HUMANITIES LAB: LOGICS AND EPISTEMOLOGY

Giacomo Sillari

Obiettivi formativi

We will tackle questions such as: Can machines think? Can they understand what they say? And what, if anything, distinguishes artificial intelligence from human intelligence?

Prerequisiti

None

Risultati di apprendimento attesi

By the end of the course, students should be able to: Explain the fundamental concepts of formal systems, algorithms, computability, and the limits of mechanical reasoning. Understand and critically compare major theories of machine intelligence and understanding, including symbolic AI, functionalism, and connectionism. Explain the basic principles underlying artificial neural networks, deep learning, and large language models. Distinguish among different conceptions of representation, meaning, knowledge, intelligence, and understanding. Critically evaluate philosophical arguments and empirical evidence concerning the cognitive and epistemic capacities of contemporary AI systems. Assess the opportunities and limitations of using artificial intelligence as a source of information, a reasoning aid, and a participant in human knowledge practices.

Contenuti Del Corso

Computers can prove mathematical theorems, defeat world champions at chess and Go, predict the structure of proteins, and produce remarkably sophisticated arguments. Large language models can write poetry, solve problems, and explain complex ideas, yet they can also make mistakes that seem almost incomprehensible to humans. How should we make sense of these achievements and failures? Are we witnessing the emergence of artificial minds, or the development of increasingly powerful machines that merely imitate some of the products of human intelligence? These questions have a surprisingly long intellectual history. The dream of mechanizing reason goes back to Leibniz and in a sense all the way back to Aristotle. In the twentieth century, mathematical logic made the notion of computation precise, while Turing and Gödel established fundamental limits on what formal procedures can accomplish. The invention of digital computers then transformed a philosophical possibility into an engineering project: building machines capable of intelligent behavior. This course follows that extraordinary intellectual journey, from formal systems and Turing machines to symbolic artificial intelligence, neural networks, deep learning, and contemporary language models. Along the way, we will encounter some of the central debates in the philosophy of mind and epistemology. Is intelligence a matter of following rules? Can understanding emerge from manipulating symbols? Can machines acquire knowledge without being explicitly taught facts? How should we evaluate the reliability of systems whose internal processes we cannot fully explain? The course combines philosophy, logic, cognitive science, and the history of AI. Students will engage with classic arguments by, among others, Alan Turing, Herbert Simon, Daniel Dennett, and John Searle, as well as contemporary research on neural networks and large language models.

Testi Di Riferimento

The principal collection of classic readings is: John Haugeland, Carl F. Craver, and Colin Klein (eds.), Mind Design III: Philosophy, Psychology, and Artificial Intelligence. MIT Press, 2023. We will read six readings from that collection Selected contemporary research articles and videos will supplement the anthology. Required sections and video segments will be specified weekly on MyLuiss. Optional material will be available for students wishing to explore particular topics in greater depth.

Metodologie Didattiche

The course combines lectures, close reading of philosophical and scientific texts, guided discussions, and practical demonstrations.

Modalità di verifica dell'apprendimento

Active Participation

Criteri per l’assegnazione dell’elaborato finale

Speak with the instructor

Settimana 1

PART A — COMPUTERS, COMPUTING, AND COMPUTATION Week 1 — Rules, Algorithms, and Formal Systems When we solve a problem by following rules, how much of the thinking lies in understanding, and how much in simply executing the procedure? Can we turn reasoning into something so precise that even a machine could do it? Reading: John Haugeland, What Is Mind Design?, in Mind Design III, Chapter 2 (selected sections).

Settimana 2

Week 2 — Turing Machines and the Limits of Computation Turing imagined an extraordinarily simple machine and showed how it could capture the power of mechanical computation. Yet the work of Turing and Gödel also revealed surprising limits to formal reasoning, raising fundamental questions about what machines (and perhaps minds) can ever accomplish. Reading: Haugeland, Chapter 2 (selected sections)

Settimana 3

Week 3 — Symbolic AI: Intelligence as Symbols and Search The pioneers of artificial intelligence made a bold conjecture: perhaps thinking is ultimately a matter of manipulating symbols according to rules. Their programs could prove theorems and solve difficult problems, but they also revealed how elusive everyday intelligence can be. Reading: Allen Newell and Herbert A. Simon, Computer Science as Empirical Inquiry: Symbols and Search, in Mind Design III, Chapter 3.

Settimana 4

PART B — INTELLIGENCE AND UNDERSTANDING Week 4 — Turing's Question: Can Machines Think? How could we tell whether a machine really thinks? Turing proposed an ingenious conversational test, challenging us to reconsider what intelligence is and what evidence could establish its presence. Reading: Alan M. Turing, Computing Machinery and Intelligence, in Mind Design III, Chapter 6.

Settimana 5

Week 5 — Minds, Meaning, and Intentionality If a machine speaks and behaves exactly like someone who understands, does it really understand anything? Dennett and Searle offer radically different answers, opening a debate about meaning and the nature of minds that remains unresolved today. Readings: Daniel C. Dennett, True Believers: The Intentional Strategy and Why It Works, Chapter 11; John R. Searle, Minds, Brains, and Programs, Chapter 12 (selected sections).

Settimana 6

PART C — NEURAL NETWORKS AND LEARNING Week 6 — The Birth of Connectionism What if machines could learn from experience instead of relying on rules written by human programmers? The earliest artificial neural networks offered a radically different approach to intelligence, but a deceptively simple logical problem exposed a fundamental limitation. Reading: David E. Rumelhart, The Architecture of Mind: A Connectionist Approach, in Mind Design III, Chapter 19 (Part I).

Settimana 7

Week 7 — How Neural Networks Learn How can a network discover useful patterns that nobody has explicitly taught it to recognize? A breakthrough in training multilayer networks showed how machines could learn increasingly complex representations, laying the foundations for modern deep learning. Reading: Rumelhart, Chapter 19 (Part II)

Settimana 8

Week 8 — Deep Learning: From Perception to Meaning Over the following decades, neural networks learned to recognize images, master complex games, and discover remarkable patterns in language. These achievements raise a fascinating question: when a machine learns to recognize, predict, or classify the world, what exactly has it learned? Readings: Instructor's lecture notes; Optional: Tomas Mikolov, Wen-tau Yih, and Geoffrey Zweig, Linguistic Regularities in Continuous Space Word Representations (2013).; Julia Haas, The Evaluative Mind, in Mind Design III, Chapter 16.

Settimana 9

PART D — LANGUAGE MODELS AND THE EPISTEMOLOGY OF AI Week 9 — How Large Language Models Work How can a machine trained to predict the next word become a remarkably capable conversationalist, programmer, and problem-solver? We look inside contemporary language models to understand how their surprising abilities emerge from relatively simple learning objectives pursued on an enormous scale. Reading/viewing: Andrej Karpathy, Deep Dive into LLMs like ChatGPT (2025), selected segments

Settimana 10

Week 10 — The Limits of Machine Knowledge: A Skeptical Perspective Language models can produce convincing explanations and sophisticated arguments while occasionally making astonishingly elementary mistakes. Do these failures reveal fundamental differences between human understanding and the kind of intelligence that machines have developed? Reading: Walter Quattrociocchi, Valerio Capraro, and Matjaž Perc, Epistemological Fault Lines Between Human and Artificial Intelligence (2025). Optional: Marina Mancoridis et al., Potemkin Understanding in Large Language Models (2025), selected sections.

Settimana 11

Week 11 — Can Language Models Have Knowledge and Understanding? But perhaps the skeptics underestimate what language models actually learn: their internal representations sometimes capture remarkably sophisticated relationships about the world. Could these systems possess genuine forms of meaning, understanding, or knowledge, even if they differ profoundly from human minds? Reading: Céline Budding, What Do Large Language Models Know? Tacit Knowledge as a Potential Causal-Explanatory Structure (2025).

Settimana 12

Sundry Topics; Q&A