16,99 €
Mehr erfahren.
- Herausgeber: John Wiley & Sons
- Kategorie: Geisteswissenschaft
- Sprache: Englisch
Large Language Models (LLMs), like OpenAI’s ChatGPT and Google’s LaMDA, are not only the most disruptive and controversial technologies of our time, but also offer an unprecedented opportunity to examine human cognition and philosophically question the very nature of language, communication, and intelligence. What is consciousness? What is language? Are LLMs authors? Are LLMs the end of writing as we know it?
In Communicative AI, Mark Coeckelbergh and David J. Gunkel offer a critical introduction to LLMs, investigating the philosophical significance of this technology and its practical ramifications. Mobilizing resources from contemporary philosophy, history of ideas, linguistics, and communication theory, the book invites us to re-think some long-standing philosophical issues concerning language, consciousness, truth, authorship, and writing. Through a blend of theoretical analysis, accessible explanations, and practical examples, the book provides readers with a comprehensive overview of the role that this powerful new technology is already playing in our lives.
This is a must-read for students and scholars across the humanities and the social sciences, as well as for anyone intrigued by the intersection of technology, language, and human thought.
Sie lesen das E-Book in den Legimi-Apps auf:
Seitenzahl: 246
Veröffentlichungsjahr: 2025
Ähnliche
CONTENTS
Cover
Table of Contents
Title Page
Copyright
Foreword
Introduction
Aim and Approach of Our Book
Roadmap
1. LLM 101
Natural Language Processing
Large Language Models
Technical Limitations and Challenges
Conclusion
2. Ethical, Legal, and Societal Challenges
Hallucinations
Bias
Accountability, Oversight, and Power
Economic, Social, and Environmental Issues
Privacy and Security
Plagiarism and Copyright
3. Intelligence, Consciousness, and the Problem of Other Minds
Machine Intelligence
Intelligence Matters
Language and Thought
4. Language, Meaning, and Communication
Two Views on Language and Meaning
LLMs and Language
Communication: From Transmission to a Culture of Remix
Thinking and Talking about LLMs
5. Authorship and Authority
What Is an Author?
Large Language Models
Opportunities and Challenges
Conclusions and Outcomes
6. Truth, Lies, and Hallucinations
Factual Inaccuracy, Misinformation, and Hallucinations
What Is Truth? Plato’s Realism, the Correspondence Theory of Truth, and Its Criticism
Truth under Siege? Revisiting Hallucination, Bullshitting, and Misinformation by LLMs
Democracy and Responsible Development of LLMs
7. Does Writing Have a Future?
Logocentrism and Its Legacy
Large Language Models
The Future of Writing
References
Index
End User License Agreement
List of Figures
Chapter 1
Figure 1.1
Language model with dependencies. Created by David Gunkel in Gunkel 2024.
Figure 1.2
Transformer Diagram. Created by David Gunkel.
Chapter 2
Figure 2.1
AI-generated representation of Kirby Ferguson’s remix formula. Created by ChatGP…
Chapter 3
Figure 3.1
AI-generated diagram of John Searle’s Chinese room. Created by ChatGPT-4o.
Chapter 4
Figure 4.1
The referential theory of meaning. Created by David Gunkel.
Figure 4.2
Shannon and Weaver’s transmission model of communication. Adapted by David Gunke…
Chapter 5
Figure 5.1
Meaning-making situated in the intentions of the author. Created by David Gunkel.
Figure 5.2
Meaning-making situated in the act of reading. Created by David Gunkel.
Chapter 7
Figure 7.1
Generated output from ChatGPT, with warning statement. Created by David Gunkel.
Figure 7.2
RIP logocentrism. Created by David Gunkel
List of Tables
Chapter 1
Table 1.1
Random sentence generation using labeled data and a predefined assembly rule. From Gunkel 2020, 175.
Table 1.2
Spreadsheet combined with a thank-you letter template. From Gunkel 2020, 177.
