Machine Consciousness: A Modern Approach

The purpose of the tutorial is to offer an overview of the theoretical and empirical issues in artificial consciousness. Is it possible to devise, project and build a conscious machine? What are the theoretical challenges? What are the technical difficulties? What is the relation between cognition and consciousness? The most promising models will be sketched and contrasted: Global Work Space, Tononi’s information integration, Embodied Cognition, Externalist approaches, bio-inspired cognitive architectures. Questions to be discussed include: What advantage does consciousness provide? Is conscious experience the hallmark of a special style of information processing? Would conscious machines be able to outperform intelligent machines? 1. Is consciousness Relevant for AI? Since 1949 – when Shannon and Weaver cast the foundation for the forthcoming information age (Shannon and Weaver 1949) – computer science, cognitive science, AI and engineering have aimed to replicate the cognitive and mental capabilities of biological beings. To this purpose, various strategies have been envisaged. By and large, we may distinguish various approaches: the symbolic and logical approach of classic AI (Haugeland 1985a; Russell and Norvig 2003), the sensori-motor approach (Pfeifer 1999), neural-network oriented design (Sporns 2011), the bioinspired strategy (Pfeifer, Lungarella et al. 2007b), and the classic AI approach (Russell and Norvig 2003). All these approaches share something – they focus mostly on the intelligent behavior showed by agents. They try to replicate the capability to react to the environment stimuli and to choose the appropriate course of actions. However, something may be missing. According to Russel & Norvig (2003) one of the main goal of AI has been that of designing system that think ... “machine with minds in the full and literal sense” (Haugeland 1985b). A full-fledged mind inevitably raises the issue of consciousness. If we take the human being as the target of our efforts, we are immediately struck by something that AI so far has not addressed properly, namely consciousness. Human beings not only act and behave. They are conscious of what they do and perceive. Somehow, human beings feel what happens to them, a condition usually defined as being conscious or as having consciousness. There is something that it like to be a certain human being (Nagel 1974). Furthermore, there seems to be some strong dependence between autonomy and consciousness. The problem of consciousness appears so difficult that it has been dubbed the hard problem (Chalmers 1996), to the extent that some scientists and philosophers have even argued that it may lie beyond our cognitive grasp (McGinn 1989; Harnad 2003). For one, there is a crucial question of paramount importance in neuroscience and AI: does consciousness provide a better way to cope with the environment? Or, to put it differently, has consciousness any selective advantage? At this point and very broadly, there are two conflicting positions. On the one hand, there are authors that set aside consciousness as a philosophical issue of no concern for AI, Cognitive Science and Neuroscience. As Ronald Arkin put it, “Most roboticists are more than happy to leave these debates on consciousness to those with more philosophical leanings” (Arkin 1998). Either because consciousness has no practical consequences or because it is a false problem, these group of authors prefer to focus on more defined issues (vision, problem solving, knowledge representation, planning, learning, language processing). For them, either consciousness is a free bonus at the end of the AI lunch, or is nothing but a by-product of biological/computational processes. On the other hand, an increasing number of scientists are taking seriously into consideration the possibility that human beings’ consciousness is more than an epiphenomenal by-product. Consciousness may be the expression of some fundamental architectural principle exploited by our brain. If this insight were true, it would mean that, in order to replicate human level of intelligence, we ought to tackle with consciousness too. In support of pro-consciousness group, there is the fact that we have a first-person experience of being conscious, which is not deniable by any amount of theoretical reasoning. In other words, when I feel a pain in my arm, there is something more than the triggering of some appropriate behavioral response. If this feeling had no practical consequences, it would follow that consciousness is epiphenomenal – namely that it has no practical consequences whatsoever. More bluntly, it would follow that consciousness is a useless phenomenon. Such a conclusion would contradict the principle of natural selection – it does not seem likely. Furthermore, in the Tutorial Volume 2, Issue 1, July 2013 7 Natural Intelligence: the INNS Magazine animal kingdom, there seems to be a correlation between highly adaptable cognitive systems (such as human beings, primates, and mammals) and consciousness. Insects, worms, arthropods, and the like that are usually considered devoid of consciousness are much less adaptable (they are adaptable as a species but not very much as individuals). As a result, many scientists are now looking for something that explicitly addresses the issue of machine consciousness (Buttazzo 2001; Holland 2003; Holland 2004; Adami 2006; Chella and Manzotti 2007; Aleksander 2008; Aleksander, Awret et al. 2008a; Buttazzo 2008; Chrisley 2008; Manzotti and Tagliasco 2008). So far, there is no accepted consensus as to what consciousness may be. There are several and often conflicting hypotheses. According to some authors, consciousness is the result of a special kind of information process related with information integration (Tononi 2004b; Tononi 2008). According to another group depend on goal generation and development (Manzotti and Tagliasco 2005b), or embodiment (Holland 2004), or a certain kind of information processing akin to the global workspace (Shanahan 2005a; Shanahan 2010), or the replication of imagination and synthetic phenomenology (Aleksander, Awret et al. 2008b; Chrisley 2009b), or emotions (Ziemke 2008a), and so forth. Furthermore, consciousness is a not only a technical challenge but also a theoretical feat. In this paper, I would like to address two lines of enquiry. On the one hand, I would like to consider and to list a series of fundamental scientific problems that consciousness research cannot set aside. On the other hand, I would like to consider a series of approaches and I will briefly evaluate their pros and cons. Among the main scientific issues, I would list: