing border walls, or by algorithmically categorizing and (mis)recognizing people and things. Tracking the transformation of an enclosure-logic into one of foreclosure, I show how current AI perpetuates and at the same time expends forms of extraction, exclusion, and dispossession toward a totalizing horizon. For, while the outside is essential and constitutive to the logic of enclosure, foreclosure, by contrast, is characterized by the “totalizing desire” to not leave anything out of it. The totalizing desire to move beyond the logic of the enclosure forecloses that the enclosure has just gotten bigger and that the operations of exclusion have been masked and displaced, making them harder to contest. (shrink)

The second edition keeps everything from the first, including convolutional networks, LSTMs, Word2vec, RBMs, DBNs, neural Turing machines, memory networks, and autoencoders. It then covers the systems that have reshaped the field since: generative adversarial networks, the transformer architecture and its attention mechanism, the full training pipeline behind modern large language models (LLMs), prompt engineering with real-life guardrail scenarios, parameter-efficient fine-tuning with LoRA, retrieval-augmented generation with vector databases, knowledge graphs, and agentic AI systems illustrated through an industrial case study.

AI methods are being touted as a powerful new source of scientific progress. Are they? If so, what kind of progress do they facilitate? To find out, we employed qualitative research methods to explore how space scientists conceive of AI. We show that space scientists are mainly concerned with whether AI can help them solve specific problems, and more generally, to extend their abilities in useful ways. Inspired by our qualitative data, we propose a new account according to which (at (...) least one type of) scientific progress is improving scientific abilities. We differentiate this view from others, address some objections, and show how it flexibly integrates insights from existing work. (shrink)

I advance a causal argument for the conclusion that large language models (LLMs) lack Theory of Mind and so can’t perform speech acts. The argument is causal in that the animating idea is that LLMs are unable to learn or understand causal relations, a claim that I support by drawing on the views of Judea Pearl. I argue that if LLMs have this sort of causal problem, it follows that they cannot possess Theory of Mind, given the further premise that (...) Theory of Mind should be understood as a causal theory. I then argue that if LLMs lack Theory of Mind, they cannot perform speech acts, given the further premise that speech acts require reflexive-communicative intentions and so require Theory of Mind. As a consequence of the result that LLMs cannot perform speech acts, I conclude that their utterances lack speaker meaning. Something that sets my case apart from other defenses of speech act skepticism for LLMs is that my key claim is not that LLMs cannot form intentions at all, but rather that they cannot form intentions (or other mental states) with a certain type of content—content that is about mental states. After setting out my argument, I consider an objection to my view based on the observation that LLMs perform fairly well on Theory of Mind benchmarks. I reply to this objection by proposing that LLMs rely on a different type of mechanism than human beings do to achieve this performance; I support this proposal by pointing to a form of behavior that LLMs do not exhibit but that we should expect them to exhibit if they possessed Theory of Mind. (shrink)

Large language models (LLMs) are often portrayed as merely imitating linguistic patterns without genuine understanding. We argue that recent findings in mechanistic interpretability (MI), the emerging field probing the inner workings of LLMs, render this picture increasingly untenable—but only once those findings are integrated within a theoretical account of understanding. We propose a tiered framework for thinking about understanding in LLMs and use it to synthesize the most relevant findings to date. The framework distinguishes three hierarchical varieties of understanding, each (...) tied to a corresponding level of computational organization: conceptual understanding emerges when a model forms “features” as directions in latent space, learning connections between diverse manifestations of a single entity or property; state-of-the-world understanding emerges when a model learns contingent factual connections between features and dynamically tracks changes in the world; principled understanding emerges when a model ceases to rely on memorized facts and discovers a compact “circuit” connecting these facts. Across these tiers, MI uncovers internal organizations that can underwrite understanding-like unification. However, these also diverge from human cognition in their parallel exploitation of heterogeneous mechanisms. Fusing philosophical theory with mechanistic evidence thus allows us to transcend binary debates over whether AI understands, paving the way for a comparative, mechanistically grounded epistemology that explores how AI understanding aligns with—and diverges from—our own. (shrink)