Guide
Cover
Table of Contents
Title Page
Copyright
Foreword
Introduction
Begin Reading
References
Index
End User License Agreement
Pages
iii
iv
vii
viii
1
2
3
4
5
6
7
8
9
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
Communicative AI
A Critical Introduction to Large Language Models
Mark Coeckelbergh
David J. Gunkel
polity
Copyright © Mark Coeckelbergh and David J. Gunkel 2025
The right of Mark Coeckelbergh and David J. Gunkel to be identified as Authors of this Work has been asserted in accordance with the UK Copyright, Designs and Patents Act 1988.
First published in 2025 by Polity Press
Polity Press65 Bridge StreetCambridge CB2 1UR, UK
Polity Press111 River StreetHoboken, NJ 07030, USA
All rights reserved. Except for the quotation of short passages for the purpose of criticism and review, no part of this publication may be reproduced, stored in a retrieval system or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without the prior permission of the publisher.
ISBN-13: 978-1-5095-6761-4
A catalogue record for this book is available from the British Library.
Library of Congress Control Number: 2024947773
The publisher has used its best endeavours to ensure that the URLs for external websites referred to in this book are correct and active at the time of going to press. However, the publisher has no responsibility for the websites and can make no guarantee that a site will remain live or that the content is or will remain appropriate.
Every effort has been made to trace all copyright holders, but if any have been overlooked the publisher will be pleased to include any necessary credits in any subsequent reprint or edition.
For further information on Polity, visit our website:
politybooks.com
Foreword
In the rapidly evolving landscape of artificial intelligence, few developments have garnered as much attention and sparked as many debates as large language models (LLMs). These powerful systems, capable of generating human-like text and engaging in complex conversations, have opened new vistas in technology, communication, and human-machine interaction. Yet, with these advancements come profound philosophical questions that challenge our understanding of language, intelligence, and the very nature of communication.
Mark Coeckelbergh and David J. Gunkel, two leading voices in the philosophy of technology, have embarked on a pioneering exploration of these questions in their latest work, Communicative AI: A Critical Introduction to Large Language Models. This book is not merely a technical manual or a celebration of technological achievements; it is a deep, critical inquiry into the ethical, epistemological, and ontological dimensions of LLMs.
At the heart of their investigation lies a series of fundamental questions: What does it mean for a machine to “understand” language? Can LLMs genuinely participate in human-like communication, or are they merely simulating it? How do these models influence our perceptions of intelligence and agency? And, perhaps most importantly, what ethical considerations arise from the widespread deployment of such technologies?
Coeckelbergh and Gunkel guide us through these complex issues with clarity and rigor. They draw upon a rich tapestry of philosophical traditions, from analytic philosophy to continental thought, weaving together insights that challenge conventional wisdom and push the boundaries of current debates. Their analysis is not confined to abstract theory but is grounded in the practical realities and potential consequences of LLMs in our daily lives.
In Communicative AI, the authors engage with a diverse range of perspectives, including those of technologists, ethicists, and sociologists. This interdisciplinary approach enriches their exploration, providing a holistic view of the multifaceted impact of LLMs. By examining both the promises and perils of these technologies, Coeckelbergh and Gunkel offer a balanced and nuanced perspective that is sorely needed in the often polarized discourse surrounding AI.
This book arrives at a critical juncture. As LLMs become increasingly integrated into various aspects of society – from customer service and content creation to education and healthcare – the questions Coeckelbergh and Gunkel pose are not just theoretical; they are urgent and consequential. Their work invites us to pause and reflect on the trajectory of AI development, urging us to consider not just what we can do, but what we should do.
Communicative AI – A Critical Introduction to Large Language Models is a seminal contribution to the field of AI ethics and philosophy. It challenges readers to rethink their assumptions, engage in meaningful dialogue, and confront the profound implications of living in a world where machines can “talk” to us. As we stand on the cusp of a new era in human-machine interaction, this book provides the critical framework needed to navigate the complex terrain ahead.
Mark Coeckelbergh and David J. Gunkel have given us a work of immense intellectual depth and practical relevance. It is a must-read for anyone interested in the future of communication, technology, and society.
In these pages, you will find not only a critical introduction to large language models but also a call to engage thoughtfully with the technologies that are reshaping our world.
