Introduction

This book is about (re)building a bridge between two different “sciences of the artificial” - Artificial Intelligence and Cognitive Science - that, nowadays, apart from some notable exceptions, do not talk as much with each other as they should. Here, I review some of the main themes that have characterized the historical paths of these two disciplines and argue that the technological maturity reached in several domains now calls for a renewed joint enterprise aimed at addressing more substantial challenges that these two disciplines have to face from a scientific viewpoint.

The book explicitly targets a multidisciplinary audience. As such, it is mainly an act of courage (or probably of irresponsibility), since experts in the specific subfields of AI and Cognitive Science will have for sure much more to say and would surely be able to communicate their own ideas in a better way than I can. However, as mentioned, this book privileges the breath of the connections between the disciplines rather than the depths of the exploration within each single discipline. As such, it is not a manual or a handbook since it presupposes the knowledge of same basic elements of each discipline that will be touched upon by our arguments. Of course, scholars and students of the diverse fields have knowledge of different pieces of the entire puzzle and need to be briefly introduced to the aspects they do not know. This service is provided in the book, though we direct the reader towards specialized literature for details.

One of the main goals of this manuscript is to show the reader that the so- called “cognitive design approach” has still an important role to play in the development of intelligent AI technologies as well as in the context of development of plausible computational models of cognition. In other words, the study of the “Cognitive Design” principles for building “Artificial Minds” will be hopefully a useful instrument for current and future generations of AI and cognitive science scholars and students. In this respect, a caveat is necessary: in the philosophical literature on AI there are many different, and well-known, positions about whether or not it is justifiable to use the terms “mind”, “intelligence”, or “thinking” to describe the constitutive or the behavioural elements of a computational system. In this book we will not foray into the details of such a monumental and decades-long debate, which also involves the attribution of such faculties to other “species” (from non-human mammals to bacteria). Given the actual purpose of the book, we will also avoid roughly summarizing it because such an attempt would be necessarily incomplete. Sometimes, however, we will refer to some instances of such a debate. For the moment, we will just mention here, as a reference for the position of why the term “mind” can be justifiably associated with the term “artificial”, the book Artificial Minds by Stan Franklin (Franklin, 1995). The position defended by Franklin - which sees the possession of a “mind” as a matter of degrees and not as a mere Boolean notion and that, as such, foresees the possibility of implementing (to some degree) a “mind” in an artificial system - can be considered our starting working hypothesis.

In Chapter 1, I review the main historical points of contact between AI and cognitive research by briefly introducing early cognitively inspired systems and paradigms and the main factors that led to a “paradigm shift” in the research agendas of these two disciplines starting from the mid-1980s. The reasons for the current renewed interest of a cognitively inspired approach in AI research are discussed. Chapter 2 introduces the main differences between AI systems of “cognitive” inspiration and AI systems that adopt machine-oriented methods to solve specific problems, and proposes a first design distinction between “functional” and “structural” models. It additionally introduces the reader to the debate over the levels of analysis of computational systems (whether they are cognitively inspired or not). The third chapter analyzes the main theories of rational behaviour developed so far and their influence in the realization of rational artificial models. As for the different types of “functional” and “structural” models introduced in Chapter 2, different types of explanatory accounts are provided. Finally, the chapter presents the “Minimal Cognitive Grid” (MCG), a pragmatic methodological tool proposed to rank the different degrees of accuracy of artificial systems built according to the “structural” design perspective in order project and predict their explanatory power with respect to the natural systems that are taken as source of inspiration. The MCG is practically tested in Chapter 4, through the analysis of different types of artificial systems (cognitively inspired and not), ranging from the IBM Watson and Alpha Go, to cognitive architectures like SOAR and ACT-R to artificial models of cognition like DUAL PECCS (a commonsense conceptual categorization system that I have developed over the last few years at the University of Turin, in collaboration with my colleagues Daniele Radicioni and Valentina Rho). The analysis concerning DUAL PECCS and its comparison with the ACT-R and SOAR cognitive architectures is mainly based on a published paper: “The Knowledge Level in Cognitive Architectures: Current Limitations and Possible Developments” (that I co-authored with Christian Lebiere and Alessandro Oltramari), which appeared in 2018 in the journal Cognitive Systems Research. Chapter 5 proceeds by providing a comparison of the different types of evaluation approaches proposed in AI and cognitive science and finalized at individuating the degree of “intelligence” or “cognitive compliance” of artificial systems. After the introduction of the Turing Test and of some of its variations, the chapter analyzes the Winograd Schema Challenge and Newell Test for a theory of cognition along with some challenges organized by the international scientific community like the RoboCup World Soccer and the RoboCup@Home. Such approaches, partially having different purposes, will be compared with the proposed MCG. Finally, Chapter 6 concludes the journey by suggesting current and future areas of possible collaboration between AI and Cognitive Science.