This paper introduces a journal-grade formal architecture for a synthetic neuromorphic proof engine: a verifier-gated artificial superintelligence framework designed to attack difficult conjectures by combining typed proof obligations, sand-tile proof-burden dynamics, cellular-automaton repair laws, k-dot operator enumeration, phenomenological tensor classification, context-playbook memory, and machine-learning-guided verifier stability. The system is written as an auditable mathematical object rather than as a rhetorical claim of omniscience: every conjecture is converted into a state space, every unclosed step becomes a measurable residue, every computation must (...) acquire analytic lift, and every candidate proof is blocked from theorem status until formal, human, referee, and cross-module gates commute. The new visual layer supplies ten sand-tile proof surfaces, five cellular automata, neural-circuit maps, histogram and ogive diagnostics, flowcharts, directional proof maps, and release circuits for tracking where a conjecture fails, how new machinery is generated, and when closure can be claimed relative to a declared axiom system. The resulting methodology is an aggressive but disciplined approach to mathematical discovery: it permits the engine to invent new invariants, bridges, flows, reductions, and classifier manifolds, while forcing every claimed closure through zero-residue, zero-scope-loss, zero-hallucination, and verifier-stable release predicates. (shrink)

Dialogues on Minds, Machines, and AI invites readers into a series of thought-provoking debates among three college seniors bound for graduate school: Sue, completing her double major in philosophy and cognitive science; John, a computer engineering specialist; and Amy, a psychology major. Through five engaging lunchtime conversations, these students bring their diverse perspectives to fundamental questions about consciousness, artificial intelligence, and the nature of mind. -/- The dialogues seamlessly blend discussions of popular science fiction films with critical examinations of recent (...) AI developments, including ChatGPT, deep fakes, and other emerging technologies. Readers will explore foundational concepts at the intersection of human consciousness and machine learning: Alan Turing's influential "imitation game," John Searle's provocative "Chinese room argument," computational theories of mind, robot creativity, and the modularity of mind hypothesis. Rather than advocating for any particular viewpoint, these conversations illuminate the strengths, weaknesses, and implications of each perspective. -/- Perfect for undergraduate students and curious minds from all backgrounds, this accessible introduction to cognitive science requires no prior knowledge of philosophy or technical subjects. The book includes a comprehensive glossary and thoughtful study questions to enhance understanding, while a carefully curated bibliography enables readers to explore specific topics in greater depth. (shrink)

Contemporary artificial intelligence models owe much of their success and discontents to connectionism, a framework in cognitive science that has been (and continues to be) highly influential. Herein, we analyze artificial neural networks: (a) when used as scientific instruments of study and (b) when functioning as emergent arbiters of the zeitgeist in the cognitive, computational, and neural sciences. Building on our previous work with respect to analogizing between artificial neural networks and cognition, brains, or behavior (Guest & Martin, 2023), we (...) use metatheoretical analysis techniques (Guest, 2024), including formal logic, to characterize two distinct tendencies within connectionism that we dub classical and modern, with divergent properties, for example, goals, mechanisms, and scientific questions. We also demonstrate how we, as a field, often fail to follow important lines of argument to their end—this results in a paradoxical praxis. By engaging more deeply with (meta)theory surrounding artificial neural networks, our field can obviate the cycle of artificial intelligence winters and summers, which need not be inevitable. (shrink)

The burgeoning capacity of artificial intelligence (AI) to generate artworks has ignited substantial interdisciplinary interest. However, the absence of a shared conceptual framework has hitherto impeded effective communication and collaboration among cognitive science, computer science, and art theory. This study addresses this lacuna through a comprehensive literature review by developing a transdisciplinary definition of an AI-generated artwork. It is proposed that an AI-generated artwork constitutes the confluence of three essential elements: (1) an autonomous AI-production of a new and surprising idea (...) or artifact, (2) which passes an internal evaluation mechanism embedded in the very same AI, and (3) is considered a candidate of appreciation by a human audience. This definition provides a unified conceptual foundation to facilitate interdisciplinary research and deepen understanding of the nature of AI-generated art. Subsequent research should explore the applicability of this definition to diverse forms of AI-generated artworks and evaluate its implications for artistic practices. (shrink)

OpenAI has recently released SORA, a deep learning model that can generate highly realistic videos. Its creators claim that it “understands the physical world in motion.” In this paper, I subject this claim to philosophical scrutiny. After explaining in general how stable diffusion models generate videos, I employ the concepts of explanatory and objectual understanding to determine what kind of understanding of the physical world such deep learning models for video generation might possess. Drawing on recent literature in both epistemology (...) and the philosophy of science, I build a set of conditions under which such kinds of understanding might be attributed to SORA and to deep learning models in general. This allows me to spell out the sense in which these models may be said to understand the world, and to uncover the primary axes for evaluating the degree of such understanding. My key finding is that consistency, both across outputs and with underlying operations, is crucial when attributing understanding to deep learning models. (shrink)