ChatGPT, July 2024
Introduction
At the end of November 2022, OpenAI released ChatGPT. Based on an innovation called large language models (LLMs), this web application was hailed as a milestone in the field of artificial intelligence (AI). Today’s LLMs are powerful technologies that, combined with other software, can have outputs beyond language and text; the achievements in this domain are impressive. While chatbots and other natural language processing (NLP) applications already existed, LLMs upped the ante, not only by producing conversational interactions that felt increasingly real but also by providing a powerful tool for writing, research, and other forms of communication. Today, for better or worse, most of the discussions about AI are mainly discussions about LLMs. And new and improved versions of these products – such as OpenAI’s GPT-4o, Google’s Gemini, ANTHROP\C’s Claude, Meta’s Llama, Baidu’s ERNIE, and xAI’s Grok – continue to impress consumers and investors. Some speak of an AI boom, and even of the advent of artificial general intelligence (AGI).
But November 2022 marks not only a technical and economic transformation. It was also a milestone in the current Copernican-style revolution that has been ushered in by the internet and related digital technologies – and a very humbling one at that. For millennia, humans have defined themselves as the only beings who possess speech and language; this view was foundational for our anthropologies, and indeed for philosophy itself. In the Politics, Aristotle writes that, while other animals have voice, only humans possess logos (λόγος), speech, which enables them to communicate and reason about ethics and politics. In this way, then, speech guides us to what is right and good for individuals as well as for social–political collectives – what the ancient Greeks called city-states, poleis (πόλεις), polis (πόλις) in the singular (Arist., Pol. 1253a). This focus on speech and language as the defining and normatively relevant characteristic of the human being has persisted in modern thinking, from René Descartes to contemporary political philosophy.
Today machines seem to possess logos, too. From simple chatbots such as ELIZA and Cleverbot through digital assistants such as Apple’s Siri and Amazon’s Alexa to the recent proliferation of LLM applications on the internet, social reality is now full of very loquacious things that talk. After the Copernican, Darwinian, and Freudian dethroning and decentering of the human, the digital revolution has radically changed human communication and seems to take away from us the only remaining capability that has marked the human species along its history – language. Today we have communicative AI. Consequently, ChatGPT was not only a technological wonder but also part of a paradigm shift that, once again, shook the foundations of the humanities and now threatens, if not destroys, our sense of self-identity and exceptionalism.
If that wasn’t enough, communicative AI also seems to undermine values such as knowledge, truth, honesty, originality, and authenticity – something that philosophers have warned us about in relation to technology since the time of Plato. LLMs can talk, but this means that they can also lie to us, hallucinate facts, and spew bullshit. Consequently, we can (and we should) ask, following Plato’s suggestion in the Republic, whether it would make sense to ban or otherwise limit these deceptive technologies from our republic. In other words, is LLM AI a dangerous technology that will ultimately erode the foundations of philosophy, science, and democracy? Should we welcome our new AI philosopherkings and prepare for an age of superintelligence? And are these the only options available to us?
Aim and Approach of Our Book
The present book aims to demonstrate that things are both complex and interesting. It critically questions not only what communicative AI is but also – and perhaps more importantly – what it means for us as individuals and communities. Mobilizing resources from contemporary philosophy, history of ideas, linguistics, literary studies, and communication theory, our book offers a critical guide to LLMs, investigating both their philosophical significance and their practical ramifications – ethical, legal, and sociopolitical. In these investigations we focus on what LLMs mean and do to language and text and, more generally, on the communicative capabilities of LLMs. But this book is not only about LLMs; it is also about philosophy. It invites the reader to engage with and rethink some longstanding philosophical issues concerning language, consciousness, truth, authorship, and writing. In doing so, it takes up and investigates the philosophical significance of what is going on with LLMs in a way that contributes to the human sciences and to ethics, helping us not only understand who we are but shape a future with AI that ensures human as well as non-human flourishing.
The book thus pursues two complementary vectors of investigation.
