Sandro Skansi

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.

This paper examines the relationship between two influential theories in cognitive science and philosophy of mind: Fodor’s modularity of mind hypothesis and the Physical Symbol System Hypothesis (PSSH) proposed by Newell and Simon. Although modularity is traditionally framed as a theory of cognitive architecture and the PSSH as a computational account of intelligence, this paper argues that the distinction between them is largely conceptual rather than substantive. The central claim is that modularity and the PSSH are, in effect, reformulations of the same un-derlying theoretical commitment. The analysis begins with a critical review of Fodor’s defining criteria for modules, such as domain specificity, mandatory operation, informational encapsula-tion, and shallow outputs. While these conditions were intended to provide an empirically gro-unded account of mental organization, many of them remain vague or metaphorical. Neverthe-less, taken together, they imply that cognitive processes must be decomposable into functionally distinct components with standardized input–output relations. Such decomposability presupposes the manipulation of structured representations. From this, the paper derives its first key claim: any modular cognitive process necessarily involves symbolic processing. Modules can only func-tion if their outputs are formatted in a way that allows systematic recombination and communica-tion with other components. This requirement aligns directly with the notion of a physical symbol system. Modularity, therefore, implicitly assumes the symbolic ontology articulated explicitly by the PSSH. The argument then proceeds in the opposite direction. If the PSSH is correct in clai-ming that a physical symbol system is sufficient for general intelligent action, then general intelli-gence must be implementable via organized subsystems that operate on specific classes of sym-bols. At the functional level, these subsystems correspond to modules. This establishes a bidirec-tional equivalence: modularity entails symbolic processing, and symbolic processing entails mo-dular realizability. The paper concludes that long-standing debates opposing modularity to sym-bolic approaches rest on a false dichotomy, and that recognizing their equivalence offers a clearer framework for understanding both human and artificial intelligence.

Computational complexity has often been ignored in the philosophy of mind, in philosophical artificial intelligence studies. The purpose of this paper is threefold. First and foremost, to show the importance of complexity rather than computability in philosophical and AI problems. Second, to rephrase the notion of computability in terms of solvability, i.e., treating computability as non-sufficient for establishing intelligence. The Church-Turing thesis is therefore revisited and rephrased in order to capture the ontological background of spatial and temporal complexity. Third, to emphasize ontological differences between different time complexities, which seem to provide a solid base for a better understanding of artificial intelligence in general.

In this paper we propose an non-machine learning artificial intelligence (AI) based approach for telecom data analysis, with a special focus on clique detection. Clique detection can be used to identify households, which is a major challenge in telecom data analysis and predictive analytics. Our approach does not use any form of machine learning, but another type of algorithm: satisfiability for propositional logic. This is a neglected approach in modern AI, and we aim to demonstrate that for certain tasks, it may be a good alternative to machine learning-based approaches. We have used a simple DPLL satisfiability solver over an artificially generated telecom dataset (due to GDPR regulations), but our approach can be implemented on any telecom data by following the SAT encoding we have developed, and the DPLL solver can be substituted by a more advanced alternative such as CDCL. This paper extends the method presented in [1] for banking logs to data containing caller information, and proposes a more efficient encoding.

In this paper, through the analysis of the division of different scientific fields, we deal with the nature of mathematics as a scientific discipline. Through the historical analysis of the division of science, but also the analysis of the nature of mathematics and the ontological status of the objects that mathematics deals with, we show that the now-established divisions among scientific fields are the result of social circumstances and that mathematics itself is closer to the humanities than the natural sciences.

Machine learning and essentialism have been connected in the past by various researchers, in order to state that the main paradigm in machine learning processes is equivalent to choosing the “essential” attributes for the machine to search for. Our goal in this paper is to show that there are connections between machine learning and essentialism, but only for some kinds of machine learning, and often not including deep learning methods. Similarity-based approaches, more connected to the overall prototype theory, spanning from psychology and linguistics, seem more suited for pattern recognition and complex deep-learning issues, while for classification problems, mostly for unsupervised learning, essentialism seems like the best choice. In order to illustrate the difference better, we will connect both paths to their sources in other disciplines and see how human psychology influences our decision in machine-learning modeling as well. This leads to a philosophically very interesting consequence: even in the setting of supervised machine learning, essences are not present in data, but in targets, which in turn means that the categories which purport to be essences are in fact human-made, and hand-coded in the targets. The success of machine learning, therefore, does not give any substantial evidence for the independent existence of essential properties. Our stance here is to state that neither the existence nor the lack of “essential” properties in machine learning can lead to metaphysical, i.e., ontological claims.

