
Deep learning, as the most abstract form of artificial intelligence (AI), is currently changing many things in our lives on a massive scale and developing human-like abilities, which the author subjects to an interdisciplinary interpretation. Based on the history of philosophy, the discussion is placed in a framework between empiricism and rationalism. Finally, the author argues a position of moderate empiricism of the human mind, which makes it possible both to evaluate the deep learning abilities achieved and to approach the question of which faculties can be modeled and which cannot. This approach also allows him to assess possible future technological developments from a philosophical perspective. In this way, a link is established between the skills achieved through deep learning and the characteristics of the human mind. Deriving abstract knowledge from experience “involves the active engagement of psychological faculties such as perception, memory, imagination, attention, and empathy, as well as their complex interplay.”
The organization and structure of the book corresponds to this approach. In the first chapter, the author describes the two opposing positions mentioned above. The very detailed discussion of the philosophical spectrum from empiricism to Plato--based on the work of numerous researchers--enables the author to establish a corresponding relationship between the computational model of deep learning and models of the faculties of the human mind on a biological basis, in order to finally present his position as a moderate empiricist in the form of his concept of a domain general modular architecture (DoGMA) for modeling human faculties in human-like cognitive structures. This approach provides the author with a way to understand deep learning by means of psychological faculties as used by an empiricist.
In the second chapter, in addition to an overview of deep learning and a description of its characteristic strengths and weaknesses, the author explains why a stratified rationality approach is preferred to a pure intelligence approach--ultimately also in support of the moderate empiricist position. Since the traditional approaches to the psychometric assessment of the characteristics of the faculties have considerable methodological disadvantages from the author’s point of view, the author opens up the possibility of a fair, qualitative assessment of the performance progress of an artificial decision agent on the basis of the layers of rationality after a possible integration of findings on the philosophical considerations of the individual mental faculties into the architecture options of deep learning networks. This makes it possible to consider the degree of biological plausibility of the necessary implementation effort.
The remaining chapters are devoted to the individual faculties of the human mind in the context of the DoGMA approach, in the interdisciplinary sense described above. For reasons of simplicity of presentation, the author concentrates on a prominent empiricist philosophical representative for each individual mental faculty who has made a significant contribution to the corresponding field, and relates the statements to relevant technological approaches. In chapter 3, John Locke’s observations on perception are taken up and discussed on the basis of realizations in deep convolutional neural networks (CNNs). In the fourth chapter on memory, the statements of Ibn Sina (Avicenna) are linked to successful models in the field of deep reinforcement learning. In chapter 5, the author examines David Hume’s view of the field of imagination in relation to powerful generative architectures. The sixth chapter is devoted to attention. Here, the influence of William James’s theory on transformer architectures is discussed. Chapter 7 deals with social perception and morality based on the reflections of Sophie de Grouchy. This is a particularly difficult part, which the author relates to research in the field of affective computing and artificial rearing. Finally, the difficult problem of the interaction (coordination and control) of several independently modeled faculties in a deep learning system, to realize an integrated and complete artificial rational decision agent, is briefly discussed. The author’s explanations are supported throughout by extensive knowledge of the psychological-philosophical background and the existing technological solutions, which allows him to establish the desired interdisciplinary connection. Where necessary, the text is supplemented by clear, informative illustrations, in addition to a very extensive bibliography (over 50 pages) and a very detailed index.
In this work, the author demonstrates very well that the development of deeply rational machines requires both a very good understanding of human rationality and an evaluative assessment of technological feasibility. He is so successful in doing this that the book should actually find a broad readership, ranging from AI researchers to cognitive scientists and philosophers who want to understand the effects of the new technologies. But with a little effort, it is also perfectly readable for the layperson, especially if they want to move beyond the usual, mostly technology-heavy presentation of the topic and try to build a bridge from their own humanity to artificial machines.
More reviews about this item: Amazon