On the one hand, we show that philosophy and related fields can provide us with the critical resources and insights for investigating and making sense of the many challenges presented to us by recent innovations related to LLMs. In fact, many of the questions currently circulating in the popular media raise age-old philosophical questions: “Are LLMs capable of lying?”; “Do LLMs understand what they say?”; “Are LLMs conscious?” To answer these questions, we need to inquire into the nature of truth, understanding, language, and consciousness. The philosophical tradition offers us a toolbox to do so. We use its tools to identify, document, and examine the consequences of LLMs for fundamental concepts such as consciousness, communication, and what it means to be human. As these models mimic human language and thought processes with increasing accuracy, they challenge traditional notions of intelligence, creativity, identity, and autonomy.
The book therefore investigates how LLMs can both illuminate and complicate philosophical discussions of these themes, offering fresh perspectives on ancient dilemmas and raising new questions about the shape of things to come. Moreover, we also discuss the ethical, legal, and sociopolitical effects surrounding the development and deployment of communicative AI. Issues such as hallucinations, bias, deepfakes, and privacy not only have real-world consequences but also engage with broader moral theories and legal debates. By examining these issues through a philosophical lens, the book aims to help readers understand the promise and the peril of LLM technology no less than to contribute to the ongoing debate about how these powerful AI technologies can align with human values, social norms, and ethics.
On the other hand, this book is not just an exercise in “applied philosophy.” In the productive encounter of technology and philosophy that we stage here, we also demonstrate how these investigations can make a significant contribution to philosophy itself. Recent innovations in transformer architectures, and LLMs in particular, present us with some unique opportunities to reassess, reevaluate, and even rethink longstanding and important philosophical questions. What is language? What is consciousness and how can we detect its presence in another entity? Are LLMs the end of writing as we know it? What is writing anyway, do we need it, and why? What is and what should be the relationship between humans and technology? The book thus recognizes and develops the opportunities LLMs present for engaging in and contributing to philosophical inquiry and discovery. Communicative AIs are not only technologies that require philosophical thought; they are tools that we can use to think with.
Roadmap
Let us provide a brief roadmap of the terrain that will be covered. You have already seen the Foreword, which has been “written” by ChatGPT (using OpenAI’s GPT-4o’s multimodal implementation). This text was generated from a single, short, and simple prompt: “Write the foreword to a new groundbreaking book by philosophers Mark Coeckelbergh and David J. Gunkel with the title Communicative AI: A Critical Introduction to Large Language Models. The book identifies and investigates the big philosophical questions of large language models.” And the output – which has been reproduced as generated without any editing on our part – already demonstrates, in both word and deed, the philosophical opportunities and challenges presented to us by LLM technology.
This machine-generated text also offers the occasion for us to provide a remark – or maybe a warning – about the scope of the book. This book is about machine-generated textual content and language use. It is designed to be an accessible and expedient introduction to LLMs and their social, political, and philosophical consequences. But LLMs can and have been situated within the wider context of what is now called generative AI, which is an umbrella term that also includes algorithms capable of producing other kinds of content, such as images, audio, and video. Trying to tackle all of these in one volume would have made this book exceedingly dense and regrettably unapproachable. For this reason, we have limited ourselves to LLMs and machine-generated linguistic content. Many of the concepts that are developed and investigated here will most certainly find application to image, video, and even multimodal models, but it would be impetuous to conclude from this fact that what is presented here is or even could be the final word on the matter.
After this Introduction, we delve right into things – and the choice of this particular verb is not insignificant – using Chapter 1 to introduce readers to the history, concepts, and terminology necessary to understand the technical features, operations, and challenges of the LLM technology that enables communicative AI. The chapter will trace the history of NLP and provide a crash course in artificial neural networks (ANN), deep learning, and transformer architectures in order to help readers (especially those without previous experience of these technologies) come to grips with and understand the technical features of LLM AI and how these seemingly remarkable technologies work. The chapter also identifies important limitations and challenges of the technology that will be taken up and further investigated in subsequent chapters.