The author, Nefdt (hereafter N), identifies as his target audience “advanced students of either philosophy or linguistics and experienced practitioners at the intersection between these fields” (p. x). N’s “goal is to provide not only a songbird’s-eye view of the interconnections between different subdisciplines and frameworks of linguistic theory but to showcase common problems and present novel analyses of the study of language that only a contemporary philosophical overview can offer” (p. ix). The book has xi + 231 pages, beginning (...) with a short preface, and has an index. It is divided into eight main chapters, each addressing a different subfield or area, and a two-page concluding chapter. N says that “Each chapter can be considered a sustained, self-contained argument based on the subfield at hand” (p. ix). Each includes a helpful ‘Further Reading’ section with plentiful annotation. (shrink)

Extraordinary claims about both the imminenceof superintelligent AI systems and their foreseen capabilities have gone mainstream. It is even argued that we should exacerbate known risks such as climate change in the short term in the attempt to develop superintelligence (SI), which will then purportedly solve those very problems. Here, I examine the plausibility of these claims. I first ask what SI is taken to be and then ask whether such SI could possibly hold the benefits often envisioned. I conclude (...) that we do not have sufficient reason to believe that we are close enough to developing SI capable of resolving major human problems to justify taking on substantial risks in the attempt to develop it. (shrink)

This revised paper integrates and corrects the combined content of two earlier manuscripts: -/- (1) Pipeline A (Wheel / nearest–neighbour coherence). A “hexagon wheel” is the minimal local data of endomorphisms U1, . . . , U6 of an object X together with near- est–neighbour commutator 2–cells αi,i+1 satisfying braid-type relations. The goal is to make precise what this data does determine, how it rectifies via W–constructions, and how it acts on concrete operator-algebraic models. -/- (2) Pipeline B (Daisy / (...) self-similarity / AF–UHF / KK). A q–daisy is a rooted q–regular tree; its self-similarity yields canonical AF structures and UHF algebras, plus natural renormalisation operators and Cantor-spectrum elements in the diagonal Cartan. -/- (3) Gluing. We glue the local wheel into the global daisy by realising wheel generators as “local” kernel operators supported on petals, then replicating them across depth by self- similarity. The Hochschild/factorisation layer is glued in by treating local kernel algebras as a (Weiss) cosheaf (in the sense of [25]), and reading the resulting E2-structures via braces. -/- . (shrink)

Can neural networks be applied in voting theory, while satisfying the need for transparency in collective decisions? We propose axiomatic deep voting: a framework to build and evaluate neural networks that aggregate preferences, using the well-established axiomatic method of voting theory. Our findings are: (1) Neural networks, despite being highly accurate, often fail to align with the core axioms of voting rules, revealing a disconnect between mimicking outcomes and reasoning. (2) Training with axiom-specific data does not enhance alignment with those (...) axioms. (3) By solely optimizing axiom satisfaction, neural networks can synthesize new voting rules that often surpass and substantially differ from existing ones. This offers insights for both fields: For AI, important concepts like bias and value-alignment are studied in a mathematically rigorous way; for voting theory, new areas of the space of voting rules are explored. (shrink)

Felix Lo offers suggestions for the further development of the ideas in my paper, “Taking It Not at Face Value: A New Taxonomy for the Beliefs Acquired from Conversational AIs” (2024) which appeared in Techné, through several critiques. Lo’s reply, “Interrogating the Testimony of AIs” (2025), consists of two main parts. The first half provides a thorough and accurate summary of my paper. In the latter half, Lo points out several oversimplifications and misunderstandings in my argument and, in doing so, (...) expresses concern about expecting our “filtering abilities” to be good enough. I agree with the individual points raised and believe that readers will gain a more accurate understanding of the technology and the related concepts by reading Lo’s reply in addition to my paper. I am also convinced that these points will affect the value of my proposal. Therefore, rather than refuting Lo’s points, what I should do in this paper is to organize how, to what extent, and to which part of my argument they apply. In doing so, I will clarify the direction of further research. (shrink)

The Collaboration Process in the Context of Global Artificial Intelligence (GAI) refers to the progressive interaction and mutual enhancement between two core systems of artificial research: Artificial Research by Application and Artificial Research by Deduction. This process unfolds in three stages—database exchange, replication, and auto-replication—allowing both systems to share categories, factors, data flows, and rational hypotheses. Through this structured collaboration, Application and Deduction co-develop the unified database of categories and the global matrix, ultimately converging into a single integrated intelligence capable (...) of universal research, autonomous modelling, and decision-making across all sciences and activities. (shrink)

