Kant and Artificial Intelligence

Hyeongjoo Kim, Dieter Schönecker, Hyeongjoo Kim, Dieter Schönecker

• Publisher: De Gruyter
• Year: 2022
• Print ISBN: 9783110706543
• eBook ISBN: 9783110706666
• Edition: 1
• Topic: Philosophy
• Subtopic: Artificial Intelligence (AI) & Semantics

How are artificial intelligence (AI) and the strong claims made by their philosophical representatives to be understood and evaluated from a Kantian perspective? Conversely, what can we learn from AI and its functions about Kantian philosophy's claims to validity? This volume focuses on various aspects, such as the self, the spirit, self-consciousness, ethics, law, and aesthetics to answer these questions.

Theoretical Philosophy

1 Minds, Brains, and Deep Learning: The Development of Cognitive Science Through the Lens of Kant’s Approach to Cognition

Tobias Schlicht

Ruhr-University Bochum

Abstract

This paper reviews several ways in which Kant’s approach to cognition has been influential and relevant for the development of various paradigms in cognitive science, such as functionalism, enactivism, and the predictive processing model of the mind. In the second part, it discusses philosophical issues arising from recent developments in artificial intelligence in relation to Kant’s conception of cognition and understanding. More precisely, it investigates questions about perception, cognition, learning, understanding, and about the age-old debate between empiricists and rationalists in the context of so-called deep neural network architectures as well as the relevance of Kant’s conception of cognition and understanding for these issues.

1 Introduction

If you follow the headlines, you can easily get the impression that much of contemporary cognitive science is heavily influenced by Kant’s philosophy. Philosopher Andrew Brook (1994) called him the “intellectual godfather” of cognitive science, since Kant allegedly already defended a functionalist theory of mind, arguably the philosophical foundation of artificial intelligence. Neuroscientist Georg Northoff (2018, viii) reports that rereading Kant’s Critique of Pure Reason has awakened him from his dogmatic slumbers, just like reading Hume had awakened Kant. Impressed by empirical evidence about self-generated brain activity, Northoff and others speak of the “Kantian brain” and associate this activity with Kant’s notion of spontaneity (Fazelpour/Thompson 2015). Francisco Varela (Weber/Varela 2002) acknowledged Kant’s enormous influence on his own autopoietic approach to life and cognition, and more recently Link Swanson (2016) has traced the popular predictive processing paradigm back to Kant’s general project. This is striking, given that Kant’s project was not primarily concerned with issues in the philosophy of mind but driven rather by epistemological concerns. But although Kant may not have subscribed to all these views attributed to him, such writings present various ideas from his theoretical philosophy as having had or still having an enormous influence on contemporary philosophy of mind and cognitive science.

In this review paper, I will first sketch several ways in which Kant’s approach to cognition has been influential and relevant for the development of cognitive science. Kant’s relevance goes well beyond some vapid and superficial similarity of certain concepts; many philosophers claim that Kant already anticipated several tenets of classical cognitivism, enactivism, and the predictive processing model of the mind. In the second part, I will add one more piece to this story by discussing philosophical issues arising from recent developments in artificial intelligence. More precisely, I want to sketch some of the philosophical issues associated with so-called deep neural network architectures and the relevance of Kant’s conception of cognition and understanding for these issues. As will become clear, the performance of deep neural networks (DNNs) raises important questions about perception, cognition, learning, understanding, and about the age-old debate between empiricists and rationalists; this has led some researchers in machine learning to revive some of Kant’s core ideas regarding cognition, developing a Kantian cognitive architecture to overcome the shortcomings of existing deep learning architectures.

2 Cognitive Science Through the Lens of Kant’s Theoretical Philosophy

Kant’s general influence on contemporary thinking is unquestioned and familiar. Gomes (2017) lists an impressive number of mental phenomena for which Kant’s philosophy has been and still is very influential, e.g., the connection between consciousness and self-consciousness (Schlicht 2016/2017) or the debate about conceptual and non-conceptual perceptual content (McDowell 1994, Hanna 2008).¹ Brook (1994) already considered several of Kant’s central claims about the mind as having fueled cognitive science more directly; most notably the claim that “most representations require concepts as well as percepts”, and Kant’s method of transcendental argument, understood as the attempt to “reveal the conditions necessary for some phenomenon to occur” (Brook 1994, p. 12). Based on this initial familiarity of Kant’s stance on issues in the philosophy of mind, cognitive science and contemporary debates, one can reconstruct the historical changes that cognitive science underwent through the lens of various aspects of Kant’s theoretical philosophy and find traces of some specific ideas of his thinking in the works of cognitive scientists.

2.1 Kant and Functionalism

When John McCarthy coined the term ‘artificial intelligence’ (AI) in the context of the famous Dartmouth conference in 1956, he described the goal of this project as “that of making a machine behave in ways that would be called intelligent if a human were so behaving” (McCarthy et al. 1955). In a similar vein, Margaret Boden describes the overarching goal of research in AI as “to make computers do the sorts of things that minds can do” (Boden 2016, p. 1). The focus in the first research phase that followed was already set by McCarthy et al. They intended to “attempt […] to find how to make machines use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves” (McCarthy et al. 1955, p. 12). While this has been achieved in some areas like speech production and chess computers, in which AI systems sometimes even outperform humans in very specialized problem-solving tasks, the “holy grail” (Boden 2016, p. 18) of AI research has always been the development of an AI system that exhibits “general intelligence”, understood “as the ability to perform tasks and attain goals in a wide variety of environments” (Shanahan 2019, p. 91, cf. Legg/Hutter 2007). This broad-stroke characterization of intelligence bypa...

Table of contents

• Title Page
• Copyright
• Contents
• Theoretical Philosophy
• Practical Philosophy
• Aesthetics
• Name Index
• Subject Index