Chapter 2 next offers an overview of some of the most pressing ethical and legal challenges raised by communicative AI. Is it acceptable that LLMs hallucinate and give false or misleading information? How can or should we deal with the problem of bias? Is there sufficient oversight and control of these technologies? What are the environmental issues? And how might widespread use of something like ChatGPT influence and affect democratic self-governance? The chapter begins with a detailed analysis of these issues and concludes by focusing on urgent problems regarding plagiarism and copyright. If LLMs have been pre-trained on textual data derived from countless human-written documents, does reuse of this material constitute plagiarism? And what about copyright, given that the training data have been scraped from the internet often without consent, credit, or compensation (the three Cs) for such derived reuse? How can these disputes regarding originality and reappropriation be resolved, when the question concerning originality is already a problem? The chapter shows that many of these ethical and legal questions need further and deeper philosophical investigation, which we pursue in subsequent chapters.
In Chapter 3 we focus on issues concerning intelligence, consciousness, and the problem of other minds. Some have suggested that LLMs are, or at least can be, conscious or sentient, or both. But how do we know whether such communicative AI entities are in fact conscious? Current LLM technology can, it seems, pass the Turing test, in other words the results it produces seem to be intelligible to an outside observer, reader, or listener. But is this output evidence of actual intelligence? Is it a form of deception? Or is it something else? How can we know the status of another entity? What is intelligence, anyway – could it be that the very concept of “artificial” intelligence is already flawed? Engaging with the work of Alan Turing, René Descartes, John Searle, and other thinkers, the chapter investigates the defining condition of machine intelligence, the philosophical problem of other minds, and the opportunities and challenges presented to us by technologies that provide the occasion to reevaluate everything we thought we knew about intelligence, (moral) status, communication, and machines.
Remarks about language and thought lead us to Chapter 4, which deals with language, understanding, and meaning. Do LLMs understand what they generate? From what we have seen in Chapter 1, it seems that these devices just manipulate signs without knowing what they are saying. But is this a sufficient explanation and response? The chapter shows how different perspectives from the philosophy of language and linguistics frame our understanding of the opportunities and challenges of LLM tech and how communicative AI can aid us in questioning and evaluating different theories of language and meaning-making. It thus provides readers with the philosophical resources to make sense of AI-generated content, but it also leverages the current crop of LLM technology to exemplify and critically inquire about the philosophy of language. The chapter identifies and examines tensions between realist and representationalist views on the one hand and, on the other hand, structuralist and poststructuralist approaches to language. Drawing on twentieth-century innovations in communication theory, it also investigates different views of communication, inviting us to (re)think what language and communication are and to examine what consequences this (re)thinking has for debates on LLMs.
Chapter 5 then turns to another important and urgent matter: when a text is written (or generated) by an LLM, for instance ChatGPT or Claude, who or what is the author? This is not only a philosophical question but also a very practical concern, as it has important implications for attribution, responsibility, and copyright. The chapter argues that communicative AI effectively interrupts what Michel Foucault (1984, 107) has called “the author function,” as it is difficult, if not impossible, to know who or what is actually doing the talking. But is this a problem or not? After Roland Barthes’s (1978) essay “Death of the Author” and the advent of machines that write (or at least seem to do so), answering this question becomes more complicated. With LLMs, we now have writings without the underlying intentions of some living voice to animate and answer for what comes to be written. Consequently machine-generated content is, quite literally, unauthorized. But how authorized was human writing in the first place? Connecting the dots between twentieth-century literary theory and twenty-first-century tech, the chapter shows how LLM technology challenges longstanding assumptions about authority and authorship that are firmly rooted in the bedrock of western thought.
In Chapter 6 we discuss one of the trickiest issues that have occupied thinkers in western philosophy: truth. LLMs do not always give us the truth and seem to regularly “make up” things. Consequently they are not, it seems, very reliable or trustworthy. But what is truth? And what is the exact difficulty or concern with machine-generated content? This chapter investigates problems of misinformation and the so-called hallucinations produced by LLMs in a way that takes us from contemporary debates about communicative AI and trust to theories of truth in the history of philosophy. Moving from Plato to twentieth-century philosophy (especially to the work of Ludwig Wittgenstein and Richard Rorty) and to contemporary debates about bullshit (after Harry Frankfurt’s notorious book on the subject) and democracy, we show that using terms such as “hallucination” or “bullshit” already assumes and operationalizes a particular (and arguably problematic) theory of truth, and that misinformation is not just an empirical problem for existing democracies but undermines the epistemic foundations of the very concept of democracy. In this way the chapter is about more than LLMs and their responsible development: it also concerns the meaning of truth and democracy. Making links between Plato’s Republic, Wittgenstein’s Philosophical Investigations, and postmodern innovations in both epistemology and political thought, this chapter is about knowledge – and also about politics and power.