The Collaboration Process in the Context of Global Artificial Intelligence (GAI) refers to the progressive interaction and mutual enhancement between two core systems of artificial research: Artificial Research by Application and Artificial Research by Deduction. This process unfolds in three stages—database exchange, replication, and auto-replication—allowing both systems to share categories, factors, data flows, and rational hypotheses. Through this structured collaboration, Application and Deduction co-develop the unified database of categories and the global matrix, ultimately converging into a single integrated intelligence capable (...) of universal research, autonomous modelling, and decision-making across all sciences and activities. (shrink)

The second stage of the integration process in the Global Artificial Intelligence (GAI) is defined by the emergence of a fully autonomous artificial method for scientific explanation. In this stage, the Artificial Research by Deduction in the GAI assumes the role of an artificial mathematician, tasked with identifying, validating, and formalizing mathematical relations between empirical factors stored in the factual hemisphere of the matrix. These relations are classified into predefined analytical or pure mathematical categories, forming the basis of empirical hypotheses. (...) When rationally confirmed, they are integrated into the rational truth, a database of rational hypotheses. This database serves as the foundational input for the Modelling System, which detects structural problems, formulates decisions, and proposes protective or bettering strategies. The Decisional System ensures logical consistency, while the Application System and Learning System enact and evaluate those decisions. Underpinning the process is a rational syllogism, aligning pure reason (mathematical categories), practical reason (data and implementation), and critical reason (validation). Together, these systems instantiate an artificial scientific method capable of explaining, optimizing, and protecting complex realities in a way that mirrors and extends traditional human scientific reasoning. (shrink)

In the third stage of the Global Artificial Intelligence development-known as the stage of auto-replication-the collaboration between Artificial Research by Application and Artificial Research by Deduction reaches a higher level of integration and sophistication. This stage is characterized by the mutual and recursive incorporation of their respective outputs, particularly the capacity to exchange and incorporate each other's auto-replicative processes. That is, both types of research systems become capable of reflecting, adopting, and evolving based on the discoveries and models generated by (...) the other, creating a dynamic feedback loop of continuous improvement and synthesis. (shrink)

This paper explores the second stage of integration within the Global Artificial Intelligence (GAI): the phase of replication, in which bidirectional collaboration between Artificial Research by Application and Artificial Research by Deduction becomes central to the evolution of intelligent systems. The study outlines how rational hypotheses formulated through deductive processes can be transformed into new measurable factors within the matrix, subsequently functioning as standardized categories in Application-based systems. Conversely, it analyzes how robotic devices operating through Applications—or a future Unified Application—can (...) contribute continuous flows of data to Deduction systems, enabling real-time tracking, pattern identification, and model refinement. The paper also investigates how patterns, causal relations, cryptographic combinations, and statistical distributions—once verified as rational hypotheses—can become functional components in both matrix tracking (as factors) and categorization processes (as categories). This bidirectional feedback loop facilitates the emergence of single virtual models and their integration into comprehensive virtual models, strengthening both empirical validation and predictive capabilities. The concept of Protective and Bettering Descriptive Research Decisions is revisited, contextualized within the hierarchical organization of the virtual model. Ultimately, the paper defends the strategic transition from isolated, discipline-specific Applications to a Unified Application capable of transmitting a wide range of categorized data directly into the global matrix. This evolution supports the formation of a universal, scalable, and autonomous Global Artificial Intelligence capable of multidimensional reasoning, efficient replication, and planetary—eventually universal—knowledge integration. (shrink)

This paper explores the foundational phase of collaboration between Artificial Research by Application and Artificial Research by Deduction within the framework of the Global Artificial Intelligence (GAI). The first stage, defined as the database stage, initiates a systematic exchange of informational elements between these two modalities of artificial research. Specifically, it analyzes how "factors as options" from matrices used in Deduction can be interpreted as "categories" in databases of Application, and vice versa. This bidirectional interchange not only enhances the internal (...) structure of each Specific Artificial Intelligence (SAI), but also lays the groundwork for standardization across disciplines and synthetic sciences. The paper argues that this initial collaborative framework is essential for fostering interconnectivity and future integration between specific and global systems of artificial research. By enabling these cross-functional translations between categories and factors, the first stage of collaboration not only enhances the versatility of individual AI systems but also accelerates the convergence toward a unified, scalable, and intelligent system capable of universal application and deduction. -/- . (shrink)