Cyberspeak, the language of cybernetics, or its metalanguage to be more precise, consists of words that are both explaining and describing human/animal and machine forms of control and communication, while in newspeak, words were value-laden, which means they had strong positive or negative connotations connected to their use. For example, a 'spy' could only be a foreign agent, while a Russian one was a 'patriot'. First, it will be shown how there are still remnants of cyberspeak in modern science, pinpointing its cybernetic background and shaping our thoughts without us realizing they had any cybernetic origin whatsoever. Second, it will be investigated how newspeak, as its counterpart, can be analyzed from the theory of speech acts. Third, the current war in Ukraine allowed us a brief analysis of newspeak present in Russian public communication today (Putin's speeches), which will again be connected to their performative aspect.

U radu se ispituju filozofski temelji dubokog učenja. Ukazivanjem na početke dubokog učenja i umjetnog neurona kao formalnog modela ljudskog neurona moguće je tvrditi da je umjetna inteligencija razvijena i prije njezinog službenog imenovanja te da je bila snažno povezana s propozicionalnom logikom. Imajući na umu neke velike zastoje u razvoju neuronskih mreža, pokazujemo da se dubinsko učenje može tretirati kao teorija umjetne inteligencije te da potpada pod paradigmu umjetne inteligencije jer je za nju dovoljno samo učenje jer se inteligentno ponašanje uči. Dakle, duboko učenje je filozofski ili epistemološki pristup u kojem se mora zagovarati radikalni empirizam. Prema tome, ne samo da ne postoji ništa u umu što nije bilo u osjetilima, već u umu ne postoji ništa što se ne može naučiti.

In this paper, we explore the possibility of applying traditional and modern aesthetical theories to logical and mathematical proofs, with the goal of better understanding the intuitive concept of mathematical beauty. This informal concept takes a central role in the work of logicians and mathematicians and can be thought of as their main motivation. In the present paper, we try to define concepts connected to mathematical beauty or beauty in mathematical proofs, so that we may lay the foundations for a more precise definition of mathematical beauty which would be obtained through a detailed survey among logicians and mathematicians, presented in a future paper. The present paper brings crucial results to be used for constructing the survey.

This textbook presents a concise, accessible and engaging first introduction to deep learning, offering a wide range of connectionist models which represent the current state-of-the-art. The text explores the most popular algorithms and architectures in a simple and intuitive style, explaining the mathematical derivations in a step-by-step manner. The content coverage includes convolutional networks, LSTMs, Word2vec, RBMs, DBNs, neural Turing machines, memory networks and autoencoders. Numerous examples in working Python code are provided throughout the book, and the code is also supplied separately at an accompanying website. Topics and features: introduces the fundamentals of machine learning, and the mathematical and computational prerequisites for deep learning; discusses feed-forward neural networks, and explores the modifications to these which can be applied to any neural network; examines convolutional neural networks, and the recurrent connections to a feed-forward neural network; describes the notion of distributed representations, the concept of the autoencoder, and the ideas behind language processing with deep learning; presents a brief history of artificial intelligence and neural networks, and reviews interesting open research problems in deep learning and connectionism. This clearly written and lively primer on deep learning is essential reading for graduate and advanced undergraduate students of computer science, cognitive science and mathematics, as well as fields such as linguistics, logic, philosophy, and psychology.

Ovaj rad istražuje mogućnost davanja kompatibilističkog argumenta iz aspekta umjetne inteligencije. Ključna pretpostavka našeg rada jest da je umjetna inteligencija načelno moguća i da se realizira na računalnim arhitekturama u bitnome nalik današnjim. Uz taj je uvjet moguće dati definiciju slobode koja je pomirljiva s determiniranim izračunom, uz pomoć načelne nedokučivosti inteligentnog procesa. Ovo se temeljem funkcionalizma može translatirati u filozofiju uma. Pitanje je li moguće naš argument adaptirati za drugačije teorije filozofije uma ostavljamo otvorenim.