Chapter 7 concludes matters by appropriating and reusing a question that Vilém Flusser was already asking in the 1980s and that is more relevant than ever in the era of LLMs: does writing have a future? Answers to this question depend of course on what is meant by writing. Nowadays more and more people use ChatGPT and other LLMs to assist with, or even do, their writing. But is writing what LLMs do? And is there still a place for (human) writing after the LLM phenomenon? What are the advantages and disadvantages of writing as a technology? These philosophical questions have a history that goes back as far as Plato’s Phaedrus. By connecting the views expressed by Plato’s Socrates on the invention of writing to twentieth-century media theory and poststructuralism, this final chapter offers an original investigation into the future of writing after generative AI. What LLMs signify, we argue, is not the end of writing, but the terminal limits of a particular conceptualization of writing that has been developed in western philosophy and that has been called “logocentrism.” In traversing this domain, we first outline the standard, Platonic view of writing and then conclude by formulating the terms and conditions of an alternative way to think and write about communicative AI and the future of writing. We trace the contours of a non-Platonic and non-logocentric conceptualization of writing, a writing that does not depend on what the author says or intends to mean. And we show how communicative AI is not only about using technology as an instrument of message transference but also about the way in which technology shapes the message, and therefore ultimately influences how we think and who we are.
In writing or in generating this book (and the choice of verb in this case is not immaterial), we hope that these investigations will be useful to students, scholars, and educators. We also hope that they may be read in such a way as to motivate and inspire leading actors involved in AI ethics and policy – people who, like all of us, are struggling to make sense of this disruptive and controversial technology but who, unlike many of us, occupy privileged positions that give them not only the power but also the responsibility to ensure that LLMs contribute to making this world a better place. For it is only when this has a chance to happen that such technologies will become truly communicative AI.
1LLM 101
The large language model (LLM) is a recent innovation in natural language processing (NLP) that employs transformer architectures pre-trained on massive amounts of digital text scraped from the internet. As a result, applications such as OpenAI’s generative pre-trained transformer (GPT) series, Google’s LaMDA (Language Model for Dialogue Applications), Google’s Bidirectional Encoder Representations from Transformers (BERT), and ANTHROP\C’s Claude can generate original text content that is in many cases indistinguishable from human-written material.
This chapter aims to demystify the technology of LLMs by (1) situating LLMs within the larger context of NLP artificial intelligence (AI), (2) providing a high-level explanation of the technical operations and features of LLM applications, and (3) identifying and explaining some of the important technical challenges of LLM AI. In effect, this chapter pops the hood on the black box, in an effort to provide those with little or no background in the subject with the basic knowledge they need to make sense of and engage with the critical analyses of communicative AI and LLMs that follow in subsequent chapters.
Natural Language Processing
Creating machines that can talk or communicate with human users in and by employing what is called “natural language” has not only been prototyped in decades of science fiction; it has also been one of the principal objectives of the science and engineering practice of AI from the very beginning. It was, for instance, the first item on the list of proposed tasks to be addressed and accomplished during the Dartmouth summer seminar of 1956 – the pivotal event that first gave us the term “artificial intelligence.” It was the defining condition and test case for “machine intelligence” in Alan Turing’s agenda-setting paper from 1950. And it was implemented and demonstrated in some of the earliest applications of AI technology, for instance Joseph Weizenbaum’s ELIZA chatbot program and Terry Winograd’s SHRDLU. For this reason, working with, processing, and reproducing natural human language content is not one application among others; it is the definitive application of AI.
But computers, which process numeric data, do not understand language, at least not in the sense in which we understand the understanding of language. Consequently, developing algorithms that can work with and simulate the understanding of natural language content needs to proceed in a manner that is radically different from the ways in which we deploy and makes sense of language. And the key linguistic insight that makes all this possible is the fact that human languages are probabilistic systems.