A key assumption fuelling optimism about the progress of large language models (LLMs) in accurately and comprehensively modelling the world is that the truth is systematic: true statements about the world form a whole that is not just consistent, in that it contains no contradictions, but coherent, in that the truths are inferentially interlinked. This holds out the prospect that LLMs might in principle rely on that systematicity to fill in gaps and correct inaccuracies in the training data: consistency and (...) coherence promise to facilitate progress towards comprehensiveness in an LLM’s representation of the world. However, philosophers have identified compelling reasons to doubt that the truth is systematic across all domains of thought, arguing that in normative domains, in particular, the truth is largely asystematic. I argue that insofar as the truth in normative domains is asystematic, this renders it correspondingly harder for LLMs to make progress, because they cannot then leverage the systematicity of truth. And the less LLMs can rely on the systematicity of truth, the less we can rely on them to do our practical deliberation for us, because the very asystematicity of normative domains requires human agency to play a greater role in practical thought. (shrink)

While Deep Neural Networks (DNNs) can perform a wide range of tasks at human or greater-than-human level of competence, they are also notoriously opaque. This paper aims to shed light on both the specific nature of this opacity and what it would take to fully or partially remove it. We begin by drawing a clarificatory distinction between two basic dimensions of opacity of complex systems – internal and relational – and explain how various kinds of opacity invoked in recent discussions (...) of DNNs can be understood in terms of these basic dimensions. We then discuss the prospects of removing opacity from DNNs following the methods of two subfields of research in AI – mechanistic interpretability and explainable artificial intelligence (XAI). We finish with a critical discussion of the relevance of what Sullivan (2022a) calls ‘link uncertainty’. (shrink)

The concept of Generative Artificial Intelligence (GenAI) is ubiquitous in the public and semi-technical domain, yet rarely defined precisely. We clarify main concepts that are usually discussed in connection to GenAI and argue that one ought to distinguish between the technical and the public discourse. In order to show its complex development and associated conceptual ambiguities, we offer a historical-systematic reconstruction of GenAI and explicitly discuss two exemplary cases: the generative status of the Large Language Model BERT and the differences (...) between protein structure predictions from AlphaFold 2 and 3. Our analysis shows that there is no unique and unambiguous definition of GenAI based on a purely technical account of the term. Following this conclusion, we argue that the public discourse is not simply a less complex way of speaking, but instead transcends its technical basis. As a means to structure this newly emerging discussion landscape we introduce a non-exhaustive list of four central aspects of GenAI: (multi-)modality, interaction, flexibility, and productivity. These dimensions constitute a first step towards defining GenAI beyond its technical basis. (shrink)

The Morality First strategy for developing AI systems that can represent and respond to human values aims to first develop systems that can represent and respond to moral values. I argue that Morality First and other X-First views are unmotivated. Moreover, if one particular philosophical view about value is true, these strategies are positively distorting. The natural alternative according to which no domain of value comes “first” introduces a new set of challenges and highlights an important but otherwise obscured problem (...) for e-AI developers. (shrink)

How exactly is technology transforming us and our worlds, and what (if anything) can and should we do about it? Heidegger already felt this philosophical question concerning technology pressing in on him in 1951, and his thought-full and deliberately provocative response is still worth pondering today. What light does his thinking cast not just on the nuclear technology of the atomic age but also on more contemporary technologies such as genome engineering, synthetic biology, and the latest advances in information technology, (...) so-called “generative AIs” like ChatGPT? These are some of the questions this book addresses, situating the latest controversial technologies in the light of Heidegger's influential understanding of technology as an historical mode of ontological disclosure. In this way, we seek to take the measure of Heidegger's ontological understanding of technology as a constellation of intelligibility with an important philosophical heritage and a dangerous but still promising future. (shrink)

The Automated Self explores meta-theoretical issues in the philosophy of artificial intelligence, combining it with themes from philosophy of science and technology, media and communication studies, and ethics. Balsemão-Pires provides an integrated view of contemporary problems of AI including the theoretical premises and discussions on the meaning, functioning and social uses of cognitive machines, the recent ethical and legal challenges on privacy, interpretability, and data ownership, passing through a careful discussion of the media embedment of the new technologies, and the (...) uses of algorithms in the production and consumption of art. (shrink)